CN109344821A - Small target detecting method based on Fusion Features and deep learning - Google Patents
Small target detecting method based on Fusion Features and deep learning Download PDFInfo
- Publication number
- CN109344821A CN109344821A CN201811005464.6A CN201811005464A CN109344821A CN 109344821 A CN109344821 A CN 109344821A CN 201811005464 A CN201811005464 A CN 201811005464A CN 109344821 A CN109344821 A CN 109344821A
- Authority
- CN
- China
- Prior art keywords
- network
- frame
- target
- characteristic pattern
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of small target detecting method based on Fusion Features and deep learning, solves to small target deteection low precision and real time problems.Its implementation is: extracting high-resolution features figure by the network model of deeper better ResNet101；5 low resolution characteristic patterns being sequentially reduced, augmented features figure scale are extracted by auxiliary convolutional layer；Analysis On Multi-scale Features figure is obtained by feature pyramid network；The profile information of deconvolution operation fusion high-level semantics layer and the profile information of shallow-layer are used in feature pyramid network structure；Target prediction is carried out using different scale and the characteristic pattern of fusion characteristics；With non-maxima suppression to multiple prediction frames and category score, the bezel locations and classification information of final target are obtained.The present invention has under the requirement for guaranteeing real-time detection, it is ensured that the high-precision advantage of small target deteection can fast and accurately detect the Small object in image, the target real-time detection that can be used in unmanned plane.
Description
Technical field
The invention belongs to technical field of image information processing, relate generally to deep learning target detection, specifically one
Small target detecting method of the kind based on Fusion Features and deep learning can be used for real-time positioning and classification to Small object.
Background technique
Target detection is a challenging project in computer vision field, currently based on the target of deep learning
Detection method is broadly divided into two classes, and one kind is the convolutional neural networks model based on candidate region, and one kind is the volume based on recurrence
Product neural network model.Depth convolutional neural networks model based on candidate region: the various optimizations of R-CNN network and R-CNN
Network.R-CNN network model passes through selective search (selective search) algorithm first and extracts candidate frame, then makes
Feature is extracted to candidate frame with depth convolutional neural networks (DCNN), is finally carried out using classifier support vector machines [5]
Classification.Since time-consuming and repeats to extract feature for the extraction algorithm of frame candidate in R-CNN, researcher has carried out a system
The optimization method of column: feature extraction is carried out to whole image first in Fast R-CNN, and the candidate region extraction stage is mentioned
The candidate region taken, which is mapped on characteristic pattern, carries out target detection；Secondly replace branch using the full articulamentum in convolutional neural networks
Hold vector machine SVM；It uses full convolutional neural networks FPN to extract network as candidate region in Faster R-CNN, and waits
Favored area extracts network and target detection network share convolutional layer, greatly has compressed the target detection time.Joseph in 2016
Redmon etc. proposes a kind of target detection network model --- YOLO (the You Only of new depth based on recurrence
Look Once), it is different from the deep learning target detection network based on candidate region, this method do not need to input picture into
Row candidate region is extracted, but input picture is divided into S*S grid, and each grid is responsible for center in the target of the grid
Detection predicts that the frame that all grids include, positioning confidence level and target belong to the probability of each classification, finally by it is non-greatly
Value inhibition obtains final testing result.Detection speed of this method on titan x GPU can achieve 45 frame per second, completely
Meet the requirement of real-time detection, but since this method gives a forecast for S*S grid, the quantity of grid is direct
Final target detection precision is influenced, the detection of Small object and dense target is unfavorable for.The problem of for YOLO, research
Person proposes YOLOV2, SSD network and RON network etc., and the advantage of these networks has: using deeper on the basis of YOLO network
Convolutional neural networks model extraction feature, improve target detection precision；Target detection is carried out using Analysis On Multi-scale Features figure, is adapted to
Multiscale target detection；The full articulamentum convolutional layer for being eventually used for target detection is replaced, can be reduced in model in this way
Parameter reduces detection time.
These two types network model is each advantageous at present, and the convolutional neural networks model inspection precision based on candidate region is general
It is higher, but real-time is poor；The number of network model based on segmentation often detection accuracy and real-time and the grid of division
There are much relations, to detect Small object, often requires that the grid number of division is larger, real-time just decreases.Both the above master
The model of stream still can not achieve higher detection accuracy to the Small object in image while guarantee real-time detection.
Summary of the invention
The purpose of the present invention is existing insufficient in view of the above technology, a kind of real-time inspection that can be realized to Small object is proposed
The small target detecting method based on Fusion Features and deep learning surveyed.
The present invention is a kind of small target detecting method based on Fusion Features and deep learning, which is characterized in that includes
Following steps:
(1) prepare atlas: use the training dataset of image set PASCAL VOC2007 and PASCAL VOC2012 as
Training set uses the test data set of image set PASCAL VOC2007 as test set, above-mentioned image set be containing size not
With the online disclosed image set of target.
(2) the small target deteection network model based on Fusion Features and deep learning is built: using residual error network as feature
The basic network of extraction adds the feature extraction network that five layers of convolution pondization operation constitute auxiliary after residual error network, obtains
The characteristic pattern of more kinds of scales is basic construction feature pyramid with a variety of scale feature figures, uses deconvolution and up-sampling side
Method obtains the characteristic pattern of resolution ratio same as shallow-layer characteristic pattern, to high-level characteristic figure and shallow-layer in such a way that element is added
Characteristic pattern carries out Fusion Features, obtains the characteristic pattern of more descriptive power；Finally addition prediction network, uses polygon frame and non-pole
Big value suppressing method obtains Small object classification and position；
(3) the target loss function of network model: the network model that training is built on training set of images, construction are constructed
The target loss function L (x, l, c, g) of network model；
(4) training network model: the training of network model is divided into the training of two stages formula, is minimized using gradient descent method
Loss function is simultaneously successively reversely adjusted the weight parameter in network, obtains final trained network model；
(5) small target deteection: original image to be detected is input in trained network model, obtains mapping to be checked
As the target category and position coordinates of Small Target.
The present invention, more consideration is given to the use of different scale characteristic pattern, passes through what is built in the design of depth model
Feature pyramid merges the characteristic pattern of different scale, so that all having shallow-layer position letter on the characteristic pattern of different resolution simultaneously
Breath and high-level semantics information, so that the feature of Small object is more retained, to improve the detection effect to Small object
Fruit.
Compared with prior art, the present invention has the advantage that
1) present invention adds supplemental characteristic after the basic network based on residual error network and extracts the complete spy that network is constituted
Sign extracts network so that the scale of characteristic pattern extracted in the depth model that the invention proposes is expanded, it is multiple dimensioned not
The information characteristics abundant of image to be detected are saved in multiple dimensions with the characteristic pattern of resolution ratio, information characteristics abundant are
Prediction provides more effective informations, improves detection accuracy；
2) the present invention is based on the characteristic patterns of a variety of scales to build feature pyramid, using deconvolution operation to low point high-rise
The characteristic pattern distinguished is up-sampled to obtain resolution ratio same as shallow-layer characteristic pattern, and to high-rise special in such a way that element is added
Sign figure carries out Fusion Features with shallow-layer characteristic pattern, the characteristic pattern of more descriptive power is obtained, so that the different resolution of prediction interval
Characteristic pattern on all there is shallow-layer location information and high-level semantics information, the just fusion feature due to having used a variety of scales simultaneously
Figure improves so that the detection accuracy finally obtained is higher than the algorithm of target detection for only using a kind of scale feature figure to small mesh
Detection accuracy, while the model of network proposed by the invention and little are marked, biggish computation burden is not present, so reaching real
The requirement of when property, to sum up, network model proposed by the present invention have preferable detection effect to Small object, are having compared with high detection
While precision, it can also guarantee the real-time of detection.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention；
Fig. 2 is the network structure constructed in the present invention；
Fig. 3 is the testing result figure of the present invention or SSD prototype network, and wherein Fig. 3 (a) is testing result of the invention, Fig. 3
(b) be SSD prototype network testing result；
Fig. 4 is the testing result figure of the present invention or SSD prototype network, and wherein Fig. 4 (a) is testing result of the invention, Fig. 4
(b) be SSD prototype network testing result；
Fig. 5 is the testing result figure of the present invention or SSD prototype network, and wherein Fig. 5 (a) is testing result of the invention, Fig. 5
(b) be SSD prototype network testing result；
Fig. 6 is the testing result figure of the present invention or SSD prototype network, and wherein Fig. 6 (a) is testing result of the invention, Fig. 6
(b) be SSD prototype network testing result.
Specific embodiment
The present invention is described in detail with example with reference to the accompanying drawing
Embodiment 1
Target detection is the farming analyte detection in the important research topic of image application field, such as wisdom agricultural, is placed
Garden safety detection in field etc..To the detection of image Small Target, there is also problems at present, for example are easy light
According to, rotation, the factors such as graphical rule influence, but the detection of Small object and great meaning, for example are taken photo by plane using unmanned plane,
Object target in Aerial Images is often smaller, is problem anxious to be resolved for small target deteection, therefore the present invention proposes
A kind of depth model for small target deteection.
It referring to Fig. 1 includes as follows that the present invention, which is a kind of small target detecting method based on Fusion Features and deep learning,
Step:
(1) prepare atlas: use the training dataset of image set PASCAL VOC2007 and PASCAL VOC2012 as
Training set uses the test data set of image set PASCAL VOC2007 as test set, above-mentioned image set be containing size not
With the online disclosed image set of target.Wherein PASCAL VOC2007 image set contains 9963 pairs containing different big altogether in this example
The image of Small object, PASCAL VOC2012 image set contain the 11540 secondary images containing different size target.Above two data
Model training of the collection for after provides training material abundant, improves the extensive degree for training rear model.
(2) the small target deteection network model based on Fusion Features and deep learning is built: referring to fig. 2, with residual error network
The preceding conv4_3 layers of basic network as feature extraction adds five layers of convolution pondization operation after residual error network and constitutes auxiliary
Feature extraction network, the characteristic pattern of more kinds of scales is obtained, with conv3, conv5, conv6, conv7, conv8, conv9 net
A variety of scale feature figures that network layers obtain are basic construction feature pyramid, are used conv5 layers and conv6 layers obtained characteristic pattern
Deconvolution and top sampling method obtain the characteristic pattern of resolution ratio same as shallow-layer characteristic pattern, divide in such a way that element is added
It is other that Fusion Features are carried out with conv5 layers of characteristic pattern to conv3 layers of high-level characteristic figure, obtain the characteristic pattern of more descriptive power；
Last conv3, conv5, conv6, conv7, conv8, the characteristic pattern that conv9 network layer obtains are input to prediction network.Prediction
Device can make duplicate prediction for the same target.It is pre- that the present invention removes the low repetition of confidence level using non-maxima suppression
It surveys.
(3) construct the target loss function of network model: the network model that training is built on training set of images considers
The classification for needing to know target to the detection of Small object and position on the image, so construction and classification error and unknown errors
The target loss function L (x, l, c, g) of related network model.
(4) for the size for further increasing training image collection, corresponding image increasing training network model: is carried out to training set
Strong operation, is then divided into the training of two stages formula for the training of network model, minimizes loss function simultaneously using gradient descent method
Weight parameter in network is successively reversely adjusted, final trained network model is obtained.
(5) small target deteection: original image to be detected is input in trained network model, obtains mapping to be checked
As the target category and position coordinates of Small Target.
The present invention is based on the target detection network models of Fusion Features, are mentioned by deeper better basic convolutional neural networks
Characteristic pattern is taken, while adding feature pyramid network structure, multiple dimensioned characteristic pattern is provided and participates in last target detection, is realized
The detection of multiscale target, and the characteristic pattern of deconvolution operation fusion high-level semantics layer is used in feature pyramid network structure
The profile information of information and shallow-layer, the preferably Small object in detection image, while being also able to satisfy requirement of real-time.
Embodiment 2
Small target detecting method based on Fusion Features and deep learning is with embodiment 1, described in step (2) of the present invention
The small target deteection network model based on Fusion Features and deep learning is built, referring to fig. 2, is carried out in accordance with the following steps:
(2a) utilizes the preceding conv4_3 layer building basic network of residual error network ResNet101: increasing not in residual error network
Connection between the identical layer of adjacent but resolution ratio, basis of formation network.The input of basic network is image to be detected, is used for
It is the characteristic pattern of each scale by image zooming-out.Due to increasing the company between the identical layer of non-conterminous but resolution ratio in residual error network
It connects, effectively prevent the loss of information after convolution operation and excitation function, while reducing appearance in deep layer network model
Gradient disappears and gradient explosion issues, therefore is conducive to the training of network model.
(2b) adds five layers of convolutional layer conv5, conv6, conv7, conv8 being sequentially reduced after basic network,
Conv9 constitutes the feature extraction network of auxiliary, obtains more kinds of scale feature figures, basic network and auxiliary to expand
Feature extraction group of networks at model proposed by the present invention feature extraction network.The characteristic pattern of multiple dimensioned multiresolution is last
Prediction provides the characteristic information in different levels, and the characteristic information of Small Target is retained, conducive to the detection of Small object.
(2c) with conv3, conv5, conv6 in the feature extraction network, conv7, conv8, conv9 network layer it is more
Kind scale feature figure is basic construction feature pyramid network structure to realize that the multiscale target in target detection detects.
