CN110059558A - A kind of orchard barrier real-time detection method based on improvement SSD network - Google Patents
A kind of orchard barrier real-time detection method based on improvement SSD network Download PDFInfo
- Publication number
- CN110059558A CN110059558A CN201910198144.5A CN201910198144A CN110059558A CN 110059558 A CN110059558 A CN 110059558A CN 201910198144 A CN201910198144 A CN 201910198144A CN 110059558 A CN110059558 A CN 110059558A
- Authority
- CN
- China
- Prior art keywords
- network
- ssd
- orchard
- model
- classification
- 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.)
- Granted
Links
Classifications
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
-
- 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
Abstract
The present invention discloses a kind of based on the orchard barrier real-time detection method for improving SSD, using improving SSD deep learning object detection method to identifying in the barrier under the environment of orchard, lightweight network MobileNetV2 is used to extract characteristics of image process spent time and operand as the basic network in SSD model to reduce, auxiliary layer combines empty convolution as basic structure using reversed residual error structure and carries out position prediction so as to avoid down-sampling from operating bring information loss while comprehensive Analysis On Multi-scale Features, use the improved SSD target detection model of corresponding image data set training, and camera acquired image is inputted trained model to detect target position, it solves in traditional detection of obstacles algorithm vulnerable to background interference, Obstacle Position position inaccurate and It is difficult to realize carry out the problems such as various disorders species do not detect simultaneously.
Description
Technical field
The invention belongs to computer visions, deep learning field, and in particular to be under a kind of environment for outdoor orchard
The obstacle detection method of mobile robot intelligent operation.
Background technique
Accurately identifying for farmland barrier is one of essential key technology of unmanned agricultural vehicle with precision agriculture
The development of theoretical proposition and intelligent robot, the self-navigation of reading intelligent agriculture vehicle is increasingly by pass both domestic and external
Note.The agricultural vehicle of independent navigation has the features such as replacing artificial, raising operating efficiency, reducing agriculture production cost.In order to protect
Intelligent System of Vehicle is demonstrate,proved in no manual intervention in the safety of field operation, it is necessary to have real-time detection of obstacles.Field conditions
Under detection of obstacles due to its complicated natural environment, barrier form the external conditions such as variability, illumination it is a wide range of
Variation etc., implements with certain challenge.Under field conditions, there is detection Obstacles position in ultrasonic sensor
Accuracy it is poor, vulnerable to interference the disadvantages of, although laser radar sensor can more intuitively detect barrier, radar system
System involves great expense.Computer Vision Detection compared to other obstacle detection methods have it is at low cost, ring can be efficiently used
The advantages that color and texture information in border.It is automatic that unmanned agricultural machinery is carried out using computer vision methods combination deep learning herein
Pedestrian's detection of obstacles in operation process,
In object detection field, the method accuracy rate based on deep learning greatly exceeds traditional based on HOG, SIFT etc.
The detection method of artificial design features.Target detection based on deep learning mainly includes two classes, and one kind is based on Area generation
Convolutional network structure, representative network is R-CNN serial (R-CNN, fast R-CNN, faster R-CNN)；It is another kind of
It is that the detection of target position is regarded as regression problem, directly whole image is handled using CNN network structure, is predicted simultaneously
The classification of target and position out, representative network have YOLO, SSD (Single Shot MultiBox Detector) etc.,
Speed a kind of method before being generally faster than.
SSD target detection model is not due to needing time-consuming Area generation and feature resampling steps, directly to entire figure
As carry out convolution operation and predict object included in image classification and corresponding coordinate, thus greatly improve detection speed
Degree, while by using the convolution kernel of small size, multi-scale prediction etc. the precision of target detection is greatly improved.SSD
Network structure is divided into basic network (base network) and auxiliary network (auxiliary network) two parts: facilities network
Network typically with the very network of high-class precision and removes its layer of classifying in image classification field to be some；Assist network be
The increased convolutional network structure for target detection on the basis of basic network, the size of these layers be gradually reduced so as into
Row multi-scale prediction.SSD network is excellent in the comprehensive performance of detection speed and precision, and detection speed and precision needs
It is promoted in further, and needs to reduce its operand to meet its requirement for disposing operation on the mobile apparatus.
Summary of the invention
The present invention in view of the above problems, use lightweight network MobileNetV2 as the basic network in SSD model with
It reduces and extracts characteristics of image process spent time and operand, auxiliary layer combines empty convolution as base using reversed residual error structure
Plinth structure carries out position prediction and avoids down-sampling from operating bring information loss while Analysis On Multi-scale Features so as to integrate, with
It carries out real-time detection of obstacles and guarantees that Intelligent System of Vehicle, in the safety of field operation, reduces depth in no manual intervention
The parameter amount and calculation amount of learning model so as to reduce requirement of the deep learning model to hardware and reach real-time with
Meet its application in outdoor mobile device.
