CN108537286B - Complex target accurate identification method based on key area detection - Google Patents

Complex target accurate identification method based on key area detection Download PDF

Info

Publication number
CN108537286B
CN108537286B CN201810345899.9A CN201810345899A CN108537286B CN 108537286 B CN108537286 B CN 108537286B CN 201810345899 A CN201810345899 A CN 201810345899A CN 108537286 B CN108537286 B CN 108537286B
Authority
CN
China
Prior art keywords
network
complex target
key area
area
sub
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.)
Active
Application number
CN201810345899.9A
Other languages
Chinese (zh)
Other versions
CN108537286A (en
Inventor
王田
李玮匡
李嘉锟
陶飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beihang University
Original Assignee
Beihang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beihang University filed Critical Beihang University
Priority to CN201810345899.9A priority Critical patent/CN108537286B/en
Publication of CN108537286A publication Critical patent/CN108537286A/en
Application granted granted Critical
Publication of CN108537286B publication Critical patent/CN108537286B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
    • 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
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a complex target accurate identification method based on key area detection, which comprises the following steps: the method comprises the steps of performing fusion training on the whole neural network by using a cross training method, extracting target features by using a convolutional neural network, detecting key regions of a complex target by using a detection sub-network and taking an anchor square frame as reference, pooling the key regions into feature maps with fixed sizes by using regional standard pooling, classifying the key regions by using a classification sub-network, and fusing classification results of all key regions so as to achieve accurate identification of the target. The whole network comprises a key area detection sub-network and a key area classification sub-network, the key area with distinguishing degree of the complex target is detected by the detection sub-network, then the key area is classified by the classification sub-network, and the classification result of each area is fused to identify the whole target. The two sub-networks share the features extracted by the VGG convolutional neural network, so that the complex target is identified quickly and accurately.

