CN109684967A - A kind of soybean plant strain stem pod recognition methods based on SSD convolutional network - Google Patents
A kind of soybean plant strain stem pod recognition methods based on SSD convolutional network Download PDFInfo
- Publication number
- CN109684967A CN109684967A CN201811540821.9A CN201811540821A CN109684967A CN 109684967 A CN109684967 A CN 109684967A CN 201811540821 A CN201811540821 A CN 201811540821A CN 109684967 A CN109684967 A CN 109684967A
- Authority
- CN
- China
- Prior art keywords
- image
- ssd
- soybean plant
- plant strain
- 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/35—Categorising the entire scene, e.g. birthday party or wedding scene
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of soybean plant strain stem pod recognition methods based on SSD convolutional network, comprising the following steps: acquisition single plant soybean sample image obtains soybean plant strain image library;Manual mark is carried out to stem pod, beanpod marks the beanpod tip not being blocked, and stalk marks stalk exposed part, and image library is divided into training set, verifying collection, test set without duplicate;Random image enhancing and data amplification are carried out to the training set image marked, and automatic mark again increases image newly;SSD convolutional network is constructed, with the characteristic pattern of different levels, carries out multiple scale detecting;The training set is randomly selected to the training for being used for SSD convolutional neural networks, and determines the learning parameter in the SSD convolutional neural networks;Test set is transported in trained SSD convolutional neural networks and carries out identification test, and by recognition result label in the original image in the test sample.A kind of soybean plant strain stem pod recognition methods based on SSD convolutional network is provided, intelligentized identification is carried out to soybean plant strain stem pod by network training, high degree of automation effectively improves the efficiency to the detection of soybean plant strain stem pod.
Description
Technical field
The present invention relates to the technical fields of Computer Image Processing recognition methods, more particularly, to a kind of based on SSD's
The soybean plant strain stem pod recognition methods of convolutional network.
Background technique
Soybean is the main source of the important grain and oil dual-purpose crop in the world and mankind's high-quality protein.It is both that China is main
One of crop, and the crop of most economic benefit.The data of whole strain soybean trait are collected, arrange by soybean species test work
And statistics, it is an important link during the genetic breeding for analyzing soybean crops.It is used currently, soybean species test work is main
Manual operation, however manual operation not only consumes a large amount of manpower and material resources, exists simultaneously human error, brings later data point
The inaccuracy of analysis.
Summary of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide a kind of plants of the soybean of convolutional network based on SSD
Machine vision is applied to soybean species test work, can not only improve soybean plant strain trait data precision by strain stem pod recognition methods,
Human error is reduced, shortens the species test period, moreover it is possible to reduce labor intensity, develop to intelligent, rapid, accuracy direction.
In order to solve the above technical problems, the technical solution adopted by the present invention is that:
A kind of soybean plant strain stem pod recognition methods of convolutional network based on SSD is provided, comprising the following steps:
S1. single plant is acquired by the position that Canon's 5D Mark II camera is fixed on apart from blue background cloth 120cm
Soybean sample image obtains soybean plant strain image library;
S2. sample image all in image library described in traversal step S1 carries out beanpod, stalk to every sample image
Manual mark, the beanpod tip not being blocked is labeled as beanpod class, stalk exposed part marks stalk class, obtains original graph
Image set;
S3. random image enhancing and data amplification are carried out for the training set image marked described in step S2.It adopts
Image enhancement is carried out with adaptive histogram equalization;Using the random adjustment in RGB color channel in certain threshold value and it is horizontal,
Vertical mirror overturning and Random-Rotation, translation carry out data amplification, and are cut between two parties to the image after rotation, translation
It takes, image after processing abandons mark if target is beyond boundary, obtains enhancing amplification training set;
S4. SSD convolutional network is constructed, with the characteristic pattern of different levels, carries out multiple scale detecting;
S5. training sample in step S2, S3 is transported to SSD convolutional neural networks progress pre-training and iteration pre-training obtains
Model after to pre-training, while determining the learning parameter in the SSD convolutional neural networks;
S6. the test set is transported in the trained SSD convolutional neural networks and carries out identification test, and will known
Output recognition result of classification results of the confidence level greater than 40% as test sample in other result.
Soybean plant strain stem pod recognition methods based on SSD convolutional network of the invention, joined fused layer for the thin of low layer
Section is directly delivered to a kind of high-rise residual error structure, takes full advantage of the more rulers of characteristic pattern progress that SSD network chooses different levels
The characteristics of degree detection, existing method is compensated for deforming, block, the defect that continuous overlapped objects detection accuracy is poor.The present invention
Have the advantages that convolutional neural networks, while reducing the interference of image background, ambient brightness have relatively by force to blocking and being overlapped
Anti-interference ability, improve soybean plant strain stem pod detection accuracy rate.