The technical solution of the present invention is as follows: it is a kind of based on the orchard barrier real-time detection method for improving SSD network, including with
Lower step:
Step 1, it constructs the data set about orchard environment and data set is divided into training set and test set；
Step 2: on the basis of TensorFlow deep learning frame, SSD network objectives detection model is built, it will
MobileNetV2 uses reversed residual error structure to the auxiliary layer of SSD and combines empty convolution as base as feature extraction network
Plinth convolutional coding structure；
Step 3: the parameter in initialization network model obtains pre-training model；
Step 4: using the training set and test set in step 1, to pre-training model using batch gradient descent algorithm into
Row training, enhances the ability of Model checking false positive using difficult sample Mining Strategy in the training process；
Step 5: deployment SSD network objectives detection model passes through camera collection image and is sent into the detection of SSD network objectives
Model, and remove excess edge frame using non-maxima suppression algorithm, obtain testing result.
Further, the detailed process of step 1 are as follows:
1.1) under the orchard environment by obtaining a large amount of different scenes on the camera that is mounted on corresponding orchard agricultural machinery
Video image obtains the video under a large amount of orchard environment, and extracts picture according to 7.5 frames/second, by all pictures according to 2:1:1 ratio
Example is divided into training set, verifying collection and test set；
1.2) above-mentioned all images are manually marked, the object of mark is obstacle target to be detected, specifically
Markup information be image in target classification and the target bounding box upper left and bottom right coordinate value；
1.3) image of training set is pre-processed, including flip horizontal and translation are to increase sample size while also right
Markup information carries out corresponding processing, and increases the quality of image by adaptive histogram equalization, reduces illumination variation pair
The influence of image.
Further, described using MobileNetV2 as feature extraction network in step 2, the auxiliary layer of SSD is used
Reversed residual error structure simultaneously combines empty convolution as basic convolutional coding structure method particularly includes:
2.1) basis for leaving feature extraction layer as SSD after removing the convolutional layer for being used to classify of MobileNetV2
Network；
2.2) empty convolution is combined with reversed residual error structure and application level Fusion Features strategy solves empty convolution institute band
The calculating discontinuous problem come, so that the feature that the basic structure as auxiliary layer is used to extract basic network carries out position
And the detection of classification.
Further, step 3 method particularly includes:
3.1) being trained on the extensive categorized data set of ImageNet to MobileNetV2 makes it to reaching higher
Classification accuracy；
3.2) the classification convolutional layer for removing MobileNetV2 takes it to be used for the convolutional layer parameter assignment of feature extraction to SSD
Corresponding feature extraction layer；
It 3.3) is mean value with 0 to each layer parameter use of SSD auxiliary layer, 0.01 carries out at random just for the Gaussian Profile of standard deviation
Beginningization.
Further, step 4 method particularly includes:
4.1) batch gradient descent algorithm is trained objective function used in process are as follows:
Wherein N is the number of matched default boundary frame, when wherein when N is 0, directly setting L is that 0, c is mark class
Not, l is the bounding box of prediction, and g is the bounding box of mark, LlocFor the smooth of corresponding position predictionL1Error, LconfIt is right
The more error in classification functions of the softmax answered:
Wherein: Pos is the positive example in sample, and cx, cy are the center point coordinate of prediction block, and w is the width of prediction block, and h is pre-
The height of frame is surveyed,Whether matched with j-th of true frame about classification K for which prediction block,For prediction block,It is true
Frame
Wherein:Whether being matched with true frame j about classification p for prediction block i, Neg is the negative example in sample,For prediction
There is no object in frame,Calculating formula are as follows:
Wherein:It is the probability of p-th of classification for target in i-th of prediction block of target.
4.2) initial positive and negative sample training detection model is first used in training process, then using the model trained to sample
This carries out detection classification, is trained wherein detecting those of mistake sample and continuing to be put into negative sample set, so as to add
The ability of strong Model checking false positive.
Further, step 5 method particularly includes:
5.1) it removes used for preventing the operation of over-fitting and fixed network parameter has been used in training process
In the SSD network objectives detection model of deployment；
5.2) by camera collection image and the input as model, thus obtain several targets classification confidence level and
Bounding box coordinates；
5.3) extra detection block is removed using non-maxima suppression algorithm, obtains more accurate testing result.
Further, non-maxima suppression algorithm specifically: for confidence level corresponding in testing result to testing result
Progress is ranked up from high to low according to confidence level, and calculates corresponding Duplication, and setting Duplication threshold value is 0.5,
Testing result adopts this testing result when having high confidence level and high Duplication threshold value.
The advantages of this programme, is:
1) by transfer learning technology, MobileNetV2 is showed preferable parameter in Imagenet classification and is transplanted to SSD
Feature extraction network model in, thus simplify target detection model training process and shorten the training time.