Description

Complex target accurate identification method based on key area detection
Technical Field
The invention relates to an image processing technology, in particular to a complex target accurate identification method based on key area detection.
Background
The classification and identification of complex objects is an important and fundamental task in the field of computer vision. Most parts of different kinds of complex targets are often the same or similar, and the differences are often reflected in some local key areas, so that a large amount of interference and redundant information exist in the images of the complex targets. The existing classification and identification methods for complex targets have the problem of low accuracy because the interference and redundant information in the complex target images cannot be removed. In order to realize accurate classification and identification of complex targets, the research of a complex target accurate identification method based on key region detection is of great significance.
Disclosure of Invention
In view of this, the present invention provides a method for accurately identifying a complex target in a high-accuracy basic key region detection, which greatly improves the detection accuracy and ensures the rapidity of identification.
In order to achieve the purpose, the technical scheme provided by the invention is as follows: a complex target accurate identification method based on key area detection is realized by the following steps:
step 1, reading a complex target picture, a coordinate label of a key area of a complex target and a complex target classification label in a training sample in a database, and performing fusion training on a complex target accurate identification network by using a cross training method.
And 2, taking the complex target picture to be recognized as the input of the complex target accurate recognition network trained in the step 1, and extracting features through a VGG convolutional neural network to obtain a feature map of the complex target picture to be recognized.
Step 3, inputting the feature map obtained in the step 2 into a key area detection sub-network, sliding the feature map by using the sub-network with the size of 3 multiplied by 3, detecting the key area of the complex target picture by using the anchor frame as reference, and giving out the keyPrediction blocks of regions and probability P of being a critical regionis,Pnot；
And 4, filtering the detected regions with higher overlapping degree by adopting non-maximum inhibition, and only keeping the possibility P of being the key region when the ratio of the intersection part area and the union part area of different prediction blocks exceeds a specified threshold IOU _ thresholdisThe largest prediction block, while the other blocks are filtered;
step 5, setting the possibility P of the key areaisWill be the critical area probability PisMapping the area larger than the set threshold value P _ threshold to a feature map extracted by the VGG network;
step 6, performing area standard pooling on the areas mapped on the feature maps obtained in the step 5, and pooling the detected areas with different sizes into feature maps with fixed sizes;
step 7, taking the feature map with fixed size obtained in the step 6 as the input of a classification sub-network, accurately classifying the feature map by using the classification sub-network, and normalizing the classification result by using a softmax function to obtain the probability of classifying the key region;
and 8, taking the mean value of the classified corresponding probabilities of the key regions obtained in the step 7 for the same complex target corresponding to the same picture, and fusing to obtain an accurate identification result of the complex target type.
In step 1, the whole network cross training process is as follows:
step 11, fine tuning is carried out on the basis of taking ImageNet database pictures as training samples and taking weights of VGG networks trained aiming at classification tasks as initial weights;
step 12, reading the complex target picture and the coordinate label of the key area corresponding to the complex target picture, and training the sub-network for detecting the key area, wherein the loss function of the training is L ═ LP+LregWherein L isPProbability P of detecting whether sub-network output is a key area for the key areais,PnotCross entropy with the true value of the tag, LregIs a critical areaThe square sum of the coordinate offset of the detection area output by the domain detection subnetwork and the coordinate offset of the actual key area in the label;
step 13, reading the complex target picture and the classification label corresponding to the complex target picture, training a classification sub-network, wherein the training loss function is the cross entropy between the network output classification result and the actual label result;
step 14 repeats steps 12 and 13 several times, cross-training the key area detection sub-network and the classification sub-network until the network is stable.
In step 3, the method for detecting the key area is as follows:
step 31, using a sliding window with the size of 3 × 3, sliding on the feature map obtained in step 2, and obtaining a 512-dimensional vector at each position;
step 32, setting 9 anchor boxes as reference at the position of each sliding window, setting the length-width ratio of the anchor boxes to be three proportions according to 1:2, 1:1 and 2:1, and setting the area size to be 1282、2562、5122The pixel has three sizes, and the center point of the anchor frame is the center of the sliding window;
step 33, outputting 9 vectors of 6 dimensions from the 512-dimensional vectors obtained at each sliding window position through a full-connection network; each vector represents the offset d of the coordinates of the center point, the length and the width of the detection area with respect to a reference anchor blockx,dy,dl,dwAnd whether it is a critical area probability Pis,PnotWherein: dx＝(x-xa)/la，dy＝(y-ya)/wa， dl＝log(l/la)，dw＝log(w/wa) X, y, l, w represent coordinates of the center point of the detected region, length and width, xa,ya,la,waRepresenting the coordinates, length and width, P, of the center point of the reference anchor regionis,PnotPerforming normalization processing by using a softmax function;
step 34 offset d from the net regressionx,dy,dl,dwCoordinate with center point of anchor box, length and width xa,ya,la,waAnd calculating the actual central point coordinates, length and width x, y, l and w of the detection area.
In step 6, the method for standard pooling of the regions is as follows:
step 61, representing the size of the region to be pooled as m × n, dividing the region to be pooled into 7 × 7 small lattices with the size of about m/7 × n/7, and rounding up approximately according to rounding up when m/7 or n/7 cannot be rounded up;
in step 62, in each small cell divided in step 61, features in the small cells are pooled into 1 × 1 dimension by using the maximum pooling method, and thus, feature regions of different sizes are pooled into a 7 × 7 dimension fixed-size feature map.
In summary, the method for accurately identifying a complex target based on key area detection according to the present invention includes: the method comprises the steps of performing fusion training on the whole neural network by using a cross training method, extracting target features by using a convolutional neural network, detecting key regions of a complex target by using a detection sub-network and taking an anchor square frame as reference, pooling the key regions into feature maps with fixed sizes by using regional standard pooling, classifying the key regions by using a classification sub-network, and fusing classification results of all key regions so as to achieve accurate identification of the target. The whole network comprises a key area detection sub-network and a key area classification sub-network, the key area with distinguishing degree of the complex target is detected by the detection sub-network, then the key area is classified by the classification sub-network, and the classification result of each area is fused to identify the whole target. The two sub-networks share the features extracted by the VGG convolutional neural network, so that the complex target is identified quickly and accurately.
Compared with the prior art, the invention has the advantages that:
(1) precision: many different complex objects tend to be similar in most places, while their differences tend to be in locally critical areas. The traditional target identification method takes the whole picture as the input of a classification network, and the whole picture contains a large amount of redundant information and interference information, which limits the accuracy of target identification. The method uses the detection sub-network to detect the key area firstly, then uses the classification sub-network to identify the key area, and fuses the identification results of all the key areas to achieve the effect of accurate target identification.
(2) Rapidity: the invention adopts a deep neural network to extract the characteristics of an original image, and a detection sub-network and a classification sub-network share the characteristics extracted by the same neural network. In the training process, the whole network is trained by adopting a cross training method. In the testing process, the detection sub-network and the classification sub-network share the features extracted by the same neural network, so that the parameter quantity and the calculated quantity of the network are greatly reduced, and the rapid target identification effect can be achieved.
Drawings
Fig. 1 is a schematic flow chart of the implementation of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
The invention relates to a complex target accurate identification method based on key area detection, which comprises the following steps: the method comprises the steps of performing fusion training on the whole neural network by using a cross training method, extracting target features by using a convolutional neural network, detecting key regions of a complex target by using a detection sub-network and taking an anchor square frame as reference, pooling the key regions into feature maps with fixed sizes by using regional standard pooling, classifying the key regions by using a classification sub-network, and fusing classification results of all key regions so as to achieve accurate identification of the target. The whole network comprises a key area detection sub-network and a key area classification sub-network, the key area with distinguishing degree of the complex target is detected by the detection sub-network, then the key area is classified by the classification sub-network, and the classification result of each area is fused to identify the whole target. The two sub-networks share the features extracted by the VGG convolutional neural network, so that the complex target is identified quickly and accurately.
As shown in fig. 1, the present invention specifically implements the following steps:
step 1, reading a complex target picture, a coordinate label of a key area corresponding to the complex target picture and a classification label corresponding to the complex target picture in a training sample in a database, and performing fusion training on a complex target accurate identification network by using a cross training method;
step 2, taking the complex target picture to be recognized as the input of the complex target accurate recognition network trained in the step 1, and extracting features through a VGG convolutional neural network to obtain a feature map of the complex target picture to be recognized;
step 3, inputting the feature map obtained in the step 2 into a key area detection sub-network, sliding the feature map by using the sub-network with the size of 3 multiplied by 3, detecting a key area of the complex target picture by using the anchor frame as a reference, and giving a prediction frame of the key area and the possibility P of whether the key area is the key areais,Pnot；
And 4, filtering the detected regions with higher overlapping degree by adopting non-maximum inhibition, and only keeping the possibility P of being the key region when the ratio of the intersection part area and the union part area of different prediction blocks exceeds a specified threshold IOU _ thresholdisThe largest prediction block, while the other blocks are filtered;
step 5, setting the possibility P of the key areaisWill be the critical area probability PisMapping the area larger than the set threshold value P _ threshold to a feature map extracted by the VGG network;
step 6, performing area standard pooling on the areas mapped on the feature maps obtained in the step 5, and pooling the detected areas with different sizes into feature maps with fixed sizes;
step 7, taking the feature map with fixed size obtained in the step 6 as the input of a classification sub-network, accurately classifying the feature map by using the classification sub-network, and normalizing the classification result by using a softmax function to obtain the probability of classifying the key region;
and 8, taking the mean value of the classified corresponding probabilities of the key regions obtained in the step 7 for the same complex target corresponding to the same picture, and fusing to obtain an accurate identification result of the complex target type.
In step 1, the whole network cross training process is as follows:
step 11, fine tuning is carried out on the basis of taking ImageNet database pictures as training samples and taking weights of VGG networks trained aiming at classification tasks as initial weights;
step 12, reading the complex target picture and the coordinate label of the key area corresponding to the complex target picture, and training the sub-network for detecting the key area, wherein the loss function of the training is L ═ LP+LregWherein L isPProbability P of detecting whether sub-network output is a key area for the key areais,PnotCross entropy with the true value of the tag, LregThe square sum of the coordinate offset of the detection area output by the sub-network for detecting the key area and the coordinate offset of the actual key area in the label;
step 13, reading the complex target picture and the classification label corresponding to the complex target picture, training a classification sub-network, wherein the training loss function is the cross entropy between the network output classification result and the actual label result;
step 14 repeats steps 12 and 13 several times, cross-training the key area detection sub-network and the classification sub-network until the network is stable.
In step 3, the method for detecting the key area is as follows:
step 31, using a sliding window with the size of 3 × 3, sliding on the feature map obtained in step 2, and obtaining a 512-dimensional vector at each position;
step 32, setting 9 anchor boxes as reference at the position of each sliding window, setting the length-width ratio of the anchor boxes to be three proportions according to 1:2, 1:1 and 2:1, and setting the area size to be 1282、2562、5122The pixel has three sizes, and the center point of the anchor frame is the center of the sliding window;
step 33, obtaining 512 dimensions of each sliding window positionOutputting 9 vectors with 6 dimensions through a full-connection network; each vector represents the offset d of the coordinates of the center point, the length and the width of the detection area with respect to a reference anchor blockx,dy,dl,dwAnd whether it is a critical area probability Pis,PnotWherein: dx＝(x-xa)/la，dy＝(y-ya)/wa， dl＝log(l/la)，dw＝log(w/wa) X, y, l, w represent coordinates of the center point of the detected region, length and width, xa,ya,la,waRepresenting the coordinates, length and width, P, of the center point of the reference anchor regionis,PnotPerforming normalization processing by using a softmax function;
step 34 offset d from the net regressionx,dy,dl,dwCoordinate with center point of anchor box, length and width xa,ya,la,waAnd calculating the actual central point coordinates, length and width x, y, l and w of the detection area.
In step 6, the process of area standard pooling is as follows:
step 61, representing the size of the region to be pooled as m × n, dividing the region to be pooled into 7 × 7 small lattices with the size of about m/7 × n/7, and rounding up approximately according to rounding up when m/7 or n/7 cannot be rounded up;
in step 62, in each small cell divided in step 61, features in the small cells are pooled into 1 × 1 dimension by using the maximum pooling method, and thus, feature regions of different sizes are pooled into a 7 × 7 dimension fixed-size feature map.
In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (4)