(2d) predicts network using multilayer convolution filter and softmax classification layer building, as based on Fusion Features and
The end prediction interval of the small target deteection network model of deep learning, processing in the pyramid network fused conv3 and
The characteristic pattern of conv5 network layer and the conv6 not merged, conv7, conv8, the small scale features figure of conv9 network layer,
Prediction as whole network model inputs.Multiple target categories predicted in frames are obtained by convolution filter and are write from memory relatively
The unknown offset of frame is recognized, then using non-maxima suppression to the target category and prediction frame phase in multiple prediction frames
The position offset of default frame is inhibited, the target category and prediction frame obtained in final prediction frame is write from memory relatively
Recognize the position offset of frame, and according to the position coordinates of the position offset of the opposite default frame of prediction frame and default frame
Find out the position coordinates of prediction frame.Small target deteection network model based on Fusion Features and deep learning builds completion.
Embodiment 3
Small target detecting method based on Fusion Features and deep learning is with embodiment 1-2, building described in step 2c
Feature pyramid network structure to realize the multiscale target in target detection detect, referring to fig. 2, specifically include:
(2c1) with conv3, conv5, conv6 in the feature extraction network, conv7, conv8, conv9 network layer
A variety of scale feature figures are basic construction feature pyramid.
(2c2) using deconvolution operate conv5 and conv6 layers of characteristic pattern of low resolution on the middle and senior level to pyramid structure into
Row up-sampling obtains resolution ratio same as conv3 and conv5 layers of shallow-layer characteristic pattern of characteristic pattern, and the side being added according to element
Formula carries out Fusion Features to high-level characteristic figure and shallow-layer characteristic pattern, and conv3 and conv5 layers of descriptive power are had more after being merged
Characteristic pattern.
The difficult point of small target deteection is target position inaccurate, on high-rise characteristic pattern based on semantic information, lacks
Location information, texture information of Small object etc., therefore it is infeasible for only relying on high-rise characteristic pattern to carry out small target deteection.?
In depth convolutional neural networks, on the characteristic pattern of shallow-layer about the information such as the position of target, texture be than more rich, but lack
Weary high-level semantics information is especially differentiated by the characteristic pattern of the available more expressive faculty of the Fusion Features of both direction
Not only the location information comprising target also contains the information of high-level semantics layer, the detection to Small object on the higher characteristic pattern of rate
It is highly beneficial.
Embodiment 4
Small target detecting method based on Fusion Features and deep learning is with embodiment 1-3, construction net described in step 3
The target loss function L (x, l, c, g) of network model is carried out as follows:
(3.1) anchor mechanism is used on the characteristic pattern of prediction interval, is predicted in each characteristic point of each characteristic pattern
The anchor box of different length-width ratios, different scale, predicts target bezel locations with this.Inventive network model is being instructed
Target loss function when practicing is originated from the target loss function of MultiBox, and is expanded to multi-class target.In view of inspection
Classification and position that result needs to export target are surveyed, target loss function L (x, l, c, g) is by Classification Loss function Lconf(x,c)
With positioning loss function Lloc(x, l, g) composition:
Wherein, x is characterized the default frame on figure, and l is prediction block, and g is mark frame, and c is characterized the default side on figure
Category score set of the frame in each classification, Lconf(x, c) indicates the default frame on characteristic pattern on category score set c
Softmax Classification Loss function, Lloc(x, l, g) indicates that positioning loss function, N indicate and the mark matched default side of frame
Frame number, parameter alpha are set as 1 by cross validation.
(3.2) the classification score set c according to the default frame on characteristic pattern on all categories calculates softmax points
Class loss function Lconf(x, c):
Wherein, whenIndicate that i-th of default frame matches with j-th of mark frame that classification is p,It indicates
J-th of mark frame that i-th of default frame and classification are p mismatches, and 0≤i≤N, N are indicated and the mark matched default of frame
Frame number, 1≤p≤H, H are total categorical measure, and 0≤j≤T, T are the quantity for marking frame,It indicates i-th in positive sample
The average on all categories of a default frame,It indicates i-th in negative sample2A default frame is in all categories
On average, 0≤i2≤N2, N2It indicates and the mark unmatched default frame number of frame.Softmax Classification Loss function
All candidate categories are considered, from the statistical significance, complete optimal class prediction, calculate simple, significant effect.
(3.3) positioning loss function L is calculatedloc(x, l, g):
Wherein (cx, cy) is by (centre coordinate of the compensated default frame x of Δ x, Δ y), w, h are by (Δ w, Δ
H) width and height of compensated default frame,Indicate that offset is i-th of prediction frame of m,Indicate that offset is the jth of m
A prediction frame, iteration optimization all carries out a refine to frame each time, and dynamic implement is best fixed to final detection frame
Position.
The design of loss function of the present invention will put together for the optimization of classification and the optimization of position, than traditional by needle
The optimization of classification and position is separately designed more efficient in two loss functions, wherein softmax Classification Loss function is considered
All candidate categories complete optimal class prediction from the statistical significance, calculate simple, significant effect；Position loss function
A refine, best orientation of the dynamic implement to final detection frame are carried out to frame in iteration optimization each time.
Embodiment 5
Small target detecting method based on Fusion Features and deep learning is with embodiment 1-4, training net described in step 4
Network model, the size of training set of images, the sample concentrated to data carry out data enhancement operations at random in order to further increase, use
To prevent network training over-fitting, carry out as follows:
(4.1) it concentrates original sample image to carry out mirror image operation data, enables testing result that figure is effectively treated
The case where there are mirror images as in；Original sample image is concentrated to carry out the scaling in scale and length-width ratio, scaling data
Ratio is [0.5,1], and scaling length-width ratio is [0.5,2], and the artificial size for expanding the object in image set especially increases instruction
Practice the small scale image concentrated, conducive to the training of Small object；It concentrates original sample image to cut data, further expands
The diversity of training set, universality are filled；
(4.2) training parameter is arranged: the parameter needs of model training comprehensively consider, and the initial stage learning rate of model training needs
Setting it is high a bit, to accelerate to train, but cannot be too high, be not so easy to happen concussion, will lead to result and do not restrain, precision is very
Difference.With trained progress, model is slowly formed, and learning rate needs accordingly decrease, and is mainly finely adjusted at this time to model,
Weight attenuation coefficient, excessive, models fitting is ineffective, too small, and fitting speed is slow.Model training needs to be arranged gradient updating power
Weight successfully manages certain abnormal prediction, makes so that the update of certain subparameter will not be adjusted according to certain primary prediction completely
It is more reliable to obtain model training.The number of final setting model training, with the increase of frequency of training, model increasingly tends to be steady
Fixed, being further added by number effect will not change, and to sum up, the initial learning rate base-lr selectable range when present invention trains is
0.01 to 0.0005, gradient updating weight momentum value selectable range is 0.9 to 0.8, weight decaying term coefficient weight-
Decay selectable range is to be set as 0.0001 to 0.00001, and maximum frequency of training selectable range is 50000 to 100000.
(4.3) first stage first trains the network model without adding Fusion Features structure.It is not added with Fusion Features structure
Network can train a common target detection model, have certain detection accuracy, but precision needs further
It improves, the training result in the stage is prepared mainly as the basic model of next stage for the training of next stage.
(4.4) when second stage training, continue to train complete network mould based on first stage trained model
Fusion Features structure is added on the basis of the training in the first stage in the type stage, after training result can be according to Fusion Features
Be adjusted on the training result of information in the first stage so that detection accuracy in the first stage on the basis of further promoted.
The size of data of training set the training effect of depth network is influenced it is very big, bigger training data tend to so that
The network that training is completed has more generalization ability, and when practice can obtain more accurately detection effect.The present invention is to training data
Collection carries out data enhancement operations and is conducive to the training of network to increase the size of training dataset.Two stage training method will
The pressure of one stage-training has effectively decomposed two stages, and second stage is completely trained on first stage,
So that training process is more efficient, it is more reliable.
Embodiment 6
Small target detecting method based on Fusion Features and deep learning is with embodiment 1-5, Small object described in step 5
Detection carries out as follows referring to Fig. 1:
(5.1) file that the sample image in test set is converted to lmdb format, passes through the network of ResNet101 first
The high-resolution features figure of a variety of scales is extracted, the high-resolution features figure extracted remains the shallow-layer conducive to small target deteection
The information such as position, the texture about target.
(5.2) the low resolution characteristic pattern that 5 scales are sequentially reduced then is extracted by auxiliary convolutional layer, referring to fig. 2,
In, 5 layers of convolutional layer of addition are on the right of the model of network, in order to the scale of augmented features figure, the choosing of 5 layers of convolutional layer
Selecting is to combine network query function amount and property is getable, will cause the reduction of characteristic pattern scale less than 5 layers, influences finally to detect
Precision increases the scale of network more than 5 layers, and computation burden aggravates, and influences real-time.
(5.3) referring to fig. 2, from conv3, conv5, conv6, conv7, conv8, conv9 in feature extraction network
A variety of scale feature figures of network layer are basic construction feature pyramid, the pyramid network of composition see from left to right it includes
Characteristic dimension is sequentially reduced, and is in tower structure.
(5.4) referring to fig. 2, conv5 and conv6 layers of low resolution on the middle and senior level to pyramid structure are operated using deconvolution
Characteristic pattern is up-sampled to obtain resolution ratio same as conv3 and conv5 layers of shallow-layer characteristic pattern of characteristic pattern, and according to element
The mode of addition carries out Fusion Features to high-level characteristic figure and shallow-layer characteristic pattern, and the conv3 of descriptive power is had more after being merged
With conv5 layers of characteristic pattern, fused characteristic pattern contains the profile information of high-level semantics layer and the characteristic pattern letter of shallow-layer
Breath, characteristic information abundant provide more effective informations for prediction.
(5.5) prediction network is predicted using the characteristic pattern of characteristic pattern and low resolution after Fusion Features simultaneously, more
Kind characteristic pattern is used in combination, and provides more more effective information than using single features figure to carry out prediction, is conducive to improve and examine
Survey precision.
(5.6) it referring to Fig. 1, predicts in network using non-maxima suppression to target categories in multiple prediction frames and pre-
The position offset for surveying the opposite default frame of frame is inhibited, and the target category and prediction side in final prediction frame are obtained
The position offset of the opposite default frame of frame, and according to prediction the frame position offset of default frame and default frame relatively
Position coordinates find out the position coordinates of prediction frame.The associated prediction of target category and position improves net than separating respective prediction
The calculated performance of network has compressed the model of network, improves efficiency.
Provide an example on the whole below, the present invention is further described
Embodiment 7
Small target detecting method based on Fusion Features and deep learning is with embodiment 1-6, referring to Fig.1, reality of the invention
Existing scheme includes the following steps:
Step 1, the deep learning based on Fusion Features is established according to residual error network ResNet101 and feature pyramid network
Network model.
The target detection network for being currently based on deep learning is divided into two major classes: one kind is the deep learning based on candidate region
Target detection network, such as R-CNN, Fast R-CNN and Faster R-CNN；Another kind of is the deep learning mesh based on recurrence
Mark detection network, such as YOLO and SSD, the present invention proposes a kind of target detection network model based on Fusion Features, by more
Deep preferably basic convolutional neural networks extract characteristic pattern, while adding feature pyramid network structure, provide multiple dimensioned spy
Sign figure participates in last target detection, realizes the detection of multiscale target, and warp is used in feature pyramid network structure
The profile information of product operation fusion high-level semantics layer and the profile information of shallow-layer, the preferably Small object in detection image.
Referring to Fig. 2, the present invention is based on the specific steps that Fusion Features and the small target deteection network model of deep learning construct
It is rapid as follows
(1a) utilizes the preceding conv4_3 layer building basic network of residual error network ResNet101: increasing not in residual error network
Connection between the identical layer of adjacent but resolution ratio, basis of formation network.The input of basic network is image to be detected, is used for
It is the characteristic pattern of each scale by image zooming-out.Due to increasing the company between the identical layer of non-conterminous but resolution ratio in residual error network
It connects, effectively prevent the loss of information after convolution operation and excitation function, while reducing appearance in deep layer network model
Gradient disappears and gradient explosion issues, therefore is conducive to the training of network model.
(1b) referring to fig. 2, adds five layers of convolutional layer conv5, conv6, conv7 being sequentially reduced after basic network,
Conv8, conv9 constitute the feature extraction network of auxiliary, to expand to obtain more kinds of scale feature figures, basic network and
The feature extraction group of networks of auxiliary at model proposed by the present invention feature extraction network.The characteristic pattern of multiple dimensioned multiresolution is
Last prediction provides the characteristic information in different levels, and the characteristic information of Small Target is retained, and is conducive to Small object
Detection.
(1c) referring to fig. 2, with conv3, conv5, conv6 in the feature extraction network, conv7, conv8, conv9 net
A variety of scale feature figures of network layers are basic construction feature pyramid network structure to realize the multiple dimensioned mesh in target detection
Mark detection.
(1c1) with conv3, conv5, conv6 in the feature extraction network, conv7, conv8, conv9 network layer
A variety of scale feature figures are basic construction feature pyramid, referring to fig. 2, the pyramid network of composition see from left to right it includes
Characteristic dimension is sequentially reduced, and is in tower structure.
(1c2) using deconvolution operate conv5 and conv6 layers of characteristic pattern of low resolution on the middle and senior level to pyramid structure into
Row up-sampling obtains resolution ratio same as conv3 and conv5 layers of shallow-layer characteristic pattern of characteristic pattern, and the side being added according to element
Formula carries out Fusion Features to high-level characteristic figure and shallow-layer characteristic pattern, and conv3 and conv5 layers of descriptive power are had more after being merged
Characteristic pattern, fused characteristic pattern contains the profile information of high-level semantics layer and the profile information of shallow-layer, abundant
Characteristic information provides more effective informations for prediction.