Preferably, the manual mark for carrying out beanpod, stalk described in step S2 to every sample image, will not be blocked
Beanpod tip be labeled as beanpod class, stalk exposed part marks stalk class.The only Partial Feature of label target can be improved and hide
Recognition accuracy under gear, overlapping cases.
Preferably, random image enhancing is carried out for the training set image marked described in step S2 and data expand
Increase.Image enhancement is carried out using adaptive histogram equalization;Using in certain threshold value RGB color channel it is random adjustment and
Horizontal, vertical mirror overturning and Random-Rotation, translation carry out data amplification, and occupy to the image after rotation, translation
Middle interception, the method that image after processing abandons mark if target is beyond boundary.
Preferably, SSD model described in step S4 increases by one layer of fused layer and four convolutional layers in VGG-16 network
It constructs, the establishment step of training pattern is as follows in step S4:
It S41. is input, the characteristic pattern obtained in convolutional layer to image convolution operation with soybean plant strain sample image;
S42. to increasing Add4_3 layers in VGG-16 network, Add4_3 is by two characteristic patterns of Maxpool3 and Conv4_2
It after merging (Add), is activated through ReLU, and by being constituted after Batch Normalization (BN) normalization, Add4_3 conduct
Conv4_3 layers of input, Conv4_3 layer in network, Fc7 layers, Conv8_2~Conv11_2 layers of characteristic pattern is with volume 3 × 3
Product core carries out convolution, realizes the confidence level of output category and the location information of output regression respectively;
S43. all export structures are merged, handles to obtain testing result by non-maxima suppression.
Characteristic pattern by choosing six different levels carries out multiple scale detecting, is retaining the detection to high-level characteristic figure
On the basis of, increase the fusion to low-level feature figure, not only takes full advantage of the rich image detailed information of low-level feature figure, but also reach
The target detection robustness effect of anti-interference ability, detection and positioning when solving deform, block, being continuously overlapped are asked
Topic.
Preferably, when the confidence level of output category, each frame generates the confidence level of two classifications;Output regression
When location information, four coordinate values (x, y, w, h) of generation of each frame.
Preferably, characteristic pattern carries out operation as follows in step S41:
Step 1: the characteristic pattern that Conv4_3 layers export being divided into 76 × 38 units, uses four kinds of defaults on each unit
Bounding box carries out convolution algorithm using the convolution kernel that size is 3 × 3 on each default boundary frame, exports four elements of frame,
It is that transverse and longitudinal coordinate x, the y for exporting the upper left corner of frame and frame return in wide w, the high h and frame of the exported frame of layer respectively
Object be belonging respectively to the confidence level of beanpod and stalk;
Step 2: according to identical method in step S411 successively in Fc7 layers, the Conv8_2~Conv11_2 layers of spy exported
It is calculated on sign figure;Wherein, each layer characteristic pattern is respectively divided into 38 × 19,20 × 10,1 × 5,6 × 3,1 × 1 units,
Default boundary frame number used in each unit is respectively 6,6,6,4,4.
Preferably, less than 15%, the average value of test error is less than the training error of model after pre-training in step S4
20%.
Compared with prior art, the beneficial effects of the present invention are:
(1) present invention can be realized the detection of soybean plant strain stem pod and positioning, accuracy rate with higher, and has and stablize
Property good, strong antijamming capability, the advantages that versatility is high, to deform, block, the detection accuracy of continuous overlay target it is high, just enough answer
With soybean plant strain character detection system.
(2) present invention has the advantages that convolutional neural networks, reduces the interference of image background, ambient brightness, to block with
It is overlapped to have strong anti-interference ability, improve the accuracy rate of soybean plant strain stem pod detection.
Detailed description of the invention
Fig. 1 is the flow chart of the soybean plant strain stem pod recognition methods of the convolutional network based on SSD;
Fig. 2 is the specific flow chart of step S4;
Fig. 3 is the sample image of the soybean plant strain after identifying in the present invention.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent;In order to better illustrate this embodiment, attached
Figure has omission, zooms in or out, and does not represent actual size;To those skilled in the art, certain public affairs in attached drawing
Know that structure and its explanation may be omitted and be will be understood by.
The present invention is further illustrated with reference to the accompanying drawings and detailed description.