1. A complex target accurate identification method based on key area detection is characterized by comprising the following steps:
step 1, reading a complex target picture, a coordinate label of a key area corresponding to the complex target picture and a classification label corresponding to the complex target picture in a training sample in a database, and performing fusion training on a complex target accurate identification network by using a cross training method;
step 2, taking the complex target picture to be recognized as the input of the complex target accurate recognition network trained in the step 1, and extracting features through a VGG convolutional neural network to obtain a feature map of the complex target picture to be recognized;
step 3, inputting the feature map obtained in the step 2 into a key area detection sub-network, sliding the feature map by using the sub-network with the size of 3 multiplied by 3, detecting a key area of the complex target picture by using the anchor frame as a reference, and giving a prediction frame of the key area and the possibility P of whether the key area is the key areais,Pnot；
And 4, filtering the detected regions with higher overlapping degree by adopting non-maximum inhibition, and only keeping the possibility P of being the key region when the ratio of the intersection part area and the union part area of different prediction blocks exceeds a specified threshold IOU _ thresholdisThe largest prediction block, while the other blocks are filtered;
step 5, setting the possibility P of the key areaisWill be the critical area probability PisMapping the area larger than the set threshold value P _ threshold to a feature map extracted by the VGG network;
step 6, performing area standard pooling on the areas mapped on the feature maps obtained in the step 5, and pooling the detected areas with different sizes into feature maps with fixed sizes;
step 7, taking the feature map with fixed size obtained in the step 6 as the input of a classification sub-network, accurately classifying the feature map by using the classification sub-network, and normalizing the classification result by using a softmax function to obtain the probability of classifying the key region;
and 8, taking the mean value of the classified corresponding probabilities of the key regions obtained in the step 7 for the same complex target corresponding to the same picture, and fusing to obtain an accurate identification result of the complex target type.
2. The method for accurately identifying the complex target based on the key area detection as claimed in claim 1, wherein: in the step 1, the cross training process is as follows:
step 11, fine tuning is carried out on the basis of taking ImageNet database pictures as training samples and taking weights of VGG networks trained aiming at classification tasks as initial weights;
step 12, reading the complex target picture and the coordinate label of the key area corresponding to the complex target picture, and training the sub-network for detecting the key area, wherein the loss function of the training is L ═ LP+LregWherein L isPProbability P of detecting whether sub-network output is a key area for the key areais,PnotCross entropy with the true value of the tag, LregThe square sum of the coordinate offset of the detection area output by the sub-network for detecting the key area and the coordinate offset of the actual key area in the label;
step 13, reading the complex target picture and the classification label corresponding to the complex target picture, training a classification sub-network, wherein the training loss function is the cross entropy between the network output classification result and the actual label result;
step 14 repeats steps 12 and 13 several times, cross-training the key area detection sub-network and the classification sub-network until the network is stable.
3. The method for accurately identifying the complex target based on the key area detection as claimed in claim 1, wherein: the step 3 specifically includes:
step 31, using a sliding window with the size of 3 × 3, sliding on the feature map obtained in step 2, and obtaining a 512-dimensional vector at each position;
step 32, setting 9 anchor boxes as reference at the position of each sliding window, setting the length-width ratio of the anchor boxes to be three proportions according to 1:2, 1:1 and 2:1, and setting the area sizeIs 1282、2562、5122The pixel has three sizes, and the center point of the anchor frame is the center of the sliding window;
step 33, outputting 9 vectors of 6 dimensions from the 512-dimensional vectors obtained at each sliding window position through a full-connection network; each vector represents the offset d of the coordinates of the center point, the length and the width of the detection area with respect to a reference anchor blockx,dy,dl,dwAnd whether it is a critical area probability Pis,PnotWherein: dx＝(x-xa)/la，dy＝(y-ya)/wa，dl＝log(l/la)，dw＝log(w/wa) X, y, l, w represent coordinates of the center point of the detected region, length and width, xa,ya,la,waRepresenting the coordinates, length and width, P, of the center point of the reference anchor regionis,PnotPerforming normalization processing by using a softmax function;
step 34 offset d from the net regressionx,dy,dl,dwCoordinate with center point of anchor box, length and width xa,ya,la,waAnd calculating the actual central point coordinates, length and width x, y, l and w of the detection area.
4. The method for accurately identifying the complex target based on the key area detection as claimed in claim 1, wherein: in step 6, the process of area standard pooling is as follows:
step 61, representing the size of the region to be pooled as m × n, dividing the region to be pooled into 7 × 7 small lattices with the size of about m/7 × n/7, and rounding up approximately according to rounding up when m/7 or n/7 cannot be rounded up;
in step 62, in each small cell divided in step 61, features in the small cells are pooled into 1 × 1 dimension by using the maximum pooling method, and thus, feature regions of different sizes are pooled into a 7 × 7 dimension fixed-size feature map.
CN201810345899.9A 2018-04-18 2018-04-18 Complex target accurate identification method based on key area detection Active CN108537286B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810345899.9A CN108537286B (en) 2018-04-18 2018-04-18 Complex target accurate identification method based on key area detection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810345899.9A CN108537286B (en) 2018-04-18 2018-04-18 Complex target accurate identification method based on key area detection