(1d) predicts network referring to fig. 2, using multilayer convolution filter and softmax classification layer building, as based on spy
The end prediction interval of the small target deteection network model of sign fusion and deep learning, processing are fused in pyramid network
The characteristic pattern of conv3 and conv5 network layer and the conv6 not merged, conv7, conv8, the small scale of conv9 network layer
Characteristic pattern, the prediction as whole network model input.The target category in multiple prediction frames is obtained by convolution filter
With the unknown offset of opposite default frame, then using non-maxima suppression to target categories in multiple prediction frames and pre-
The position offset for surveying the opposite default frame of frame is inhibited, and the target category and prediction side in final prediction frame are obtained
The position offset of the opposite default frame of frame, and according to prediction the frame position offset of default frame and default frame relatively
Position coordinates find out the position coordinates of prediction frame.The associated prediction of target category and position improves net than separating respective prediction
The calculated performance of network has compressed the model of network, improves efficiency.
Step 2, the network model that training is built on training set of images.
The method being trained at present to deep learning network is broadly divided into two classes: from the unsupervised learning of lower rising and oneself
Downward supervised learning is pushed up, is trained in the present invention using top-down supervised learning method, realizes that steps are as follows:
(2a) uses the training dataset of image set PASCAL VOC2007 and PASCAL VOC2012 as training set, uses
The test data set of image set PASCAL VOC2007 is as test set.
The target loss function L (x, l, c, g) of (2b) tectonic network model.
(2b1) is predicted on characteristic pattern using convolution filter, obtains classification of the default frame on all categories
Position offset (Δ x, Δ y, the Δ w, Δ h), wherein (Δ x, Δ y) of score set c and prediction frame relative to default frame
Indicate offset of the prediction frame centre coordinate relative to default frame centre coordinate, wherein Δ w indicates that prediction frame is wide opposite
In the wide offset of default frame, wherein Δ h indicates the prediction frame high offset high relative to default frame.
The classification score set c of (2b2) according to the default frame on characteristic pattern on all categories calculates softmax points
Class loss function Lconf(x, c):
Wherein, whenIndicate that i-th of default frame matches with j-th of mark frame that classification is p,Table
Show that i-th of default frame and classification mismatch for j-th of mark frame of p, 0≤i≤N, N, which are indicated, and mark frame is matched writes from memory
Recognizing frame number, 1≤p≤H, H are total categorical measure, and 0≤j≤T, T are the quantity for marking frame,It indicates in positive sample
The average on all categories of i-th of default frame,It indicates i-th in negative sample2A default frame is in all classes
Average on not, 0≤i2≤N2, N2It indicates and the mark unmatched default frame number of frame.
(2b3) calculates positioning loss function Lloc(x, l, g):
Wherein (cx, cy) is by (centre coordinate of the compensated default frame x of Δ x, Δ y), w, h are by (Δ w, Δ
H) width and height of compensated default frame,Indicate that offset is i-th of prediction frame of m,Indicate that offset is the jth of m
A prediction frame；
(2b4) is according to Classification Loss function Lconf(x, c) and positioning loss function Lloc(x, l, g) obtains target loss letter
Number L (x, l, c, g):
Wherein, x is characterized the default frame on figure, and l is prediction block, and g is mark frame, and c is characterized the default side on figure
Category score set of the frame in each classification, Lconf(x, c) indicates the default frame on characteristic pattern on category score set c
Softmax Classification Loss function, Lloc(x, l, g) indicates that positioning loss function, N indicate and the mark matched default side of frame
Frame number, parameter alpha are set as 1 by cross validation.The design of loss function of the present invention by for classification optimization and position it is excellent
Change is put together, than it is traditional will be separately designed for the optimization of classification and position it is more efficient in two loss functions, wherein
Softmax Classification Loss function considers all candidate categories, from the statistical significance, completes optimal class prediction, calculates
Simply, significant effect；Position loss function carries out a refine to frame in iteration optimization each time, and dynamic implement is to final
Detect the best orientation of frame.
Initial learning rate base-lr when (2c) present invention training is set as 0.001, and gradient updating weight momentum value is set
It is set to 0.9, weight decaying term coefficient weight-decay is set as 0.0005, and maximum frequency of training is 80000.By network model
Training be divided into two stages formula training, using gradient descent method minimize loss function simultaneously it is layer-by-layer to the weight parameter in network
It is reversed to adjust, obtain trained network model.Wherein, two stage training method effectively decomposes the pressure of a stage-training
To two stages, the main training characteristics of first stage extract network, and second stage carries out completely on first stage
Training it is more reliable so that training process is more efficient.
Step 3, original image to be detected is input in trained network model, obtains small mesh in image to be detected
Target target category and position coordinates.
Sample image in test set is converted to the file of lmdb format by (3a), passes through the network of ResNet101 first
The high-resolution features figure of a variety of scales is extracted, the high-resolution features figure extracted remains the shallow-layer conducive to small target deteection
About information such as target position abundant, textures；
(3b) then extracts the low resolution characteristic pattern that 5 scales are sequentially reduced by auxiliary convolutional layer, wherein addition
The purpose of 5 layers of convolutional layer is for the scale of augmented features figure, and the selection of 5 layers of convolutional layer is to combine network query function amount and performance
It obtains, will cause the reduction of characteristic pattern scale less than 5 layers, influence the precision finally detected, increase network more than 5 layers
Scale, computation burden aggravate, and influence real-time.
(3c) with conv3, conv5, conv6 in the feature extraction network, conv7, conv8, conv9 network layer it is more
Kind scale feature figure is basic construction feature pyramid；
(3d) using deconvolution operate conv5 and conv6 layers of characteristic pattern of low resolution on the middle and senior level to pyramid structure into
Row up-sampling obtains resolution ratio same as conv3 and conv5 layers of shallow-layer characteristic pattern of characteristic pattern, and the side being added according to element
Formula carries out Fusion Features to high-level characteristic figure and shallow-layer characteristic pattern, and conv3 and conv5 layers of descriptive power are had more after being merged
Characteristic pattern, fused characteristic pattern contains the profile information of high-level semantics layer and the profile information of shallow-layer, abundant
Characteristic information provides more effective informations for prediction.
(3e) prediction network is predicted using the characteristic pattern of characteristic pattern and low resolution after Fusion Features simultaneously, a variety of
Characteristic pattern is used in combination, and provides more more effective information than using single features figure to carry out prediction, is conducive to improve detection
Precision；
(3f) is predicted in network using non-maxima suppression to the target category and prediction frame phase in multiple prediction frames
The position offset of default frame is inhibited, the target category and prediction frame obtained in final prediction frame is write from memory relatively
Recognize the position offset of frame, and according to the position coordinates of the position offset of the opposite default frame of prediction frame and default frame
Find out the position coordinates of prediction frame.The associated prediction of target category and position improves the calculating of network than separately respective prediction
Performance has compressed the model of network, improves efficiency.
Target detection is carried out using same image network model through the invention and SSD network model to obtain referring to Fig. 3
The testing result of the testing result and (b) SSD network model of the invention to Fig. 3 (a) observes the above two width testing result figures, can
Seeing, the present invention completely detected all three humanoid Small objects in figure, and there is no missing inspections and false retrieval, still, SSD net
Network model only detected two humanoid Small objects therein, and missing inspection has occurred.It is obvious herein for the testing result of Small object
Higher than SSD network model.
Technical effect of the invention is illustrated by following experiment.
Embodiment 8
Small target detecting method based on Fusion Features and deep learning with embodiment 1-7,
Experimental subjects
Experimental subjects is the test data set of PASCAL VOC2007.
Experimental procedure
1) Fast-RCNN network model, Faster-RCNN network model, YOLO network model, SSD300 net are used respectively
The training on the training set of image set PASCAL VOC2007 and PASCAL VOC2012 of network model and network model of the invention
Model；
2) trained Fast-RCNN network model, Faster-RCNN network model, YOLO network in 1) are used respectively
Model, SSD300 network model and network model of the invention are tested in PASCAL VOC2007 test data set, are obtained
Detection accuracy and detection speed to network model is as shown in table 1；
3) it is chosen on image set PASCAL VOC2007 using the trained model of the present invention and SSD prototype network model
Four pictures containing Small object successively carry out target detection, wherein testing result of the invention is as shown in figure 3, SSD model net
The testing result of network model is as shown in Figure 4.
Experimental data statistics:
Respectively using trained Fast-RCNN network model, Faster-RCNN network model, YOLO network model,
SSD300 network model and network model of the invention are tested in PASCAL VOC2007 test data set, obtained inspection
It surveys precision and detection speed is as shown in table 1:
Table 1
| Algorithm model | Data set | Detection accuracy (%) | It detects speed (frame/second) |
| Fast-RCNN | 07++12 | 68.4 | 3 |
| Faster-RCNN | 07++12 | 70.4 | 5 |
| YOLO | 07++12 | 57.9 | 47 |
| SSD300 | 07++12 | 72.4 | 59 |
| The present invention | 07++12 | 78.5 | 27 |
As it can be seen from table 1 detection accuracy and detection speed ratio that network model of the invention is tested on test set
Fast-RCNN network model, the detection accuracy of Faster-RCNN network model and detection speed are all significantly increased, with
SSD300 network model, YOLO network model are compared, although network model of the invention does not have SSD300 net in detection speed
Network model, YOLO network model are fast, but detection accuracy has and is obviously improved.To sum up, inspection of the present invention in five kinds of networks
It surveys precision and has reached 78.5%, detection effect is that best, of the invention detection speed reaches that 27 frames are per second, and real-time detection
It is per second that rate request is greater than 25 frames, therefore the present invention meets real-time detection requirement.
Embodiment 9
Small target detecting method based on Fusion Features and deep learning is the same as embodiment 1-7, experiment condition and experiment content
With embodiment 8
The testing result of comparison diagram 3, Fig. 4, Fig. 5 and Fig. 6, wherein success in Fig. 3 inventive network model inspection result (a)
It detected all three humanoid Small objects, and Fig. 3, SSD300 network model testing result (b) only detected therein two
A humanoid Small object, the case where missing inspection has occurred.Fig. 4 is that an image about dinner party dinner has there are 8 personages in figure
Different degrees of blocks, while dinner scene determines that the light of entire image is integrally darker, observes Fig. 4 (b) SSD network model
Testing result, only detected 4 people in figure, cannot be effectively treated and block and the case where dark, Fig. 4 (a) is the present invention
Testing result, successfully detected 7 personages in figure, model inspection precision of the present invention is apparently higher than SSD network model, right
Blocking in image has preferable robustness with light.Fig. 5 is an image about personage's landscape, and background is mountain, more empty
The case where spacious, there are the personages' of four different sizes in figure, and all back to camera lens, and character features are relative to face camera lens meeting
There are many missing, and there are more hiding relation between personage, Fig. 5 (b) is the testing result of SSD model, only detected one most
Large-sized personage, missing inspection its excess-three individual's object.Fig. 5 (a) is testing result of the invention, successfully detected three people
Situations such as blocking in image can be effectively treated in object, and character features lack.Fig. 6 is an image about family party, figure
There are the things of plurality of classes as in, and have 11 personages of serious mutually hiding relation, and scene is very complicated, and Fig. 6 (b) is
The testing result of SSD model only detected two personages in image, a large amount of missing inspection situation have occurred.Fig. 6 (a) is this
The testing result of invention successfully detected four personages therein, although missing inspection also has occurred, final detection accuracy
Still it is higher than SSD, in contrast to SSD model, also there is preferable detection effect to complex scene.To sum up, network mould of the invention
Type is more preferable to the detection effect of Small object, and it is darker to be effectively treated image light, and scene is complicated, and multiple target and target, which have, blocks
Situation, the rare generation of missing inspection situation.The difficult point of small target deteection is target position inaccurate, with semanteme on high-rise characteristic pattern
Based on information, lack location information, texture information of Small object etc., therefore only relies on high-rise characteristic pattern and carry out small target deteection
It is infeasible.It about information such as the position of target, textures is to compare on the characteristic pattern of shallow-layer in depth convolutional neural networks
It is abundant, but it is a lack of high-level semantics information, the available more expressive faculty of Fusion Features that the present invention passes through both direction
Characteristic pattern, not only also contain high-level semantics layer comprising the location information of target especially on the higher characteristic pattern of resolution ratio
Information, it is highly beneficial to the detection of Small object.
The invention discloses a kind of small target detecting method based on Fusion Features and deep learning, solves to Small object
Detection accuracy difference and real time problems.Its implementation is: extracting high score by the network model of deeper better ResNet101
Resolution characteristic pattern；5 low resolution characteristic patterns being sequentially reduced, the scale of augmented features figure are extracted by auxiliary convolutional layer；It is logical
Feature pyramid network model is crossed, multiple dimensioned characteristic pattern is obtained；It is operated in feature pyramid network structure using deconvolution
Merge the profile information of high-level semantics layer and the profile information of shallow-layer；By different scale and the characteristic pattern of fusion characteristics into
Row target prediction；Using non-maxima suppression to multiple prediction frames and multiple category scores, the frame of final target is obtained
Position and classification information.The present invention has under the requirement for guaranteeing real-time detection, it is ensured that the high-precision advantage of small target deteection, energy
Small object in image is fast and accurately detected, the target real-time detection that can be used in unmanned plane.