2) by improving the feature extraction network of original SSD, the more light-weighted MobileNetV2 network model of use into
Row feature extraction, auxiliary layer carries out convolution algorithm using improved reversed residual error structure, so as to utilize multicharacteristic information
And operand is reduced, to improve the accuracy rate and detection speed of model inspection.
3) improved SSD target detection model is used, pedestrian's detection of obstacles under field conditions is carried out, model occupies empty
Between smaller and lightweight, be suitable for disposing on the mobile apparatus, model have preferable robustness, can preferably realize orchard
The detection of barrier under environment provides foundation for avoidance decision.
Detailed description of the invention
Fig. 1 is step figure of the invention.
The improved reversed residual error structure chart of Fig. 2
Fig. 3 hierarchy characteristic fusion structure figure
Empty convolutional layer representation is (# input channel, receptive field, # output channel), wherein empty convolution kernel is effective
Receptive field is nk*nk, nk=(n-1) * 2k-1+1, k=1 ..., K.
The improved SSD target detection model of Fig. 4.
Specific embodiment
Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.
The present invention provides a kind of orchard barrier real-time detection method based on improvement SSD, and this method mainly includes following
Step:
Step 1: data set is simultaneously divided into training set and test set by construction data set, which includes following sub-step:
1.1) under the orchard environment by obtaining a large amount of different scenes on the camera that is mounted on corresponding orchard agricultural machinery
Video image obtains the video under a large amount of orchard environment, and extracts picture according to 7.5 frames/second, by all pictures according to 2:1:1 ratio
Example is divided into training set, verifying collection and test set；
1.2) above-mentioned all images are manually marked, the object of mark is obstacle target to be detected, specifically
Markup information be image in target classification and the target bounding box upper left and bottom right coordinate value.
1.3) image of training set is pre-processed, including flip horizontal and translation are to increase sample size while also right
Markup information carries out corresponding processing, and increases the quality of image by adaptive histogram equalization, reduces illumination variation pair
The influence of image.
Step 2: on the basis of TensorFlow deep learning frame, using MobileNetV2 as feature extraction net
Network uses reversed residual error structure to the auxiliary layer of SSD and combines empty convolution as basic convolutional coding structure.
2.1) basis for leaving feature extraction layer as SSD after removing the convolutional layer for being used to classify of MobileNetV2
Network；
2.2) empty convolution is combined with reversed residual error structure and application level Fusion Features strategy solves empty convolution institute band
The calculating discontinuous problem come, so that the feature that the basic structure as auxiliary layer is used to extract basic network carries out position
And the detection of classification.
It mainly comprises the steps that
1, it is built in TensorFlow deep learning frame and improves SSD algorithm of target detection, by lightweight network model
Convolutional layer conv2d 1x1, avgpool 7x7, the conv2d 1x1 that MobileNetV2 is eventually used for classification are used as SSD after removing
Basal layer for extracting feature.
2, the auxiliary layer convolutional coding structure of SSD target detection model is improved, uses reversed residual error structured set cavity
Convolutional coding structure improves convolutional coding structure, and basic convolution structural unit as auxiliary layer, it is specific as shown in Fig. 2,
The receptive field of convolution kernel can be increased in the case where not having to down-sampling and operating using empty convolution, it is non-in learning process to reduce
Information loss and convolution kernel caused by linear transformation have multiple dimensioned receptive field.
3, it since the introducing of empty convolution will lead to the discontinuous problem of convolution kernel operation, further uses hierarchy characteristic and melts
Hop algorithm negatively affects to eliminate empty convolution bring, and specific implementation is each convolution unit to empty convolutional layer
Output successively sum, and the result after each summation is attached (concatenate) and operates to the end
Output result.See attached drawing 3.
Wherein, the activation primitive that improved reversed residual error structure uses is ReLU6, and ReLU6 is compared to ReLU in low essence
Spending has better robustness in operation scene, in addition, convolution kernel size still uses the convolution kernel of typical 3x3 size,
ReLU6 function is shown in formula (1).
ReLU6=min (max (features, 0), 6) (1)
Finally obtained improved SSD target detection model such as attached drawing 4.
Step 3: the parameter in initialization network model obtains pre-training model.It mainly comprises the steps that
3.1) being trained on the extensive categorized data set of ImageNet to MobileNetV2 makes it to reaching higher
Classification accuracy；
3.2) the classification convolutional layer for removing MobileNetV2 takes it to be used for the convolutional layer parameter assignment of feature extraction to SSD
Corresponding feature extraction layer；To the part basic network MobilenetV2, using being instructed on ImageNet classification task data set
The initialization value of the MobilenetV2 network perfected and the parameter for extracting corresponding network structure as basic network.