Publications (2)

Publication Number Publication Date
CN108537286A CN108537286A (en) 2018-09-14
CN108537286B true CN108537286B (en) 2020-11-24

Family

ID=63481345

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810345899.9A Active CN108537286B (en) 2018-04-18 2018-04-18 Complex target accurate identification method based on key area detection

Country Status (1)

Country Link
CN (1) CN108537286B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109242030A (en) * 2018-09-21 2019-01-18 京东方科技集团股份有限公司 Draw single generation method and device, electronic equipment, computer readable storage medium
CN110955380B (en) * 2018-09-21 2021-01-12 中科寒武纪科技股份有限公司 Access data generation method, storage medium, computer device and apparatus
CN109410601A (en) * 2018-12-04 2019-03-01 北京英泰智科技股份有限公司 Method for controlling traffic signal lights, device, electronic equipment and storage medium
CN109829398B (en) * 2019-01-16 2020-03-31 北京航空航天大学 Target detection method in video based on three-dimensional convolution network
CN110852285B (en) * 2019-11-14 2023-04-18 腾讯科技（深圳）有限公司 Object detection method and device, computer equipment and storage medium
CN110929678B (en) * 2019-12-04 2023-04-25 山东省计算中心（国家超级计算济南中心） Method for detecting vulvovaginal candida spores
CN111612797B (en) * 2020-03-03 2021-05-25 江苏大学 Rice image information processing system
CN111931877B (en) * 2020-10-12 2021-01-05 腾讯科技（深圳）有限公司 Target detection method, device, equipment and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106250812A (en) * 2016-07-15 2016-12-21 汤平 A kind of model recognizing method based on quick R CNN deep neural network
CN106599939A (en) * 2016-12-30 2017-04-26 深圳市唯特视科技有限公司 Real-time target detection method based on region convolutional neural network
CN107798335A (en) * 2017-08-28 2018-03-13 浙江工业大学 A kind of automobile logo identification method for merging sliding window and Faster R CNN convolutional neural networks