Claims (6)
1. a kind of small target detecting method based on Fusion Features and deep learning, which is characterized in that comprise the following steps that
(1) prepare atlas: using the training dataset of image set PASCAL VOC2007 and PASCAL VOC2012 as training
Collection, uses the test data set of image set PASCAL VOC2007 as test set, above-mentioned image set is containing mesh of different sizes
Target disclosed image set on the net；
(2) the small target deteection network model based on Fusion Features and deep learning is built: using residual error network as feature extraction
Basic network, the feature extraction network that the operation of five layers of convolution pondization constitutes auxiliary is added after residual error network, is obtained more
The characteristic pattern of kind scale is that basic construction feature pyramid is obtained using deconvolution and top sampling method with a variety of scale feature figures
To the characteristic pattern of resolution ratio same as shallow-layer characteristic pattern, to high-level characteristic figure and shallow-layer characteristic pattern in such a way that element is added
Fusion Features are carried out, the characteristic pattern of more descriptive power is obtained；Finally addition prediction network, is pressed down using polygon frame and non-maximum
Method processed obtains Small object classification and position；
(3) the target loss function of network model: the network model that training is built on training set of images, tectonic network is constructed
The target loss function L (x, l, c, g) of model；
(4) training network model: the training of network model is divided into the training of two stages formula, is minimized and is lost using gradient descent method
Function is simultaneously successively reversely adjusted the weight parameter in network, obtains final trained network model；
(5) small target deteection: original image to be detected is input in trained network model, is obtained in image to be detected
The target category and position coordinates of Small object.
2. the small target detecting method according to claim 1 based on Fusion Features and deep learning, which is characterized in that step
Suddenly the small target deteection network model based on Fusion Features and deep learning is built described in (2), is comprised the following steps that
(2a) constructs basic network using residual error network ResNet101: it is identical to increase non-conterminous but resolution ratio in residual error network
Layer between connection, basis of formation network；The input of basic network is image to be detected, for being each ruler by image zooming-out
The characteristic pattern of degree；
(2b) adds five layers of convolutional layer being sequentially reduced after basic network, the feature extraction network of auxiliary is constituted, to expand
Fill to obtain more kinds of scale feature figures, the feature extraction network of basic network and auxiliary forms the feature extraction net of the model
Network；
(2c) construction feature pyramid network structure is to realize that the multiscale target in target detection detects:
(2d) predicts network using multilayer convolution filter and softmax classification layer building, as based on Fusion Features and depth
The end prediction interval of the small target deteection network model of study handles fused a variety of scale features from pyramid network
Small scale features figure in figure and feature extraction network makees the prediction output of whole network model.
3. the small target detecting method according to claim 1 based on Fusion Features and deep learning, which is characterized in that step
Construction feature pyramid network structure described in rapid 2c is specific to wrap to realize that the multiscale target in target detection detects
It includes:
(2c1) is basic construction feature pyramid with a variety of scale feature figures from feature extraction network；
(2c2) operates to be up-sampled to obtain to the characteristic pattern of high-rise low resolution using deconvolution similarly to be divided with shallow-layer characteristic pattern
Resolution, and Fusion Features are carried out to high-level characteristic figure and shallow-layer characteristic pattern in such a way that element is added, energy is more described
The characteristic pattern of power.
4. the small target detecting method according to claim 1 based on Fusion Features and deep learning, which is characterized in that step
The target loss function L (x, l, c, g) of tectonic network model described in rapid 3 is carried out as follows:
(3.1) target loss function L (x, l, c, g) is by Classification Loss function Lconf(x, c) and positioning loss function Lloc(x,l,g)
Composition:
Wherein, x is characterized the default frame on figure, and l is prediction block, and g is mark frame, and the default frame that c is characterized on figure exists
Category score set in each classification, Lconf(x, c) indicates the default frame on characteristic pattern on category score set c
Softmax Classification Loss function, Lloc(x, l, g) indicates that positioning loss function, N indicate and the mark matched default frame of frame
Number, parameter alpha are set as 1 by cross validation；
(3.2) the classification score set c according to the default frame on characteristic pattern on all categories calculates softmax classification damage
Lose function Lconf(x, c):
Wherein, whenIndicate that i-th of default frame matches with j-th of mark frame that classification is p,Indicate i-th
J-th of mark frame that a default frame and classification are p mismatches, and 0≤i≤N, N are indicated and the mark matched default side of frame
Frame number, 1≤p≤H, H are total categorical measure, and 0≤j≤T, T are the quantity for marking frame,It indicates i-th in positive sample
Default the average on all categories of frame,It indicates i-th in negative sample2A default frame is on all categories
Average, 0≤i2≤N2, N2It indicates and the mark unmatched default frame number of frame；
(3.3) positioning loss function L is calculatedloc(x, l, g):
Wherein (cx, cy) is by (centre coordinate of the compensated default frame x of Δ x, Δ y), w, h are by (Δ w, Δ h) benefit
The width and height of default frame after repaying,Indicate that offset is i-th of prediction frame of m,Expression offset is that j-th of m is pre-
Survey frame.
5. the small target detecting method according to claim 1 based on Fusion Features and deep learning, which is characterized in that step
Training network model described in rapid 4, the sample concentrated to data carry out data enhancement operations at random, prevent network training excessively quasi-
It closes, comprises the following steps that
(4.1) original sample image is concentrated to carry out mirror image operation data；Original sample image is concentrated to carry out ruler data
Scaling on degree and length-width ratio, scaling ratio are [0.5,1], and scaling length-width ratio is [0.5,2]；Data are concentrated originally
Sample image is cut；
(4.2) training parameter is arranged: initial learning rate base-lr when training is set as 0.001, gradient updating weight momentum
Value is set as 0.9, and weight decaying term coefficient weight-decay is set as 0.0005, and maximum frequency of training is 80000；
(4.3) first stage first trains the network model without adding Fusion Features structure；
(4.4) when second stage training, continue to train complete network model based on first stage trained model.
6. the small target detecting method according to claim 1 based on Fusion Features and deep learning, which is characterized in that step
Small target deteection described in rapid 5 carries out as follows:
(5.1) network that image to be detected passes through ResNet101 first extracts the high-resolution features figure of a variety of scales；
(5.2) the low resolution characteristic pattern that 5 scales are sequentially reduced then is extracted by auxiliary convolutional layer；
(5.3) characteristic pattern of a variety of scales extracted before is built into feature pyramid network；
(5.4) profile information and shallow-layer of deconvolution operation fusion high-level semantics layer are used in feature pyramid network structure
Profile information；
(5.5) prediction network is predicted using the characteristic pattern of characteristic pattern and low resolution after Fusion Features simultaneously；
(5.6) using non-maxima suppression to the position of target category and the opposite default frame of prediction frame in multiple prediction frames
It sets offset to be inhibited, the position for obtaining the target category and the opposite default frame of prediction frame in final prediction frame is inclined
Shifting amount, and prediction frame is found out according to the position offset of the opposite default frame of prediction frame and the position coordinates of default frame
Position coordinates.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811005464.6A CN109344821A (en) | 2018-08-30 | 2018-08-30 | Small target detecting method based on Fusion Features and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811005464.6A CN109344821A (en) | 2018-08-30 | 2018-08-30 | Small target detecting method based on Fusion Features and deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109344821A true CN109344821A (en) | 2019-02-15 |
Family
ID=65296917
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811005464.6A Pending CN109344821A (en) | 2018-08-30 | 2018-08-30 | Small target detecting method based on Fusion Features and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109344821A (en) |
Cited By (143)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109871903A (en) * | 2019-03-11 | 2019-06-11 | 南京邮电大学 | A kind of object detection method based on end-to-end depth network and confrontation study |
| CN109886357A (en) * | 2019-03-13 | 2019-06-14 | 哈尔滨工程大学 | A kind of adaptive weighting deep learning objective classification method based on Fusion Features |
| CN109886230A (en) * | 2019-02-28 | 2019-06-14 | 中南大学 | A kind of image object detection method and device |
| CN109919215A (en) * | 2019-02-27 | 2019-06-21 | 中国电子科技集团公司第二十八研究所 | The object detection method of feature pyramid network is improved based on clustering algorithm |
| CN109978014A (en) * | 2019-03-06 | 2019-07-05 | 华南理工大学 | A kind of flexible base board defect inspection method merging intensive connection structure |
| CN110008953A (en) * | 2019-03-29 | 2019-07-12 | 华南理工大学 | Potential target Area generation method based on the fusion of convolutional neural networks multilayer feature |
| CN110008853A (en) * | 2019-03-15 | 2019-07-12 | 华南理工大学 | Pedestrian detection network and model training method, detection method, medium, equipment |
| CN110047058A (en) * | 2019-03-25 | 2019-07-23 | 杭州电子科技大学 | A kind of image interfusion method based on residual pyramid |
| CN110084299A (en) * | 2019-04-24 | 2019-08-02 | 中国人民解放军国防科技大学 | target detection method and device based on multi-head fusion attention |
| CN110084124A (en) * | 2019-03-28 | 2019-08-02 | 北京大学 | Feature based on feature pyramid network enhances object detection method |
| CN110110599A (en) * | 2019-04-03 | 2019-08-09 | 天津大学 | A kind of Remote Sensing Target detection method based on multi-scale feature fusion |
| CN110110719A (en) * | 2019-03-27 | 2019-08-09 | 浙江工业大学 | A kind of object detection method based on attention layer region convolutional neural networks |
| CN110135373A (en) * | 2019-05-20 | 2019-08-16 | 北京探境科技有限公司 | A kind of multiple dimensioned face recognition method, system and electronic equipment |
| CN110135503A (en) * | 2019-05-19 | 2019-08-16 | 重庆理工大学 | One kind put together machines people's part depth study recognition methods |
| CN110287927A (en) * | 2019-07-01 | 2019-09-27 | 西安电子科技大学 | Based on the multiple dimensioned remote sensing image object detection method with context study of depth |
| CN110287849A (en) * | 2019-06-20 | 2019-09-27 | 北京工业大学 | A kind of lightweight depth network image object detection method suitable for raspberry pie |
| CN110309792A (en) * | 2019-07-04 | 2019-10-08 | 电子科技大学 | Indoor human detection method based on component model |
| CN110348447A (en) * | 2019-06-27 | 2019-10-18 | 电子科技大学 | A kind of multiple-model integration object detection method with rich space information |
| CN110400305A (en) * | 2019-07-26 | 2019-11-01 | 哈尔滨理工大学 | A kind of object detection method based on deep learning |
| CN110415795A (en) * | 2019-08-02 | 2019-11-05 | 杭州智团信息技术有限公司 | A kind of recognition methods of fluorescent staining CTC image |
| CN110427970A (en) * | 2019-07-05 | 2019-11-08 | 平安科技（深圳）有限公司 | Image classification method, device, computer equipment and storage medium |