Referring to Fig. 1, the present embodiment is the soybean plant strain stem pod recognition methods of the convolutional network of the invention based on SSD
First embodiment, comprising the following steps:
S1. the position apart from blue background cloth 120cm is fixed on by Canon's 5D Mark II camera and carries out single plant soybean
Sample image acquisition, obtains soybean plant strain image library;
S2. sample image all in image set described in traversal step S1 carries out beanpod, stalk to every sample image
Manual mark, the beanpod tip not being blocked is labeled as beanpod class, stalk exposed part marks stalk class, obtains original instruction
Practice collection;
S3. random image enhancing and data amplification are carried out for the training set image marked described in step S2.It adopts
Image enhancement is carried out with adaptive histogram equalization;Using the random adjustment in RGB color channel in certain threshold value and it is horizontal,
Vertical mirror overturning and Random-Rotation, translation carry out data amplification, and are cut between two parties to the image after rotation, translation
It takes, image after processing abandons mark if target is beyond boundary, obtains enhancing amplification training set;
Referring to Fig. 2, the step S4. of the present embodiment constructs SSD convolutional network, with the characteristic pattern of different levels, carry out more
Size measurement;
S5. training sample in step S2, S3 is transported to SSD convolutional neural networks progress pre-training and iteration pre-training obtains
Model after to pre-training, while determining the learning parameter in the SSD convolutional neural networks;
S6. the test set is transported in the trained SSD convolutional neural networks and carries out identification test, and will known
Output recognition result of classification results of the confidence level greater than 40% as test sample in other result.
In step S1, the position apart from blue background cloth 120cm is fixed on by Canon's 5D Mark II camera and carries out list
Strain soybean sample image acquisition, obtains soybean plant strain image library.Specifically, plant sample image collection stores sample using following form
Notebook data:
{ image_name, x, y }
Wherein, image_name indicates that soybean plant strain Image Name, x indicate image horizontal pixel value, and y indicates image longitudinal direction picture
Element value.
In step S2, all sample images in image library described in traversal step S1 carry out beans to every sample image
Pod, the manual of stalk mark, and the beanpod tip not being blocked are labeled as beanpod class, stalk exposed part marks stalk class, obtains
Obtain original image set.Specifically, plant sample image is labeled to form image labeling collection to wherein each true frame, image mark
Note collection stores flag data using following form:
{ label, xmin, ymin, xmax, ymax }
Wherein, label indicates the classification of mark, and xmin indicates that the abscissa of the minimum image vegetarian refreshments of mark, ymin indicate mark
The ordinate of the minimum image vegetarian refreshments of note, xmax indicate that the abscissa of the maximum pixel point of mark, ymax indicate the maximum picture of mark
The ordinate of vegetarian refreshments.
In step S4, the establishment step of model is as follows after pre-training:
S41. it is input with soybean plant strain sample image, the feature that convolution algorithm obtains is carried out to image in convolutional layer
Figure;
S42. to increasing Add4_3 layers in VGG-16 network, Add4_3 is by two characteristic patterns of Maxpool3 and Conv4_2
It after merging (Add), is activated through ReLU, and by being constituted after Batch Normalization (BN) normalization, Add4_3 conduct
Conv4_3 layers of input, Conv4_3 layer in network, Fc7 layers, Conv8_2~Conv11_2 layers of characteristic pattern is with volume 3 × 3
Product core carries out convolution, realizes the confidence level of output category and the location information of output regression respectively;
S43. all export structures are merged, handles to obtain testing result by non-maxima suppression;Wherein, output point
The confidence level of class is the confidence level of classification in each prediction block;The location information of output regression is four of each prediction block
Coordinate value (x, y, w, h).
Wherein, the characteristic pattern in step S41 carries out operation as follows:
Step 1: the characteristic pattern that Conv4_3 layers export being divided into 76 × 38 units, uses four kinds of defaults on each unit
Bounding box carries out convolution algorithm using the convolution kernel that size is 3 × 3 on each default boundary frame, exports four elements of frame,
It is that transverse and longitudinal coordinate x, the y for exporting the upper left corner of frame and frame return in wide w, the high h and frame of the exported frame of layer respectively
Object be belonging respectively to the confidence level of beanpod and stalk;
Step 2: according to identical method in step S411 successively in Fc7 layers, the Conv8_2~Conv11_2 layers of spy exported
It is calculated on sign figure;Wherein, each layer characteristic pattern is respectively divided into 38 × 19,20 × 10,10 × 5,6 × 3,1 × 1 units,
Default boundary frame number used in each unit is respectively 6,6,6,4,4.
Characteristic pattern by choosing six different levels carries out multiple scale detecting, is retaining the detection to high-level characteristic figure
On the basis of, increase the fusion to low-level feature figure, not only takes full advantage of the rich image detailed information of low-level feature figure, but also reach
The target detection robustness effect of anti-interference ability, detection and positioning when solving deform, block, being continuously overlapped are asked
Topic.