It 3.3) is mean value with 0 to each layer parameter use of SSD auxiliary layer, 0.01 carries out at random just for the Gaussian Profile of standard deviation
Beginningization.
Step 4: being trained pre-training model using batch gradient descent algorithm, in the training process using difficulty
Sample Mining Strategy is to enhance the ability of Model checking false positive.Specific training process are as follows:
4.1) batch gradient descent algorithm is trained objective function used in process are as follows:
Wherein N is the number of matched default boundary frame, when wherein when N is 0, directly setting L is that 0, c is mark class
Not, l is the bounding box of prediction, and g is the bounding box of mark, LlocFor the smooth of corresponding position predictionL1Error, LconfIt is right
The more error in classification functions of the softmax answered:
Wherein: Pos is the positive example in sample, and cx, cy are the center point coordinate of prediction block, and w is the width of prediction block, and h is pre-
The height of frame is surveyed,Whether matched with j-th of true frame about classification K for which prediction block,For prediction block,It is true
Frame
Wherein:Whether being matched with true frame j about classification p for prediction block i, Neg is the negative example in sample,For prediction
There is no object in frame,Calculating formula are as follows:
Wherein:It is the probability of p-th of classification for target in i-th of prediction block of target.
4.2) initial positive and negative sample training detection model is first used in training process, then using the model trained to sample
This carries out detection classification, is trained wherein detecting those of mistake sample and continuing to be put into negative sample set, so as to add
The ability of strong Model checking false positive.
Above-mentioned batch gradient descent algorithm setting sample batch size is 128, momentum 0.9, weight attenuation coefficient is 2 ×
10-3, maximum number of iterations is set as 100k, and initial learning rate is 0.004, attenuation rate 0.95, is decayed after every 10000 iteration
Once, and at interval of a model is saved after 10000 iteration, the highest model of precision is finally chosen.
(hard negative mining) strategy is excavated using difficult sample in training process, i.e., is first used in training process
Initial positive and negative sample training detection model, then carries out detection classification to sample using the model trained, wherein detecting
Those of mistake sample, which continues to be put into negative sample set, to be trained, so as to reinforce the ability of Model checking false positive.
Step 5: deployment SSD model passes through camera collection image and is sent into SSD target detection model, and uses non-pole
Big value restrainable algorithms remove excess edge frame, obtain testing result.Specific implementation are as follows:
5.1) it removes used for preventing the operation of over-fitting and fixed network parameter has been used in training process
In the network model of deployment；5.2) by camera collection image and the input as model, to obtain the class of several targets
Other confidence level and bounding box coordinates；5.3) extra detection block is removed using non-maxima suppression algorithm, is more accurately examined
Survey result.
1, trained model parameter and removing dropout etc. prevents the operation of over-fitting to obtain in fixing step four
Final network model.
2, network model is tested and assesses, evaluation index is equal using the reconciliation of precision ratio (P) and recall ratio (R) and the two
Value F1, respectively as shown in formula (2), (3), (4).
TP is the quantity for being correctly detecting pedestrian in formula (2), (3), and it is pedestrian target that FP, which is accidentally non-pedestrian target detection,
Quantity, FN is the quantity for being accidentally background, F pedestrian detection1Value is the harmomic mean to precision ratio and recall ratio, closer to
1, show that model performance is better.
3, it fixes and is deployed in corresponding mobile device expected network model parameter is reached, it is real-time by camera
It obtains in the picture under the environment of orchard and input model, extra detection block is removed using non-maxima suppression, wherein IOU threshold
Value is selected as 0.4, and confidence threshold value is selected as 0.5.
To sum up, of the invention to be primarily adapted for use under the environment of orchard based on the orchard barrier real-time detection method for improving SSD
The self-navigation scene of unmanned agricultural machinery is proposed a kind of based on improvement by the depth of investigation learning objective detection algorithm basic principle
The obstacle detection method of SSD target detection network, it is enterprising in network structure and training process based on SSD detection network
Row improves, and accelerates detection speed to reduce operand, improves detection accuracy, reach the requirement of real-time, and reduce depth
Requirement of the model to hardware is practised, so as to meet the application deployment in movement.This method acquires corresponding video counts first
According to, and picture is extracted with certain frame per second and is labeled, it produces for trained data set.And by transfer learning to depth
Learning model is initialized, and is then trained using batch gradient descent algorithm to model, and trained mould is finally used
Type is in practical obstacle analyte detection task.The present invention can quickly accurately examine to front obstacle under the environment of orchard
It surveys, is to realize that reading intelligent agriculture improves the effective measure of its reliability.