| CN110427823A (en) * | 2019-06-28 | 2019-11-08 | 北京大学 | Joint objective detection method and device based on video frame and pulse array signals |
| CN110503112A (en) * | 2019-08-27 | 2019-11-26 | 电子科技大学 | A kind of small target deteection of Enhanced feature study and recognition methods |
| CN110533084A (en) * | 2019-08-12 | 2019-12-03 | 长安大学 | A kind of multiscale target detection method based on from attention mechanism |
| CN110533105A (en) * | 2019-08-30 | 2019-12-03 | 北京市商汤科技开发有限公司 | A kind of object detection method and device, electronic equipment and storage medium |
| CN110598788A (en) * | 2019-09-12 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN110647816A (en) * | 2019-08-26 | 2020-01-03 | 合肥工业大学 | Target detection method for real-time monitoring of goods shelf medicines |
| CN110689021A (en) * | 2019-10-17 | 2020-01-14 | 哈尔滨理工大学 | Real-time target detection method in low-visibility environment based on deep learning |
| CN110728200A (en) * | 2019-09-23 | 2020-01-24 | 武汉大学 | Real-time pedestrian detection method and system based on deep learning |
| CN110728658A (en) * | 2019-09-16 | 2020-01-24 | 武汉大学 | High-resolution remote sensing image weak target detection method based on deep learning |
| CN110738647A (en) * | 2019-10-12 | 2020-01-31 | 成都考拉悠然科技有限公司 | Mouse detection method integrating multi-receptive-field feature mapping and Gaussian probability model |
| CN110766721A (en) * | 2019-09-30 | 2020-02-07 | 南京航空航天大学 | Carrier landing cooperative target detection method based on airborne vision |
| CN110796115A (en) * | 2019-11-08 | 2020-02-14 | 厦门美图之家科技有限公司 | Image detection method and device, electronic equipment and readable storage medium |
| CN110796640A (en) * | 2019-09-29 | 2020-02-14 | 郑州金惠计算机系统工程有限公司 | Small target defect detection method and device, electronic equipment and storage medium |
| CN110807384A (en) * | 2019-10-24 | 2020-02-18 | 华东计算技术研究所（中国电子科技集团公司第三十二研究所） | Small target detection method and system under low visibility |
| CN110807372A (en) * | 2019-10-15 | 2020-02-18 | 哈尔滨工程大学 | Rapid optical remote sensing target identification method based on depth feature recombination |
| CN110826514A (en) * | 2019-11-13 | 2020-02-21 | 国网青海省电力公司海东供电公司 | Construction site violation intelligent identification method based on deep learning |
| CN110852241A (en) * | 2019-11-06 | 2020-02-28 | 西安交通大学 | Small target detection method applied to nursing robot |
| CN110852317A (en) * | 2019-09-16 | 2020-02-28 | 浙江工业大学 | Small-scale target detection method based on weak edge |
| CN110956108A (en) * | 2019-11-22 | 2020-04-03 | 华南理工大学 | Small frequency scale detection method based on characteristic pyramid |
| CN111008562A (en) * | 2019-10-31 | 2020-04-14 | 北京城建设计发展集团股份有限公司 | Human-vehicle target detection method with feature map depth fusion |
| CN111079620A (en) * | 2019-12-10 | 2020-04-28 | 北京小蝇科技有限责任公司 | Leukocyte image detection and identification model construction method based on transfer learning and application |
| CN111144494A (en) * | 2019-12-27 | 2020-05-12 | 睿魔智能科技（深圳）有限公司 | Object detection model training method, object detection device, object detection equipment and object detection medium |
| CN111178213A (en) * | 2019-12-23 | 2020-05-19 | 大连理工大学 | Aerial photography vehicle detection method based on deep learning |
| CN111191535A (en) * | 2019-12-18 | 2020-05-22 | 南京理工大学 | Pedestrian detection model construction method based on deep learning and pedestrian detection method |
| CN111191508A (en) * | 2019-11-28 | 2020-05-22 | 浙江省北大信息技术高等研究院 | Face recognition method and device |
| CN111209887A (en) * | 2020-01-15 | 2020-05-29 | 西安电子科技大学 | SSD model optimization method for small target detection |
| CN111259758A (en) * | 2020-01-13 | 2020-06-09 | 中国矿业大学 | Two-stage remote sensing image target detection method for dense area |
| CN111274970A (en) * | 2020-01-21 | 2020-06-12 | 南京航空航天大学 | Traffic sign detection method based on improved YOLO v3 algorithm |
| CN111291687A (en) * | 2020-02-11 | 2020-06-16 | 青岛联合创智科技有限公司 | 3D human body action standard identification method |
| CN111292377A (en) * | 2020-03-11 | 2020-06-16 | 南京旷云科技有限公司 | Target detection method, target detection device, computer equipment and storage medium |
| CN111340039A (en) * | 2020-02-12 | 2020-06-26 | 杰创智能科技股份有限公司 | Target detection method based on feature selection |
| CN111340126A (en) * | 2020-03-03 | 2020-06-26 | 腾讯云计算（北京）有限责任公司 | Article identification method and device, computer equipment and storage medium |
| CN111382787A (en) * | 2020-03-06 | 2020-07-07 | 芯薇(上海)智能科技有限公司 | Target detection method based on deep learning |
| CN111401290A (en) * | 2020-03-24 | 2020-07-10 | 杭州博雅鸿图视频技术有限公司 | Face detection method and system and computer readable storage medium |
| CN111444939A (en) * | 2020-02-19 | 2020-07-24 | 山东大学 | Small-scale equipment component detection method based on weak supervision cooperative learning in open scene of power field |
| CN111444865A (en) * | 2020-03-31 | 2020-07-24 | 盐城禅图智能科技有限公司 | Multi-scale target detection method based on gradual refinement |
| CN111461217A (en) * | 2020-03-31 | 2020-07-28 | 华南理工大学 | Aerial image small target detection method based on feature fusion and up-sampling |
| CN111462090A (en) * | 2020-04-01 | 2020-07-28 | 上饶市中科院云计算中心大数据研究院 | Multi-scale image target detection method |
| CN111461039A (en) * | 2020-04-07 | 2020-07-28 | 电子科技大学 | Landmark identification method based on multi-scale feature fusion |
| CN111461085A (en) * | 2020-06-01 | 2020-07-28 | 青岛大学 | Log detection method based on weight sharing and upper and lower feature fusion |
| CN111539782A (en) * | 2020-01-21 | 2020-08-14 | 中国银联股份有限公司 | Merchant information data processing method and system based on deep learning |
| CN111553280A (en) * | 2020-04-28 | 2020-08-18 | 上海无线电设备研究所 | Target part identification method based on deep learning |
| CN111582093A (en) * | 2020-04-27 | 2020-08-25 | 北京工业大学 | Automatic small target detection method in high-resolution image based on computer vision and deep learning |
| CN111597857A (en) * | 2019-02-20 | 2020-08-28 | 中科院微电子研究所昆山分所 | Logistics package detection method, device and equipment and readable storage medium |
| CN111612797A (en) * | 2020-03-03 | 2020-09-01 | 江苏大学 | Rice image information processing system |
| CN111626208A (en) * | 2020-05-27 | 2020-09-04 | 北京百度网讯科技有限公司 | Method and apparatus for detecting small targets |
| CN111652288A (en) * | 2020-05-11 | 2020-09-11 | 北京航天自动控制研究所 | Improved SSD small target detection method based on dense feature pyramid |
| CN111666822A (en) * | 2020-05-13 | 2020-09-15 | 飒铂智能科技有限责任公司 | Low-altitude unmanned aerial vehicle target detection method and system based on deep learning |
| CN111723737A (en) * | 2020-06-19 | 2020-09-29 | 河南科技大学 | Target detection method based on multi-scale matching strategy deep feature learning |
| CN111738212A (en) * | 2020-07-20 | 2020-10-02 | 平安国际智慧城市科技股份有限公司 | Traffic signal lamp identification method, device, equipment and medium based on artificial intelligence |
| CN111738036A (en) * | 2019-03-25 | 2020-10-02 | 北京四维图新科技股份有限公司 | Image processing method, device, equipment and storage medium |
| CN111767878A (en) * | 2020-07-03 | 2020-10-13 | 中国科学院自动化研究所 | Deep learning-based traffic sign detection method and system in embedded device |
| CN111767944A (en) * | 2020-05-27 | 2020-10-13 | 重庆大学 | Deep learning-based single-stage detector design method suitable for multi-scale target detection |
| CN111783685A (en) * | 2020-05-08 | 2020-10-16 | 西安建筑科技大学 | Target detection improved algorithm based on single-stage network model |
| CN111797846A (en) * | 2019-04-08 | 2020-10-20 | 四川大学 | Feedback type target detection method based on characteristic pyramid network |
| CN111860587A (en) * | 2020-06-12 | 2020-10-30 | 长安大学 | Method for detecting small target of picture |
| CN111860637A (en) * | 2020-07-17 | 2020-10-30 | 河南科技大学 | Single-shot multi-frame infrared target detection method |
| CN111898668A (en) * | 2020-07-24 | 2020-11-06 | 佛山市南海区广工大数控装备协同创新研究院 | Small target object detection method based on deep learning |
| CN111914599A (en) * | 2019-05-09 | 2020-11-10 | 四川大学 | Fine-grained bird recognition method based on semantic information multi-layer feature fusion |
| CN111914861A (en) * | 2019-05-08 | 2020-11-10 | 北京字节跳动网络技术有限公司 | Target detection method and device |
| CN111931799A (en) * | 2019-05-13 | 2020-11-13 | 百度在线网络技术（北京）有限公司 | Image recognition method and device |
| CN111950515A (en) * | 2020-08-26 | 2020-11-17 | 重庆邮电大学 | Semantic feature pyramid network-based small face detection method |
| CN111985325A (en) * | 2020-07-15 | 2020-11-24 | 国网河南省电力公司 | Aerial small target rapid identification method in extra-high voltage environment evaluation |
| CN112001403A (en) * | 2020-08-11 | 2020-11-27 | 北京化工大学 | Image contour detection method and system |
| CN112016512A (en) * | 2020-09-08 | 2020-12-01 | 重庆市地理信息和遥感应用中心 | Remote sensing image small target detection method based on feedback type multi-scale training |
| CN112068111A (en) * | 2020-08-13 | 2020-12-11 | 中国人民解放军海军工程大学 | Unmanned aerial vehicle target detection method based on multi-sensor information fusion |
| CN112070043A (en) * | 2020-09-15 | 2020-12-11 | 常熟理工学院 | Safety helmet wearing convolutional network based on feature fusion, training and detecting method |
| CN112084838A (en) * | 2020-07-20 | 2020-12-15 | 河海大学 | Workshop safety helmet detection method |
| CN112131996A (en) * | 2020-09-17 | 2020-12-25 | 东南大学 | Road side image multi-scale pedestrian rapid detection method based on channel separation convolution |
| CN112132828A (en) * | 2020-10-15 | 2020-12-25 | 浙江工商大学 | Film flaw detection method based on deep learning |
| CN112149533A (en) * | 2020-09-10 | 2020-12-29 | 上海电力大学 | Target detection method based on improved SSD model |
| CN112184711A (en) * | 2020-11-05 | 2021-01-05 | 杭州青枭科技有限公司 | Photovoltaic module defect detection and positioning method and system |
| CN112183203A (en) * | 2020-08-26 | 2021-01-05 | 北京工业大学 | Real-time traffic sign detection method based on multi-scale pixel feature fusion |
| CN112183579A (en) * | 2020-09-01 | 2021-01-05 | 国网宁夏电力有限公司检修公司 | Method, medium and system for detecting micro target |
| CN112270279A (en) * | 2020-11-02 | 2021-01-26 | 重庆邮电大学 | Multi-dimensional-based remote sensing image micro-target detection method |
| CN112270366A (en) * | 2020-11-02 | 2021-01-26 | 重庆邮电大学 | Micro target detection method based on self-adaptive multi-feature fusion |
| CN112287895A (en) * | 2020-11-26 | 2021-01-29 | 山东捷讯通信技术有限公司 | Model construction method, recognition method and system for river drain outlet detection |
| CN112307889A (en) * | 2020-09-22 | 2021-02-02 | 北京航空航天大学 | Face detection algorithm based on small auxiliary network |
| CN112308040A (en) * | 2020-11-26 | 2021-02-02 | 山东捷讯通信技术有限公司 | River sewage outlet detection method and system based on high-definition images |
| CN112307853A (en) * | 2019-08-02 | 2021-02-02 | 成都天府新区光启未来技术研究院 | Detection method of aerial image, storage medium and electronic device |
| CN112329808A (en) * | 2020-09-25 | 2021-02-05 | 武汉光谷信息技术股份有限公司 | Optimization method and system of Deeplab semantic segmentation algorithm |
| CN112348036A (en) * | 2020-11-26 | 2021-02-09 | 北京工业大学 | Self-adaptive target detection method based on lightweight residual learning and deconvolution cascade |
| CN112348042A (en) * | 2019-08-08 | 2021-02-09 | 中国人民解放军火箭军工程大学 | Improved YOLOv 3-based infrared target detection method |
| CN112418165A (en) * | 2020-12-07 | 2021-02-26 | 武汉工程大学 | Small-size target detection method and device based on improved cascade neural network |
| CN112508935A (en) * | 2020-12-22 | 2021-03-16 | 郑州金惠计算机系统工程有限公司 | Product packaging detection method and system based on deep learning and product packaging sorting system |
| CN112633352A (en) * | 2020-12-18 | 2021-04-09 | 浙江大华技术股份有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN112651441A (en) * | 2020-12-25 | 2021-04-13 | 深圳市信义科技有限公司 | Fine-grained non-motor vehicle feature detection method, storage medium and computer equipment |
| WO2021083126A1 (en) * | 2019-10-31 | 2021-05-06 | 北京市商汤科技开发有限公司 | Target detection and intelligent driving methods and apparatuses, device, and storage medium |
| CN112766361A (en) * | 2021-01-18 | 2021-05-07 | 山东师范大学 | Target fruit detection method and detection system under homochromatic background |
| CN112841154A (en) * | 2020-12-29 | 2021-05-28 | 长沙湘丰智能装备股份有限公司 | Disease and pest control system based on artificial intelligence |
| CN112966659A (en) * | 2021-03-31 | 2021-06-15 | 北湾科技(武汉)有限公司 | Video image small target detection method based on deep learning |
| CN113052210A (en) * | 2021-03-11 | 2021-06-29 | 北京工业大学 | Fast low-illumination target detection method based on convolutional neural network |
| CN113095316A (en) * | 2021-04-15 | 2021-07-09 | 西安电子科技大学 | Image rotation target detection method based on multilevel fusion and angular point offset |