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107169421B (en) * 2017-04-20 2020-04-28 华南理工大学 Automobile driving scene target detection method based on deep convolutional neural network
CN107368845B (en) * 2017-06-15 2020-09-22 华南理工大学 Optimized candidate region-based Faster R-CNN target detection method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106250812A (en) * 2016-07-15 2016-12-21 汤平 A kind of model recognizing method based on quick R CNN deep neural network
CN106599939A (en) * 2016-12-30 2017-04-26 深圳市唯特视科技有限公司 Real-time target detection method based on region convolutional neural network
CN107798335A (en) * 2017-08-28 2018-03-13 浙江工业大学 A kind of automobile logo identification method for merging sliding window and Faster R CNN convolutional neural networks

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
"A survey on deep learning-based fine-grained object classification and semantic segmentation";Bo Zhao等;《International Journal of Automation and Computing》;20170430;第14卷(第2期);第119-135页 *
"Faster r-cnn: Towards real-time object detection with region proposal networks";Shaoqing Ren等;《IEEE Transactions on Pattern Analysis and Machine Intelligence》;20170601;第39卷(第6期);第1137-1149页 *
"Fine-grained Discriminative Localization via Saliency-guided Faster R-CNN";Xiangteng He等;《MM’17 proceedings of the 25th ACM international conference on multimedia》;20171027;第1-9页 *
"Scene-Adaptive Vehicle Detection Algorithm Based on a Composite Deep Structure";YINGFENG CAI等;《IEEE Access》;20171114;第5卷;第22804-22811页 *
"基于卷积神经网络语义检测的细粒度鸟类识别";李新叶等;《科学技术与工程》;20180408;第18卷(第10期);第240-244页 *
"基于深度学习的车型细粒度识别研究";吴凡;《http://www.doc88.com/p-7708621280922.html》;20171102;第3节 *

Also Published As

Publication number Publication date
CN108537286A (en) 2018-09-14

Similar Documents

Publication Publication Date Title
CN108537286B (en) Complex target accurate identification method based on key area detection
WO2020125216A1 (en) Pedestrian re-identification method, device, electronic device and computer-readable storage medium
CN107833213B (en) Weak supervision object detection method based on false-true value self-adaptive method
CN111080693A (en) Robot autonomous classification grabbing method based on YOLOv3
CN105809651B (en) Image significance detection method based on the comparison of edge non-similarity
CN105512683A (en) Target positioning method and device based on convolution neural network
CN108492298B (en) Multispectral image change detection method based on generation countermeasure network
CN108305260B (en) Method, device and equipment for detecting angular points in image
CN111274926B (en) Image data screening method, device, computer equipment and storage medium
CN105404886A (en) Feature model generating method and feature model generating device
CN108537790B (en) Different-source image change detection method based on coupling translation network
CN110909656B (en) Pedestrian detection method and system integrating radar and camera
CN110610123A (en) Multi-target vehicle detection method and device, electronic equipment and storage medium
CN113705570A (en) Few-sample target detection method based on deep learning
CN111738319A (en) Clustering result evaluation method and device based on large-scale samples
CN112712066B (en) Image recognition method and device, computer equipment and storage medium
CN117315578A (en) Monitoring method and system for rust area expansion by combining classification network
CN111860623A (en) Method and system for counting tree number based on improved SSD neural network
CN109416745B (en) Structured image matching method and system
CN110751623A (en) Joint feature-based defect detection method, device, equipment and storage medium
CN112241736A (en) Text detection method and device
Promsuk et al. Numerical Reader System for Digital Measurement Instruments Embedded Industrial Internet of Things.
CN111027399B (en) Remote sensing image water surface submarine recognition method based on deep learning
CN113658089A (en) Double-data-stream fusion object identification method based on depth camera
CN112926610A (en) Construction method of license plate image screening model and license plate image screening method

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
GR01 Patent grant
GR01 Patent grant