Claims (7)
1. a kind of based on the orchard barrier real-time detection method for improving SSD network, characterized in that the following steps are included:
Step 1, it constructs the data set about orchard environment and data set is divided into training set and test set；
Step 2: on the basis of TensorFlow deep learning frame, SSD network objectives detection model is built, it will
MobileNetV2 uses reversed residual error structure to the auxiliary layer of SSD and combines empty convolution as base as feature extraction network
Plinth convolutional coding structure；
Step 3: the parameter in initialization network model obtains pre-training model；
Step 4: using the training set and test set in step 1, pre-training model being instructed using batch gradient descent algorithm
Practice, enhances the ability of Model checking false positive using difficult sample Mining Strategy in the training process；
Step 5: deployment SSD network objectives detection model passes through camera collection image and is sent into SSD network objectives detection mould
Type, and remove excess edge frame using non-maxima suppression algorithm, obtain testing result.
2. according to claim 1 a kind of based on the orchard barrier real-time detection method for improving SSD network, feature exists
In the detailed process of step 1 are as follows:
1.1) video under the orchard environment by obtaining a large amount of different scenes on the camera that is mounted on corresponding orchard agricultural machinery
Image obtains the video under a large amount of orchard environment, and extracts picture according to 7.5 frames/second, by all pictures according to 2:1:1 ratio point
For training set, verifying collection and test set；
1.2) above-mentioned all images are manually marked, the object of mark is obstacle target to be detected, specific to mark
Infuse the coordinate value that information is the upper left and bottom right of the bounding box of the classification and target of target in image；
1.3) image of training set is pre-processed, including flip horizontal and translation are to increase sample size while also to mark
Information carries out corresponding processing, and increases the quality of image by adaptive histogram equalization, reduces illumination variation to image
Influence.
3. according to claim 1 a kind of based on the orchard barrier real-time detection method for improving SSD network, feature exists
In, it is described using MobileNetV2 as feature extraction network in step 2, reversed residual error structure is used to the auxiliary layer of SSD
And combine empty convolution as basic convolutional coding structure method particularly includes:
2.1) basic network for leaving feature extraction layer as SSD after removing the convolutional layer for being used to classify of MobileNetV2；
2.2) empty convolution is combined with reversed residual error structure and application level Fusion Features strategy solves brought by empty convolution
Discontinuous problem is calculated, so that the feature that the basic structure as auxiliary layer is used to extract basic network carries out position and class
Other detection.
4. according to claim 1 a kind of based on the orchard barrier real-time detection method for improving SSD network, feature exists
In step 3 method particularly includes:
3.1) being trained on the extensive categorized data set of ImageNet to MobileNetV2 makes it to reaching higher classification
Accuracy；
3.2) the classification convolutional layer for removing MobileNetV2 takes it to be used for the convolutional layer parameter assignment of feature extraction corresponding to SSD
Feature extraction layer；
It 3.3) is mean value with 0 to each layer parameter use of SSD auxiliary layer, 0.01 is random initial for the Gaussian Profile progress of standard deviation
Change.
5. according to claim 1 a kind of based on the orchard barrier real-time detection method for improving SSD network, feature exists
In step 4 method particularly includes:
4.1) batch gradient descent algorithm is trained objective function used in process are as follows:
Wherein N is the number of matched default boundary frame, when wherein when N is 0, it is mark classification that directly setting L, which is 0, c, and l is
The bounding box of prediction, g are the bounding box of mark, LlocFor the smooth of corresponding position predictionL1Error, LconfIt is corresponding
The more error in classification functions of softmax:
Wherein: Pos is the positive example in sample, and cx, cy are the center point coordinate of prediction block, and w is the width of prediction block, and h is prediction block
Height,Whether matched with j-th of true frame about classification K for which prediction block,For prediction block,For true frame
Wherein:Whether being matched with true frame j about classification p for prediction block i, Neg is the negative example in sample,For in prediction block
There is no object,Calculating formula are as follows:
Wherein:It is the probability of p-th of classification for target in i-th of prediction block of target.
4.2) first with initial positive and negative sample training detection model in training process, then using the model trained to sample into
Row detection classification continues wherein detection those of mistake sample to be put into negative sample set being trained, so as to reinforce mould
The ability of type differentiation false positive.
6. according to claim 1 a kind of based on the orchard barrier real-time detection method for improving SSD network, feature exists
In step 5 method particularly includes:
5.1) it removes used for preventing the operation of over-fitting and fixed network parameter has been obtained for portion in training process
The SSD network objectives detection model of administration；
5.2) by camera collection image and the input as model, to obtain classification confidence level and the boundary of several targets
Frame coordinate；
5.3) extra detection block is removed using non-maxima suppression algorithm, obtains more accurate testing result.