| CN113112866A (en) * | 2021-04-14 | 2021-07-13 | 深圳市旗扬特种装备技术工程有限公司 | Intelligent traffic early warning method and intelligent traffic early warning system |
| CN113159079A (en) * | 2020-01-07 | 2021-07-23 | 顺丰科技有限公司 | Target detection method, target detection device, computer equipment and storage medium |
| CN113283499A (en) * | 2021-05-24 | 2021-08-20 | 南京航空航天大学 | Three-dimensional fabric weaving density detection method based on deep learning |
| CN113408321A (en) * | 2020-03-16 | 2021-09-17 | 中国人民解放军战略支援部队信息工程大学 | Real-time target detection method and device for lightweight image and video data |
| CN113420660A (en) * | 2021-06-23 | 2021-09-21 | 西安电子科技大学 | Infrared image target detection model construction method, prediction method and system |
| CN113420819A (en) * | 2021-06-25 | 2021-09-21 | 西北工业大学 | Lightweight underwater target detection method based on CenterNet |
| CN113486781A (en) * | 2021-07-02 | 2021-10-08 | 国网电力科学研究院有限公司 | Electric power inspection method and device based on deep learning model |
| CN113591568A (en) * | 2021-06-28 | 2021-11-02 | 北京百度网讯科技有限公司 | Target detection method, training method of target detection model and device thereof |
| CN113657285A (en) * | 2021-08-18 | 2021-11-16 | 中国人民解放军陆军装甲兵学院 | Real-time target detection method based on small-scale target |
| CN113657196A (en) * | 2021-07-27 | 2021-11-16 | 中国科学院自动化研究所 | SAR image target detection method and device, electronic equipment and storage medium |
| CN113705341A (en) * | 2021-07-16 | 2021-11-26 | 国家石油天然气管网集团有限公司 | Small-scale face detection method based on generation countermeasure network |
| CN113705359A (en) * | 2021-08-03 | 2021-11-26 | 江南大学 | Multi-scale clothes detection system and method based on washing machine drum image |
| CN113723181A (en) * | 2021-07-20 | 2021-11-30 | 深圳大学 | Unmanned aerial vehicle aerial photography target detection method and device |
| CN113792660A (en) * | 2021-09-15 | 2021-12-14 | 江苏科技大学 | Pedestrian detection method, system, medium and equipment based on improved YOLOv3 network |
| CN113792631A (en) * | 2021-08-31 | 2021-12-14 | 电子科技大学 | Aircraft detection and tracking method based on multi-scale self-adaption and side-domain attention |
| CN113807231A (en) * | 2021-09-14 | 2021-12-17 | 西安电子科技大学 | X-ray contraband detection method based on UNET (UNET) downsampling convolutional neural network |
| CN113887649A (en) * | 2021-10-19 | 2022-01-04 | 齐鲁工业大学 | Target detection method based on fusion of deep-layer features and shallow-layer features |
| CN113963265A (en) * | 2021-09-13 | 2022-01-21 | 北京理工雷科电子信息技术有限公司 | Small sample small target rapid detection and identification method for complex remote sensing land environment |
| CN114241274A (en) * | 2021-11-30 | 2022-03-25 | 电子科技大学 | Small target detection method based on super-resolution multi-scale feature fusion |
| CN114359680A (en) * | 2021-12-17 | 2022-04-15 | 中国人民解放军海军工程大学 | Panoramic vision water surface target detection method based on deep learning |
| CN114373121A (en) * | 2021-09-08 | 2022-04-19 | 武汉众智数字技术有限公司 | Method and system for improving small target detection of yolov5 network |
| CN114863267A (en) * | 2022-03-30 | 2022-08-05 | 南京邮电大学 | Aerial tree number accurate statistical method based on multi-track intelligent prediction |
| CN115205855A (en) * | 2022-07-08 | 2022-10-18 | 中国人民解放军国防科技大学 | Vehicle target identification method, device and equipment fusing multi-scale semantic information |
| CN115272700A (en) * | 2022-08-03 | 2022-11-01 | 河南大学 | Improved SSD small target detection method based on multi-scale feature self-enhancement |
| WO2022227770A1 (en) * | 2021-04-28 | 2022-11-03 | 北京百度网讯科技有限公司 | Method for training target object detection model, target object detection method, and device |
| CN115841460A (en) * | 2022-11-21 | 2023-03-24 | 国网湖北省电力有限公司超高压公司 | High-precision hardware crack image detection and feature extraction method under complex background |
| CN115984846A (en) * | 2023-02-06 | 2023-04-18 | 山东省人工智能研究院 | Intelligent identification method for small target in high-resolution image based on deep learning |
| CN116070789A (en) * | 2023-03-17 | 2023-05-05 | 北京茗禾科技有限公司 | Artificial intelligence-based single-yield prediction method for mature-period rice and wheat |
| JP2023527615A (en) * | 2021-04-28 | 2023-06-30 | ベイジン バイドゥ ネットコム サイエンス テクノロジー カンパニー リミテッド | Target object detection model training method, target object detection method, device, electronic device, storage medium and computer program |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
| CN108345911A (en) * | 2018-04-16 | 2018-07-31 | 东北大学 | Surface Defects in Steel Plate detection method based on convolutional neural networks multi-stage characteristics |
-
2018
- 2018-08-30 CN CN201811005464.6A patent/CN109344821A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
| CN108345911A (en) * | 2018-04-16 | 2018-07-31 | 东北大学 | Surface Defects in Steel Plate detection method based on convolutional neural networks multi-stage characteristics |
Non-Patent Citations (1)
| Title |
|---|
| CHENG-YANG FU ET AL: "DSSD : Deconvolutional Single Shot Detector", 《ARXIV:1701.06659V1 [CS.CV]》 * |
Cited By (217)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111597857A (en) * | 2019-02-20 | 2020-08-28 | 中科院微电子研究所昆山分所 | Logistics package detection method, device and equipment and readable storage medium |
| CN111597857B (en) * | 2019-02-20 | 2023-09-26 | 中科微至科技股份有限公司 | Logistics package detection method, device, equipment and readable storage medium |
| CN109919215A (en) * | 2019-02-27 | 2019-06-21 | 中国电子科技集团公司第二十八研究所 | The object detection method of feature pyramid network is improved based on clustering algorithm |
| CN109886230A (en) * | 2019-02-28 | 2019-06-14 | 中南大学 | A kind of image object detection method and device |
| CN109978014A (en) * | 2019-03-06 | 2019-07-05 | 华南理工大学 | A kind of flexible base board defect inspection method merging intensive connection structure |
| CN109871903B (en) * | 2019-03-11 | 2022-08-26 | 南京邮电大学 | Target detection method based on end-to-end deep network and counterstudy |
| CN109871903A (en) * | 2019-03-11 | 2019-06-11 | 南京邮电大学 | A kind of object detection method based on end-to-end depth network and confrontation study |
| CN109886357A (en) * | 2019-03-13 | 2019-06-14 | 哈尔滨工程大学 | A kind of adaptive weighting deep learning objective classification method based on Fusion Features |
| CN109886357B (en) * | 2019-03-13 | 2022-12-13 | 哈尔滨工程大学 | Feature fusion-based adaptive weight deep learning target classification method |
| CN110008853A (en) * | 2019-03-15 | 2019-07-12 | 华南理工大学 | Pedestrian detection network and model training method, detection method, medium, equipment |
| CN111738036B (en) * | 2019-03-25 | 2023-09-29 | 北京四维图新科技股份有限公司 | Image processing method, device, equipment and storage medium |
| CN110047058A (en) * | 2019-03-25 | 2019-07-23 | 杭州电子科技大学 | A kind of image interfusion method based on residual pyramid |
| CN111738036A (en) * | 2019-03-25 | 2020-10-02 | 北京四维图新科技股份有限公司 | Image processing method, device, equipment and storage medium |
| CN110110719A (en) * | 2019-03-27 | 2019-08-09 | 浙江工业大学 | A kind of object detection method based on attention layer region convolutional neural networks |
| CN110084124A (en) * | 2019-03-28 | 2019-08-02 | 北京大学 | Feature based on feature pyramid network enhances object detection method |
| CN110008953B (en) * | 2019-03-29 | 2023-04-28 | 华南理工大学 | Potential target area generation method based on convolution neural network multi-layer feature fusion |
| CN110008953A (en) * | 2019-03-29 | 2019-07-12 | 华南理工大学 | Potential target Area generation method based on the fusion of convolutional neural networks multilayer feature |
| CN110110599B (en) * | 2019-04-03 | 2023-05-09 | 天津大学 | Remote sensing image target detection method based on multi-scale feature fusion |
| CN110110599A (en) * | 2019-04-03 | 2019-08-09 | 天津大学 | A kind of Remote Sensing Target detection method based on multi-scale feature fusion |
| CN111797846A (en) * | 2019-04-08 | 2020-10-20 | 四川大学 | Feedback type target detection method based on characteristic pyramid network |
| CN111797846B (en) * | 2019-04-08 | 2022-06-21 | 四川大学 | Feedback type target detection method based on characteristic pyramid network |
| CN110084299B (en) * | 2019-04-24 | 2021-08-06 | 中国人民解放军国防科技大学 | Target detection method and device based on multi-head fusion attention |
| CN110084299A (en) * | 2019-04-24 | 2019-08-02 | 中国人民解放军国防科技大学 | target detection method and device based on multi-head fusion attention |
| CN111914861A (en) * | 2019-05-08 | 2020-11-10 | 北京字节跳动网络技术有限公司 | Target detection method and device |
| CN111914599A (en) * | 2019-05-09 | 2020-11-10 | 四川大学 | Fine-grained bird recognition method based on semantic information multi-layer feature fusion |
| CN111914599B (en) * | 2019-05-09 | 2022-09-02 | 四川大学 | Fine-grained bird recognition method based on semantic information multi-layer feature fusion |
| CN111931799A (en) * | 2019-05-13 | 2020-11-13 | 百度在线网络技术（北京）有限公司 | Image recognition method and device |
| CN110135503B (en) * | 2019-05-19 | 2022-07-12 | 重庆理工大学 | Deep learning identification method for parts of assembly robot |
| CN110135503A (en) * | 2019-05-19 | 2019-08-16 | 重庆理工大学 | One kind put together machines people's part depth study recognition methods |
| CN110135373A (en) * | 2019-05-20 | 2019-08-16 | 北京探境科技有限公司 | A kind of multiple dimensioned face recognition method, system and electronic equipment |
| CN110287849A (en) * | 2019-06-20 | 2019-09-27 | 北京工业大学 | A kind of lightweight depth network image object detection method suitable for raspberry pie |
| CN110348447A (en) * | 2019-06-27 | 2019-10-18 | 电子科技大学 | A kind of multiple-model integration object detection method with rich space information |
| CN110348447B (en) * | 2019-06-27 | 2022-04-19 | 电子科技大学 | Multi-model integrated target detection method with abundant spatial information |
| CN110427823A (en) * | 2019-06-28 | 2019-11-08 | 北京大学 | Joint objective detection method and device based on video frame and pulse array signals |
| CN110287927B (en) * | 2019-07-01 | 2021-07-27 | 西安电子科技大学 | Remote sensing image target detection method based on depth multi-scale and context learning |
| CN110287927A (en) * | 2019-07-01 | 2019-09-27 | 西安电子科技大学 | Based on the multiple dimensioned remote sensing image object detection method with context study of depth |
| CN110309792A (en) * | 2019-07-04 | 2019-10-08 | 电子科技大学 | Indoor human detection method based on component model |
| CN110309792B (en) * | 2019-07-04 | 2022-07-01 | 电子科技大学 | Indoor person detection method based on component template |
| CN110427970B (en) * | 2019-07-05 | 2023-08-01 | 平安科技（深圳）有限公司 | Image classification method, apparatus, computer device and storage medium |
| CN110427970A (en) * | 2019-07-05 | 2019-11-08 | 平安科技（深圳）有限公司 | Image classification method, device, computer equipment and storage medium |
| CN110400305A (en) * | 2019-07-26 | 2019-11-01 | 哈尔滨理工大学 | A kind of object detection method based on deep learning |
| CN110415795A (en) * | 2019-08-02 | 2019-11-05 | 杭州智团信息技术有限公司 | A kind of recognition methods of fluorescent staining CTC image |
| CN112307853A (en) * | 2019-08-02 | 2021-02-02 | 成都天府新区光启未来技术研究院 | Detection method of aerial image, storage medium and electronic device |
| CN112348042A (en) * | 2019-08-08 | 2021-02-09 | 中国人民解放军火箭军工程大学 | Improved YOLOv 3-based infrared target detection method |
| CN110533084B (en) * | 2019-08-12 | 2022-09-30 | 长安大学 | Multi-scale target detection method based on self-attention mechanism |
| CN110533084A (en) * | 2019-08-12 | 2019-12-03 | 长安大学 | A kind of multiscale target detection method based on from attention mechanism |
| CN110647816B (en) * | 2019-08-26 | 2022-11-22 | 合肥工业大学 | Target detection method for real-time monitoring of goods shelf medicines |
| CN110647816A (en) * | 2019-08-26 | 2020-01-03 | 合肥工业大学 | Target detection method for real-time monitoring of goods shelf medicines |
| CN110503112A (en) * | 2019-08-27 | 2019-11-26 | 电子科技大学 | A kind of small target deteection of Enhanced feature study and recognition methods |
| CN110503112B (en) * | 2019-08-27 | 2023-02-03 | 电子科技大学 | Small target detection and identification method for enhancing feature learning |
| CN110533105B (en) * | 2019-08-30 | 2022-04-05 | 北京市商汤科技开发有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN110533105A (en) * | 2019-08-30 | 2019-12-03 | 北京市商汤科技开发有限公司 | A kind of object detection method and device, electronic equipment and storage medium |
| CN110598788A (en) * | 2019-09-12 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN110598788B (en) * | 2019-09-12 | 2023-06-30 | 腾讯科技（深圳）有限公司 | Target detection method, target detection device, electronic equipment and storage medium |
| CN110728658A (en) * | 2019-09-16 | 2020-01-24 | 武汉大学 | High-resolution remote sensing image weak target detection method based on deep learning |
| CN110852317A (en) * | 2019-09-16 | 2020-02-28 | 浙江工业大学 | Small-scale target detection method based on weak edge |