7. according to claim 6 a kind of based on the orchard barrier real-time detection method for improving SSD network, feature exists
In non-maxima suppression algorithm specifically: confidence level corresponding in testing result carries out according to confidence testing result
Degree is ranked up from high to low, and calculates corresponding Duplication, and setting Duplication threshold value is 0.5, is had in testing result
This testing result is adopted when high confidence level and high Duplication threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910198144.5A CN110059558B (en) | 2019-03-15 | 2019-03-15 | Orchard obstacle real-time detection method based on improved SSD network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910198144.5A CN110059558B (en) | 2019-03-15 | 2019-03-15 | Orchard obstacle real-time detection method based on improved SSD network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110059558A true CN110059558A (en) | 2019-07-26 |
| CN110059558B CN110059558B (en) | 2023-08-25 |
Family
ID=67316148
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910198144.5A Active CN110059558B (en) | 2019-03-15 | 2019-03-15 | Orchard obstacle real-time detection method based on improved SSD network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110059558B (en) |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110633702A (en) * | 2019-11-20 | 2019-12-31 | 江西博微新技术有限公司 | Unmanned aerial vehicle-based line maintenance charge calculation method, server and storage medium |
| CN110728200A (en) * | 2019-09-23 | 2020-01-24 | 武汉大学 | Real-time pedestrian detection method and system based on deep learning |
| CN110807429A (en) * | 2019-10-23 | 2020-02-18 | 西安科技大学 | Construction safety detection method and system based on tiny-YOLOv3 |
| 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 |
| CN110826634A (en) * | 2019-11-11 | 2020-02-21 | 北京百度网讯科技有限公司 | Training method and device of target detection model, electronic equipment and storage medium |
| CN110837771A (en) * | 2019-09-17 | 2020-02-25 | 常熟理工学院 | Bird identification method and device based on convolutional neural network |
| CN110852222A (en) * | 2019-10-31 | 2020-02-28 | 上海交通大学 | Campus corridor scene intelligent monitoring method based on target detection |
| CN110866472A (en) * | 2019-11-04 | 2020-03-06 | 西北工业大学 | Unmanned aerial vehicle ground moving target identification and image enhancement system and method |
| CN111310631A (en) * | 2020-02-10 | 2020-06-19 | 湖南大学 | Target tracking method and system for rotor operation flying robot |
| CN111429425A (en) * | 2020-03-20 | 2020-07-17 | 中冶赛迪重庆信息技术有限公司 | Rolling mill inlet abnormity identification method based on deep learning |
| CN111476129A (en) * | 2020-03-27 | 2020-07-31 | 潍坊申海科技有限公司 | Soil impurity detection method based on deep learning |
| CN111539399A (en) * | 2020-07-13 | 2020-08-14 | 追创科技(苏州)有限公司 | Control method and device of self-moving equipment, storage medium and self-moving equipment |
| CN111666872A (en) * | 2020-06-04 | 2020-09-15 | 电子科技大学 | Efficient behavior identification method under data imbalance |
| CN111860514A (en) * | 2020-05-21 | 2020-10-30 | 江苏大学 | Orchard scene multi-class real-time segmentation method based on improved deep Lab |
| CN111914773A (en) * | 2020-08-07 | 2020-11-10 | 杭州微胜智能科技有限公司 | Equipment and method for capturing illegal boarding and alighting of passengers |
| CN112084985A (en) * | 2020-09-15 | 2020-12-15 | 山东鲁能软件技术有限公司 | CSSD-based multi-target tracking algorithm for electric power overhaul pedestrians |
| CN112270347A (en) * | 2020-10-20 | 2021-01-26 | 西安工程大学 | Medical waste classification detection method based on improved SSD |
| CN112288044A (en) * | 2020-12-24 | 2021-01-29 | 成都索贝数码科技股份有限公司 | News picture attribute identification method of multi-scale residual error network based on tree structure |
| CN112801148A (en) * | 2021-01-14 | 2021-05-14 | 西安电子科技大学 | Fire recognition and positioning system and method based on deep learning |
| CN112818806A (en) * | 2021-01-27 | 2021-05-18 | 国网四川省电力公司电力科学研究院 | Transformer substation inspection robot auxiliary navigation method based on deep learning |
| CN112836623A (en) * | 2021-01-29 | 2021-05-25 | 北京农业智能装备技术研究中心 | Facility tomato farming decision auxiliary method and device |
| CN113392927A (en) * | 2021-07-01 | 2021-09-14 | 哈尔滨理工大学 | Animal target detection method based on single-order deep neural network |
| CN113628206A (en) * | 2021-08-25 | 2021-11-09 | 深圳市捷顺科技实业股份有限公司 | License plate detection method, device and medium |