| CN110728200A (en) * | 2019-09-23 | 2020-01-24 | 武汉大学 | Real-time pedestrian detection method and system based on deep learning |
| CN110796640A (en) * | 2019-09-29 | 2020-02-14 | 郑州金惠计算机系统工程有限公司 | Small target defect detection method and device, electronic equipment and storage medium |
| CN110766721B (en) * | 2019-09-30 | 2023-06-06 | 南京航空航天大学 | Carrier landing cooperative target detection method based on airborne vision |
| CN110766721A (en) * | 2019-09-30 | 2020-02-07 | 南京航空航天大学 | Carrier landing cooperative target detection method based on airborne vision |
| CN110738647A (en) * | 2019-10-12 | 2020-01-31 | 成都考拉悠然科技有限公司 | Mouse detection method integrating multi-receptive-field feature mapping and Gaussian probability model |
| CN110807372A (en) * | 2019-10-15 | 2020-02-18 | 哈尔滨工程大学 | Rapid optical remote sensing target identification method based on depth feature recombination |
| CN110689021A (en) * | 2019-10-17 | 2020-01-14 | 哈尔滨理工大学 | Real-time target detection method in low-visibility environment based on deep learning |
| CN110807384A (en) * | 2019-10-24 | 2020-02-18 | 华东计算技术研究所（中国电子科技集团公司第三十二研究所） | Small target detection method and system under low visibility |
| CN111008562A (en) * | 2019-10-31 | 2020-04-14 | 北京城建设计发展集团股份有限公司 | Human-vehicle target detection method with feature map depth fusion |
| WO2021083126A1 (en) * | 2019-10-31 | 2021-05-06 | 北京市商汤科技开发有限公司 | Target detection and intelligent driving methods and apparatuses, device, and storage medium |
| JP2022535473A (en) * | 2019-10-31 | 2022-08-09 | ベイジン センスタイム テクノロジー デベロップメント シーオー．，エルティーディー | Target detection, intelligent driving methods, devices, equipment and storage media |
| CN111008562B (en) * | 2019-10-31 | 2023-04-18 | 北京城建设计发展集团股份有限公司 | Human-vehicle target detection method with feature map depth fusion |
| CN110852241A (en) * | 2019-11-06 | 2020-02-28 | 西安交通大学 | Small target detection method applied to nursing robot |
| CN110852241B (en) * | 2019-11-06 | 2022-08-16 | 西安交通大学 | Small target detection method applied to nursing robot |
| CN110796115A (en) * | 2019-11-08 | 2020-02-14 | 厦门美图之家科技有限公司 | Image detection method and device, electronic equipment and readable storage medium |
| CN110796115B (en) * | 2019-11-08 | 2022-12-23 | 厦门美图宜肤科技有限公司 | Image detection method and device, electronic equipment and readable storage medium |
| CN110826514A (en) * | 2019-11-13 | 2020-02-21 | 国网青海省电力公司海东供电公司 | Construction site violation intelligent identification method based on deep learning |
| CN110956108A (en) * | 2019-11-22 | 2020-04-03 | 华南理工大学 | Small frequency scale detection method based on characteristic pyramid |
| CN110956108B (en) * | 2019-11-22 | 2023-04-18 | 华南理工大学 | Small frequency scale detection method based on characteristic pyramid |
| CN111191508A (en) * | 2019-11-28 | 2020-05-22 | 浙江省北大信息技术高等研究院 | Face recognition method and device |
| CN111079620B (en) * | 2019-12-10 | 2023-10-17 | 北京小蝇科技有限责任公司 | White blood cell image detection and identification model construction method and application based on transfer learning |
| CN111079620A (en) * | 2019-12-10 | 2020-04-28 | 北京小蝇科技有限责任公司 | Leukocyte image detection and identification model construction method based on transfer learning and application |
| CN111191535A (en) * | 2019-12-18 | 2020-05-22 | 南京理工大学 | Pedestrian detection model construction method based on deep learning and pedestrian detection method |
| CN111191535B (en) * | 2019-12-18 | 2022-08-09 | 南京理工大学 | Pedestrian detection model construction method based on deep learning and pedestrian detection method |
| CN111178213B (en) * | 2019-12-23 | 2022-11-18 | 大连理工大学 | Aerial photography vehicle detection method based on deep learning |
| CN111178213A (en) * | 2019-12-23 | 2020-05-19 | 大连理工大学 | Aerial photography vehicle detection method based on deep learning |
| CN111144494A (en) * | 2019-12-27 | 2020-05-12 | 睿魔智能科技（深圳）有限公司 | Object detection model training method, object detection device, object detection equipment and object detection medium |
| CN113159079A (en) * | 2020-01-07 | 2021-07-23 | 顺丰科技有限公司 | Target detection method, target detection device, computer equipment and storage medium |
| CN111259758A (en) * | 2020-01-13 | 2020-06-09 | 中国矿业大学 | Two-stage remote sensing image target detection method for dense area |
| CN111259758B (en) * | 2020-01-13 | 2023-04-07 | 中国矿业大学 | Two-stage remote sensing image target detection method for dense area |
| CN111209887A (en) * | 2020-01-15 | 2020-05-29 | 西安电子科技大学 | SSD model optimization method for small target detection |
| CN111274970A (en) * | 2020-01-21 | 2020-06-12 | 南京航空航天大学 | Traffic sign detection method based on improved YOLO v3 algorithm |
| CN111539782B (en) * | 2020-01-21 | 2024-04-30 | 中国银联股份有限公司 | Deep learning-based merchant information data processing method and system |
| CN111539782A (en) * | 2020-01-21 | 2020-08-14 | 中国银联股份有限公司 | Merchant information data processing method and system based on deep learning |
| CN111291687B (en) * | 2020-02-11 | 2022-11-11 | 青岛联合创智科技有限公司 | 3D human body action standard identification method |
| CN111291687A (en) * | 2020-02-11 | 2020-06-16 | 青岛联合创智科技有限公司 | 3D human body action standard identification method |
| CN111340039A (en) * | 2020-02-12 | 2020-06-26 | 杰创智能科技股份有限公司 | Target detection method based on feature selection |
| CN111340039B (en) * | 2020-02-12 | 2023-10-17 | 杰创智能科技股份有限公司 | Target detection method based on feature selection |
| CN111444939B (en) * | 2020-02-19 | 2022-06-28 | 山东大学 | Small-scale equipment component detection method based on weak supervision cooperative learning in open scene of power field |
| CN111444939A (en) * | 2020-02-19 | 2020-07-24 | 山东大学 | Small-scale equipment component detection method based on weak supervision cooperative learning in open scene of power field |
| CN111340126A (en) * | 2020-03-03 | 2020-06-26 | 腾讯云计算（北京）有限责任公司 | Article identification method and device, computer equipment and storage medium |
| CN111612797B (en) * | 2020-03-03 | 2021-05-25 | 江苏大学 | Rice image information processing system |
| CN111612797A (en) * | 2020-03-03 | 2020-09-01 | 江苏大学 | Rice image information processing system |
| CN111382787A (en) * | 2020-03-06 | 2020-07-07 | 芯薇(上海)智能科技有限公司 | Target detection method based on deep learning |
| CN111292377B (en) * | 2020-03-11 | 2024-01-23 | 南京旷云科技有限公司 | Target detection method, device, computer equipment and storage medium |
| CN111292377A (en) * | 2020-03-11 | 2020-06-16 | 南京旷云科技有限公司 | Target detection method, target detection device, computer equipment and storage medium |
| CN113408321B (en) * | 2020-03-16 | 2023-08-22 | 中国人民解放军战略支援部队信息工程大学 | Real-time target detection method and device for lightweight image and video data |
| CN113408321A (en) * | 2020-03-16 | 2021-09-17 | 中国人民解放军战略支援部队信息工程大学 | Real-time target detection method and device for lightweight image and video data |
| CN111401290A (en) * | 2020-03-24 | 2020-07-10 | 杭州博雅鸿图视频技术有限公司 | Face detection method and system and computer readable storage medium |
| CN111444865A (en) * | 2020-03-31 | 2020-07-24 | 盐城禅图智能科技有限公司 | Multi-scale target detection method based on gradual refinement |
| CN111461217B (en) * | 2020-03-31 | 2023-05-23 | 华南理工大学 | Aerial image small target detection method based on feature fusion and up-sampling |
| CN111461217A (en) * | 2020-03-31 | 2020-07-28 | 华南理工大学 | Aerial image small target detection method based on feature fusion and up-sampling |
| CN111462090B (en) * | 2020-04-01 | 2023-09-01 | 上饶市中科院云计算中心大数据研究院 | Multi-scale image target detection method |
| CN111462090A (en) * | 2020-04-01 | 2020-07-28 | 上饶市中科院云计算中心大数据研究院 | Multi-scale image target detection method |
| CN111461039A (en) * | 2020-04-07 | 2020-07-28 | 电子科技大学 | Landmark identification method based on multi-scale feature fusion |
| CN111461039B (en) * | 2020-04-07 | 2022-03-25 | 电子科技大学 | Landmark identification method based on multi-scale feature fusion |
| CN111582093A (en) * | 2020-04-27 | 2020-08-25 | 北京工业大学 | Automatic small target detection method in high-resolution image based on computer vision and deep learning |
| CN111553280A (en) * | 2020-04-28 | 2020-08-18 | 上海无线电设备研究所 | Target part identification method based on deep learning |
| CN111783685A (en) * | 2020-05-08 | 2020-10-16 | 西安建筑科技大学 | Target detection improved algorithm based on single-stage network model |
| CN111652288B (en) * | 2020-05-11 | 2023-08-01 | 北京航天自动控制研究所 | Improved SSD small target detection method based on dense feature pyramid |
| CN111652288A (en) * | 2020-05-11 | 2020-09-11 | 北京航天自动控制研究所 | Improved SSD small target detection method based on dense feature pyramid |
| CN111666822A (en) * | 2020-05-13 | 2020-09-15 | 飒铂智能科技有限责任公司 | Low-altitude unmanned aerial vehicle target detection method and system based on deep learning |
| CN111626208A (en) * | 2020-05-27 | 2020-09-04 | 北京百度网讯科技有限公司 | Method and apparatus for detecting small targets |
| CN111767944B (en) * | 2020-05-27 | 2023-08-15 | 重庆大学 | Single-stage detector design method suitable for multi-scale target detection based on deep learning |
| CN111767944A (en) * | 2020-05-27 | 2020-10-13 | 重庆大学 | Deep learning-based single-stage detector design method suitable for multi-scale target detection |
| CN111461085A (en) * | 2020-06-01 | 2020-07-28 | 青岛大学 | Log detection method based on weight sharing and upper and lower feature fusion |
| CN111461085B (en) * | 2020-06-01 | 2022-05-31 | 青岛大学 | Log detection method based on weight sharing and upper and lower feature fusion |
| CN111860587A (en) * | 2020-06-12 | 2020-10-30 | 长安大学 | Method for detecting small target of picture |
| CN111860587B (en) * | 2020-06-12 | 2024-02-02 | 长安大学 | Detection method for small targets of pictures |
| CN111723737B (en) * | 2020-06-19 | 2023-11-17 | 河南科技大学 | Target detection method based on multi-scale matching strategy deep feature learning |
| CN111723737A (en) * | 2020-06-19 | 2020-09-29 | 河南科技大学 | Target detection method based on multi-scale matching strategy deep feature learning |
| CN111767878A (en) * | 2020-07-03 | 2020-10-13 | 中国科学院自动化研究所 | Deep learning-based traffic sign detection method and system in embedded device |
| CN111985325A (en) * | 2020-07-15 | 2020-11-24 | 国网河南省电力公司 | Aerial small target rapid identification method in extra-high voltage environment evaluation |
| CN111985325B (en) * | 2020-07-15 | 2023-04-07 | 国网河南省电力公司 | Aerial small target rapid identification method in extra-high voltage environment evaluation |
| CN111860637A (en) * | 2020-07-17 | 2020-10-30 | 河南科技大学 | Single-shot multi-frame infrared target detection method |
| CN111860637B (en) * | 2020-07-17 | 2023-11-21 | 河南科技大学 | Single-shot multi-frame infrared target detection method |
| CN112084838A (en) * | 2020-07-20 | 2020-12-15 | 河海大学 | Workshop safety helmet detection method |
| CN111738212B (en) * | 2020-07-20 | 2020-11-20 | 平安国际智慧城市科技股份有限公司 | Traffic signal lamp identification method, device, equipment and medium based on artificial intelligence |
| CN112084838B (en) * | 2020-07-20 | 2022-07-29 | 河海大学 | Workshop safety helmet detection method |
| CN111738212A (en) * | 2020-07-20 | 2020-10-02 | 平安国际智慧城市科技股份有限公司 | Traffic signal lamp identification method, device, equipment and medium based on artificial intelligence |
| CN111898668A (en) * | 2020-07-24 | 2020-11-06 | 佛山市南海区广工大数控装备协同创新研究院 | Small target object detection method based on deep learning |
| CN112001403A (en) * | 2020-08-11 | 2020-11-27 | 北京化工大学 | Image contour detection method and system |
| CN112001403B (en) * | 2020-08-11 | 2023-12-15 | 北京化工大学 | Image contour detection method and system |
| CN112068111A (en) * | 2020-08-13 | 2020-12-11 | 中国人民解放军海军工程大学 | Unmanned aerial vehicle target detection method based on multi-sensor information fusion |
| CN111950515A (en) * | 2020-08-26 | 2020-11-17 | 重庆邮电大学 | Semantic feature pyramid network-based small face detection method |
| CN112183203A (en) * | 2020-08-26 | 2021-01-05 | 北京工业大学 | Real-time traffic sign detection method based on multi-scale pixel feature fusion |
| CN112183203B (en) * | 2020-08-26 | 2024-05-28 | 北京工业大学 | Real-time traffic sign detection method based on multi-scale pixel feature fusion |
| CN111950515B (en) * | 2020-08-26 | 2022-10-18 | 重庆邮电大学 | Semantic feature pyramid network-based small face detection method |
| CN112183579B (en) * | 2020-09-01 | 2023-05-30 | 国网宁夏电力有限公司检修公司 | Method, medium and system for detecting micro target |
| CN112183579A (en) * | 2020-09-01 | 2021-01-05 | 国网宁夏电力有限公司检修公司 | Method, medium and system for detecting micro target |
| CN112016512A (en) * | 2020-09-08 | 2020-12-01 | 重庆市地理信息和遥感应用中心 | Remote sensing image small target detection method based on feedback type multi-scale training |
| CN112149533A (en) * | 2020-09-10 | 2020-12-29 | 上海电力大学 | Target detection method based on improved SSD model |
| CN112070043A (en) * | 2020-09-15 | 2020-12-11 | 常熟理工学院 | Safety helmet wearing convolutional network based on feature fusion, training and detecting method |