| CN113657161A (en) * | 2021-07-15 | 2021-11-16 | 北京中科慧眼科技有限公司 | Non-standard small obstacle detection method and device and automatic driving system |
| CN114266977A (en) * | 2021-12-27 | 2022-04-01 | 青岛澎湃海洋探索技术有限公司 | Multi-AUV underwater target identification method based on super-resolution selectable network |
| CN115272665A (en) * | 2022-09-27 | 2022-11-01 | 南京信息工程大学 | Traffic obstacle detection method and system based on improved SSD algorithm |
| CN116612122A (en) * | 2023-07-20 | 2023-08-18 | 湖南快乐阳光互动娱乐传媒有限公司 | Image significance region detection method and device, storage medium and electronic equipment |
| WO2024051296A1 (en) * | 2022-09-07 | 2024-03-14 | 南京邮电大学 | Method and apparatus for obstacle detection in complex weather |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086678A (en) * | 2018-07-09 | 2018-12-25 | 天津大学 | A kind of pedestrian detection method extracting image multi-stage characteristics based on depth supervised learning |
| CN109146967A (en) * | 2018-07-09 | 2019-01-04 | 上海斐讯数据通信技术有限公司 | The localization method and device of target object in image |
-
2019
- 2019-03-15 CN CN201910198144.5A patent/CN110059558B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086678A (en) * | 2018-07-09 | 2018-12-25 | 天津大学 | A kind of pedestrian detection method extracting image multi-stage characteristics based on depth supervised learning |
| CN109146967A (en) * | 2018-07-09 | 2019-01-04 | 上海斐讯数据通信技术有限公司 | The localization method and device of target object in image |
Cited By (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110837771A (en) * | 2019-09-17 | 2020-02-25 | 常熟理工学院 | Bird identification method and device based on convolutional neural network |
| CN110728200A (en) * | 2019-09-23 | 2020-01-24 | 武汉大学 | Real-time pedestrian detection method and system based on deep learning |
| CN110807429A (en) * | 2019-10-23 | 2020-02-18 | 西安科技大学 | Construction safety detection method and system based on tiny-YOLOv3 |
| CN110807429B (en) * | 2019-10-23 | 2023-04-07 | 西安科技大学 | Construction safety detection method and system based on tiny-YOLOv3 |
| CN110852222A (en) * | 2019-10-31 | 2020-02-28 | 上海交通大学 | Campus corridor scene intelligent monitoring method based on target detection |
| CN110866472A (en) * | 2019-11-04 | 2020-03-06 | 西北工业大学 | Unmanned aerial vehicle ground moving target identification and image enhancement system and method |
| CN110826634A (en) * | 2019-11-11 | 2020-02-21 | 北京百度网讯科技有限公司 | Training method and device of target detection model, electronic equipment and 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 |
| CN110633702A (en) * | 2019-11-20 | 2019-12-31 | 江西博微新技术有限公司 | Unmanned aerial vehicle-based line maintenance charge calculation method, server and storage medium |
| CN111310631B (en) * | 2020-02-10 | 2021-05-07 | 湖南大学 | Target tracking method and system for rotor operation flying robot |
| CN111310631A (en) * | 2020-02-10 | 2020-06-19 | 湖南大学 | Target tracking method and system for rotor operation flying robot |
| CN111429425A (en) * | 2020-03-20 | 2020-07-17 | 中冶赛迪重庆信息技术有限公司 | Rolling mill inlet abnormity identification method based on deep learning |
| CN111429425B (en) * | 2020-03-20 | 2023-04-07 | 中冶赛迪信息技术(重庆)有限公司 | Rolling mill entrance abnormity identification method based on deep learning |
| CN111476129A (en) * | 2020-03-27 | 2020-07-31 | 潍坊申海科技有限公司 | Soil impurity detection method based on deep learning |
| CN111860514A (en) * | 2020-05-21 | 2020-10-30 | 江苏大学 | Orchard scene multi-class real-time segmentation method based on improved deep Lab |
| CN111666872B (en) * | 2020-06-04 | 2022-08-05 | 电子科技大学 | Efficient behavior identification method under data imbalance |
| CN111666872A (en) * | 2020-06-04 | 2020-09-15 | 电子科技大学 | Efficient behavior identification method under data imbalance |
| CN111539399B (en) * | 2020-07-13 | 2021-06-29 | 追创科技(苏州)有限公司 | Control method and device of self-moving equipment, storage medium and self-moving equipment |
| CN111539399A (en) * | 2020-07-13 | 2020-08-14 | 追创科技(苏州)有限公司 | Control method and device of self-moving equipment, storage medium and self-moving equipment |
| CN111914773A (en) * | 2020-08-07 | 2020-11-10 | 杭州微胜智能科技有限公司 | Equipment and method for capturing illegal boarding and alighting of passengers |
| CN112084985A (en) * | 2020-09-15 | 2020-12-15 | 山东鲁能软件技术有限公司 | CSSD-based multi-target tracking algorithm for electric power overhaul pedestrians |
| CN112270347A (en) * | 2020-10-20 | 2021-01-26 | 西安工程大学 | Medical waste classification detection method based on improved SSD |
| CN112288044A (en) * | 2020-12-24 | 2021-01-29 | 成都索贝数码科技股份有限公司 | News picture attribute identification method of multi-scale residual error network based on tree structure |
| CN112801148A (en) * | 2021-01-14 | 2021-05-14 | 西安电子科技大学 | Fire recognition and positioning system and method based on deep learning |
| CN112818806A (en) * | 2021-01-27 | 2021-05-18 | 国网四川省电力公司电力科学研究院 | Transformer substation inspection robot auxiliary navigation method based on deep learning |
| CN112836623A (en) * | 2021-01-29 | 2021-05-25 | 北京农业智能装备技术研究中心 | Facility tomato farming decision auxiliary method and device |
| CN112836623B (en) * | 2021-01-29 | 2024-04-16 | 北京农业智能装备技术研究中心 | Auxiliary method and device for agricultural decision of facility tomatoes |
| CN113392927A (en) * | 2021-07-01 | 2021-09-14 | 哈尔滨理工大学 | Animal target detection method based on single-order deep neural network |
| CN113657161A (en) * | 2021-07-15 | 2021-11-16 | 北京中科慧眼科技有限公司 | Non-standard small obstacle detection method and device and automatic driving system |
| CN113628206A (en) * | 2021-08-25 | 2021-11-09 | 深圳市捷顺科技实业股份有限公司 | License plate detection method, device and medium |
| CN114266977B (en) * | 2021-12-27 | 2023-04-07 | 青岛澎湃海洋探索技术有限公司 | Multi-AUV underwater target identification method based on super-resolution selectable network |
| CN114266977A (en) * | 2021-12-27 | 2022-04-01 | 青岛澎湃海洋探索技术有限公司 | Multi-AUV underwater target identification method based on super-resolution selectable network |
| WO2024051296A1 (en) * | 2022-09-07 | 2024-03-14 | 南京邮电大学 | Method and apparatus for obstacle detection in complex weather |
| CN115272665A (en) * | 2022-09-27 | 2022-11-01 | 南京信息工程大学 | Traffic obstacle detection method and system based on improved SSD algorithm |
| CN116612122A (en) * | 2023-07-20 | 2023-08-18 | 湖南快乐阳光互动娱乐传媒有限公司 | Image significance region detection method and device, storage medium and electronic equipment |
| CN116612122B (en) * | 2023-07-20 | 2023-10-10 | 湖南快乐阳光互动娱乐传媒有限公司 | Image significance region detection method and device, storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110059558B (en) | 2023-08-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059558A (en) | A kind of orchard barrier real-time detection method based on improvement SSD network | |
| Yang et al. | Real-time face detection based on YOLO | |
| CN109934115B (en) | Face recognition model construction method, face recognition method and electronic equipment | |
| CN109800689B (en) | Target tracking method based on space-time feature fusion learning | |
| CN109785337B (en) | In-column mammal counting method based on example segmentation algorithm | |
| CN111178197B (en) | Mass R-CNN and Soft-NMS fusion based group-fed adherent pig example segmentation method | |
| CN113065558A (en) | Lightweight small target detection method combined with attention mechanism | |
| Xie et al. | Detecting trees in street images via deep learning with attention module | |
| CN110765865B (en) | Underwater target detection method based on improved YOLO algorithm | |
| CN111401293B (en) | Gesture recognition method based on Head lightweight Mask scanning R-CNN | |
| CN113160062B (en) | Infrared image target detection method, device, equipment and storage medium | |
| Shen et al. | Biomimetic vision for zoom object detection based on improved vertical grid number YOLO algorithm | |
| Wan et al. | A real-time branch detection and reconstruction mechanism for harvesting robot via convolutional neural network and image segmentation | |
| CN110991444A (en) | Complex scene-oriented license plate recognition method and device | |
| CN113435355A (en) | Multi-target cow identity identification method and system | |
| CN114898470A (en) | Fall behavior detection method and system based on improved YOLOv5 | |
| CN116824413A (en) | Aerial image target detection method based on multi-scale cavity convolution | |
| Srinidhi et al. | Pothole detection using CNN and AlexNet | |
| Liu et al. | An improved YOLOv5-based approach to soybean phenotype information perception | |
| CN117079095A (en) | Deep learning-based high-altitude parabolic detection method, system, medium and equipment | |
| CN116740758A (en) | Bird image recognition method and system for preventing misjudgment | |
| CN110826575A (en) | Underwater target identification method based on machine learning | |
| CN115995017A (en) | Fruit identification and positioning method, device and medium | |
| Wu et al. | Research on Asphalt Pavement Disease Detection Based on Improved YOLOv5s | |
| CN113379738A (en) | Method and system for detecting and positioning epidemic trees based on images |
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 |