| CN112070043B (en) * | 2020-09-15 | 2023-11-10 | 常熟理工学院 | Feature fusion-based safety helmet wearing convolution network, training and detection method |
| CN112131996A (en) * | 2020-09-17 | 2020-12-25 | 东南大学 | Road side image multi-scale pedestrian rapid detection method based on channel separation convolution |
| CN112307889B (en) * | 2020-09-22 | 2022-07-26 | 北京航空航天大学 | Face detection algorithm based on small auxiliary network |
| CN112307889A (en) * | 2020-09-22 | 2021-02-02 | 北京航空航天大学 | Face detection algorithm based on small auxiliary network |
| CN112329808A (en) * | 2020-09-25 | 2021-02-05 | 武汉光谷信息技术股份有限公司 | Optimization method and system of Deeplab semantic segmentation algorithm |
| CN112132828A (en) * | 2020-10-15 | 2020-12-25 | 浙江工商大学 | Film flaw detection method based on deep learning |
| CN112270279A (en) * | 2020-11-02 | 2021-01-26 | 重庆邮电大学 | Multi-dimensional-based remote sensing image micro-target detection method |
| CN112270366A (en) * | 2020-11-02 | 2021-01-26 | 重庆邮电大学 | Micro target detection method based on self-adaptive multi-feature fusion |
| CN112270279B (en) * | 2020-11-02 | 2022-04-12 | 重庆邮电大学 | Multi-dimensional-based remote sensing image micro-target detection method |
| CN112184711A (en) * | 2020-11-05 | 2021-01-05 | 杭州青枭科技有限公司 | Photovoltaic module defect detection and positioning method and system |
| CN112184711B (en) * | 2020-11-05 | 2024-04-02 | 杭州青枭科技有限公司 | Photovoltaic module defect detection and positioning method and system |
| CN112308040A (en) * | 2020-11-26 | 2021-02-02 | 山东捷讯通信技术有限公司 | River sewage outlet detection method and system based on high-definition images |
| CN112348036A (en) * | 2020-11-26 | 2021-02-09 | 北京工业大学 | Self-adaptive target detection method based on lightweight residual learning and deconvolution cascade |
| CN112287895A (en) * | 2020-11-26 | 2021-01-29 | 山东捷讯通信技术有限公司 | Model construction method, recognition method and system for river drain outlet detection |
| CN112418165A (en) * | 2020-12-07 | 2021-02-26 | 武汉工程大学 | Small-size target detection method and device based on improved cascade neural network |
| CN112633352A (en) * | 2020-12-18 | 2021-04-09 | 浙江大华技术股份有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN112633352B (en) * | 2020-12-18 | 2023-08-29 | 浙江大华技术股份有限公司 | Target detection method and device, electronic equipment and storage medium |
| CN112508935A (en) * | 2020-12-22 | 2021-03-16 | 郑州金惠计算机系统工程有限公司 | Product packaging detection method and system based on deep learning and product packaging sorting system |
| CN112651441A (en) * | 2020-12-25 | 2021-04-13 | 深圳市信义科技有限公司 | Fine-grained non-motor vehicle feature detection method, storage medium and computer equipment |
| CN112651441B (en) * | 2020-12-25 | 2022-08-16 | 深圳市信义科技有限公司 | Fine-grained non-motor vehicle feature detection method, storage medium and computer equipment |
| CN112841154A (en) * | 2020-12-29 | 2021-05-28 | 长沙湘丰智能装备股份有限公司 | Disease and pest control system based on artificial intelligence |
| CN112766361A (en) * | 2021-01-18 | 2021-05-07 | 山东师范大学 | Target fruit detection method and detection system under homochromatic background |
| CN113052210B (en) * | 2021-03-11 | 2024-04-26 | 北京工业大学 | Rapid low-light target detection method based on convolutional neural network |
| CN113052210A (en) * | 2021-03-11 | 2021-06-29 | 北京工业大学 | Fast low-illumination target detection method based on convolutional neural network |
| CN112966659B (en) * | 2021-03-31 | 2022-08-23 | 北湾科技(武汉)有限公司 | Video image small target detection method based on deep learning |
| CN112966659A (en) * | 2021-03-31 | 2021-06-15 | 北湾科技(武汉)有限公司 | Video image small target detection method based on deep learning |
| CN113112866A (en) * | 2021-04-14 | 2021-07-13 | 深圳市旗扬特种装备技术工程有限公司 | Intelligent traffic early warning method and intelligent traffic early warning system |
| CN113112866B (en) * | 2021-04-14 | 2022-06-03 | 深圳市旗扬特种装备技术工程有限公司 | Intelligent traffic early warning method and intelligent traffic early warning system |
| CN113095316A (en) * | 2021-04-15 | 2021-07-09 | 西安电子科技大学 | Image rotation target detection method based on multilevel fusion and angular point offset |
| CN113095316B (en) * | 2021-04-15 | 2023-04-07 | 西安电子科技大学 | Image rotation target detection method based on multilevel fusion and angular point offset |
| JP2023527615A (en) * | 2021-04-28 | 2023-06-30 | ベイジン バイドゥ ネットコム サイエンス テクノロジー カンパニー リミテッド | Target object detection model training method, target object detection method, device, electronic device, storage medium and computer program |
| WO2022227770A1 (en) * | 2021-04-28 | 2022-11-03 | 北京百度网讯科技有限公司 | Method for training target object detection model, target object detection method, and device |
| CN113283499A (en) * | 2021-05-24 | 2021-08-20 | 南京航空航天大学 | Three-dimensional fabric weaving density detection method based on deep learning |
| CN113420660A (en) * | 2021-06-23 | 2021-09-21 | 西安电子科技大学 | Infrared image target detection model construction method, prediction method and system |
| CN113420819A (en) * | 2021-06-25 | 2021-09-21 | 西北工业大学 | Lightweight underwater target detection method based on CenterNet |
| CN113591568A (en) * | 2021-06-28 | 2021-11-02 | 北京百度网讯科技有限公司 | Target detection method, training method of target detection model and device thereof |
| CN113486781B (en) * | 2021-07-02 | 2023-10-24 | 国网电力科学研究院有限公司 | Electric power inspection method and device based on deep learning model |
| CN113486781A (en) * | 2021-07-02 | 2021-10-08 | 国网电力科学研究院有限公司 | Electric power inspection method and device based on deep learning model |
| CN113705341A (en) * | 2021-07-16 | 2021-11-26 | 国家石油天然气管网集团有限公司 | Small-scale face detection method based on generation countermeasure network |
| CN113723181B (en) * | 2021-07-20 | 2023-10-20 | 深圳大学 | Unmanned aerial vehicle aerial photographing target detection method and device |
| CN113723181A (en) * | 2021-07-20 | 2021-11-30 | 深圳大学 | Unmanned aerial vehicle aerial photography target detection method and device |
| CN113657196A (en) * | 2021-07-27 | 2021-11-16 | 中国科学院自动化研究所 | SAR image target detection method and device, electronic equipment and storage medium |
| CN113657196B (en) * | 2021-07-27 | 2024-05-28 | 中国科学院自动化研究所 | SAR image target detection method, SAR image target detection device, electronic equipment and storage medium |
| CN113705359A (en) * | 2021-08-03 | 2021-11-26 | 江南大学 | Multi-scale clothes detection system and method based on washing machine drum image |
| CN113705359B (en) * | 2021-08-03 | 2024-05-03 | 江南大学 | Multi-scale clothes detection system and method based on drum images of washing machine |
| CN113657285A (en) * | 2021-08-18 | 2021-11-16 | 中国人民解放军陆军装甲兵学院 | Real-time target detection method based on small-scale target |
| CN113792631A (en) * | 2021-08-31 | 2021-12-14 | 电子科技大学 | Aircraft detection and tracking method based on multi-scale self-adaption and side-domain attention |
| CN114373121A (en) * | 2021-09-08 | 2022-04-19 | 武汉众智数字技术有限公司 | Method and system for improving small target detection of yolov5 network |
| CN113963265A (en) * | 2021-09-13 | 2022-01-21 | 北京理工雷科电子信息技术有限公司 | Small sample small target rapid detection and identification method for complex remote sensing land environment |
| CN113807231A (en) * | 2021-09-14 | 2021-12-17 | 西安电子科技大学 | X-ray contraband detection method based on UNET (UNET) downsampling convolutional neural network |
| CN113807231B (en) * | 2021-09-14 | 2024-02-13 | 西安电子科技大学 | X-ray contraband detection method based on UNET downsampling convolutional neural network |
| CN113792660B (en) * | 2021-09-15 | 2024-03-01 | 江苏科技大学 | Pedestrian detection method, system, medium and equipment based on improved YOLOv3 network |
| CN113792660A (en) * | 2021-09-15 | 2021-12-14 | 江苏科技大学 | Pedestrian detection method, system, medium and equipment based on improved YOLOv3 network |
| CN113887649A (en) * | 2021-10-19 | 2022-01-04 | 齐鲁工业大学 | Target detection method based on fusion of deep-layer features and shallow-layer features |
| CN113887649B (en) * | 2021-10-19 | 2022-05-27 | 齐鲁工业大学 | Target detection method based on fusion of deep layer features and shallow layer features |
| CN114241274A (en) * | 2021-11-30 | 2022-03-25 | 电子科技大学 | Small target detection method based on super-resolution multi-scale feature fusion |
| CN114241274B (en) * | 2021-11-30 | 2023-04-07 | 电子科技大学 | Small target detection method based on super-resolution multi-scale feature fusion |
| CN114359680A (en) * | 2021-12-17 | 2022-04-15 | 中国人民解放军海军工程大学 | Panoramic vision water surface target detection method based on deep learning |
| CN114863267A (en) * | 2022-03-30 | 2022-08-05 | 南京邮电大学 | Aerial tree number accurate statistical method based on multi-track intelligent prediction |
| CN115205855B (en) * | 2022-07-08 | 2023-10-27 | 中国人民解放军国防科技大学 | Vehicle target identification method, device and equipment integrating multi-scale semantic information |
| CN115205855A (en) * | 2022-07-08 | 2022-10-18 | 中国人民解放军国防科技大学 | Vehicle target identification method, device and equipment fusing multi-scale semantic information |
| CN115272700A (en) * | 2022-08-03 | 2022-11-01 | 河南大学 | Improved SSD small target detection method based on multi-scale feature self-enhancement |
| CN115272700B (en) * | 2022-08-03 | 2023-06-02 | 河南大学 | Improved SSD small target detection method based on multi-scale feature self-enhancement |
| CN115841460A (en) * | 2022-11-21 | 2023-03-24 | 国网湖北省电力有限公司超高压公司 | High-precision hardware crack image detection and feature extraction method under complex background |
| CN115984846A (en) * | 2023-02-06 | 2023-04-18 | 山东省人工智能研究院 | Intelligent identification method for small target in high-resolution image based on deep learning |
| CN115984846B (en) * | 2023-02-06 | 2023-10-10 | 山东省人工智能研究院 | Intelligent recognition method for small targets in high-resolution image based on deep learning |
| CN116070789A (en) * | 2023-03-17 | 2023-05-05 | 北京茗禾科技有限公司 | Artificial intelligence-based single-yield prediction method for mature-period rice and wheat |
| CN116070789B (en) * | 2023-03-17 | 2023-06-02 | 北京茗禾科技有限公司 | Artificial intelligence-based single-yield prediction method for mature-period rice and wheat |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109344821A (en) | Small target detecting method based on Fusion Features and deep learning | |
| CN108334847B (en) | A kind of face identification method based on deep learning under real scene | |
| Yao et al. | Application of convolutional neural network in classification of high resolution agricultural remote sensing images | |
| CN101714262B (en) | Method for reconstructing three-dimensional scene of single image | |
| CN113850825A (en) | Remote sensing image road segmentation method based on context information and multi-scale feature fusion | |
| CN106683091A (en) | Target classification and attitude detection method based on depth convolution neural network | |
| CN110163836A (en) | Based on deep learning for the excavator detection method under the inspection of high-altitude | |
| CN105303195B (en) | A kind of bag of words image classification method | |
| CN113160062B (en) | Infrared image target detection method, device, equipment and storage medium | |
| CN111553406A (en) | Target detection system, method and terminal based on improved YOLO-V3 | |
| CN109785344A (en) | The remote sensing image segmentation method of binary channel residual error network based on feature recalibration | |
| CN109784283A (en) | Based on the Remote Sensing Target extracting method under scene Recognition task | |
| CN110378239A (en) | A kind of real-time traffic marker detection method based on deep learning | |
| CN105243154A (en) | Remote sensing image retrieval method and system based on significant point characteristics and spare self-encodings | |
| CN109948457B (en) | Real-time target recognition method based on convolutional neural network and CUDA acceleration | |
| CN107944459A (en) | A kind of RGB D object identification methods | |
| CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
| CN109063549A (en) | High-resolution based on deep neural network is taken photo by plane video moving object detection method | |
| CN114842208A (en) | Power grid harmful bird species target detection method based on deep learning | |
| CN107908646A (en) | A kind of image search method based on layering convolutional neural networks | |
| CN107506792A (en) | A kind of semi-supervised notable method for checking object | |
| CN110322509A (en) | Object localization method, system and computer equipment based on level Class Activation figure | |
| CN115223017B (en) | Multi-scale feature fusion bridge detection method based on depth separable convolution | |
| Bragilevsky et al. | Deep learning for Amazon satellite image analysis | |
| CN108961385A (en) | A kind of SLAM patterning process and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190215 |