Increase the VGG-16 subnetwork structure of residual error structure in the present embodiment are as follows:
First layer, continuous use 64 convolution filters that size is 3 × 3 twice, stride 1 are filled (padding)
It is 1, obtains two 600 × 300 × 64 convolutional layers (Conv1_1, Conv1_2), after the output for obtaining convolutional layer, use BN layers
(batch normalization) is normalized, and then uses ReLU function (Rectified Linear Units)
It is activated as nonlinear activation function, the maximum pond layer (Maxpooling) for being again finally 2 × 2 with a window size
Pond is carried out, the sampling stride of maximum pond layer (Maxpooling) is 2.
The second layer, continuous use 128 convolution filters that size is 3 × 3 twice, stride 1 are filled (padding)
It is 1, obtains two 300 × 150 × 128 convolutional layers (Conv2_1, Conv2_2) and use BN after the output for obtaining convolutional layer
Layer (batch normalization) is normalized, and then uses ReLU function (Rectified Linear
Units it) is activated as nonlinear activation function, the maximum pond layer for being again finally 2 × 2 with a window size
(Maxpooling) pond is carried out, the sampling stride of maximum pond layer (Maxpooling) is 2.
Third layer, continuous use 256 convolution filters that size is 3 × 3 three times, stride 1 are filled (padding)
It is 1, obtains three 150 × 75 × 256 convolutional layers (Conv3_1, Conv3_2, Conv3_3), after the output for obtaining convolutional layer,
It is normalized using BN layers (batch normalization), then uses ReLU function (Rectified Linear
Units it) is activated as nonlinear activation function, the maximum pond layer for being again finally 2 × 2 with a window size
(Maxpooling) pond is carried out, the sampling stride of maximum pond layer (Maxpooling) is 2.
4th layer, 512 convolution filters for the use of secondary size being respectively 3 × 3 a, size is 512 of 1 × 1
Convolution filter, 512 convolution filters that a size is 3 × 3, stride 1, filling (padding) are 1, obtain three
76 × 38 × 512 convolutional layer (Conv4_1, Conv4_2, Add4_3, Conv4_3) uses BN after the output for obtaining convolutional layer
Layer (batch normalization) is normalized, and then uses ReLU function (Rectified Linear
Units it) is activated as nonlinear activation function, the maximum pond layer for being again finally 2 × 2 with a window size
(Maxpooling) pond is carried out, the sampling stride of maximum pond layer (Maxpooling) is 2.
Layer 5, continuous use 512 convolution filters that size is 3 × 3 three times, stride 1 are filled (padding)
It is 1, obtains three 38 × 19 × 512 convolutional layers (Conv5_1, Conv5_2, Conv5_3), after the output for obtaining convolutional layer,
It is normalized using BN layers (batch normalization), then uses ReLU function (Rectified Linear
Units it) is activated as nonlinear activation function.
Then, 1024 convolution filters for the use of size being 3 × 3 to the output of Conv5_3, stride 1, filling
(padding) it is 1, obtains the Fc6 layer that size is 38 × 19 × 1024, then use 1024 volumes that size is 1 × 1 to Fc6 layers
Product filter, stride 1, filling (padding) obtain the Fc7 layer that size is 38 × 19 × 1024 for 1.
Finally, Fc7 layers below increase by four convolutional layers, be respectively size be 20 × 10 × 512 Conv8 layer, 10 ×
5 × 256 Conv9 layer, 6 × 3 × 256 Conv10 layer, 1 × 1 × 256 Conv11 layer.
The training error of model is less than 15% after pre-training in step S4, and the average value of test error is less than 20%.Model
The calculation method of training error is as follows:
Step 1: each true frame being matched with the default boundary frame that corresponding maximum jaccard coefficient is overlapped, and will be write from memory
Recognize bounding box any true frame matching greater than 0.5 Chong Die with jaccard coefficient;
Step 2:i indicates default frame serial number, and j indicates true frame serial number, and p indicates classification sequence number, and 0 is background, and 1 is beanpod,
2 be stalk,Whether to match, the threshold value that is greater than Chong Die with true frame maximum jaccard coefficient be matching value is 1,
It otherwise is 0.
Step 3: total target loss function L (x, c, l, g) loses L by positioninglocL is lost with confidence levelconfWeighted sum
It obtains:
In formula, N is the number of the default boundary frame to match with true frame, LlocFor positioning loss, LconfFor confidence level damage
It losing, x indicates that training sample, c indicate the confidence level of each type objects, and l represents prediction block, and g represents true frame, and α indicates weight, this
α in embodiment is set as 0.8;
Positioning loss LlocIn, f (x) is the sectionally smooth function controlled with σ, and d represents default frame, w indicate true frame or
The width of default boundary frame, h indicate the height of true frame or default boundary frame, and i indicates i-th of default frame, and j-th of j expression true
Real frame, m indicate that (wherein, cx represents central point x-axis coordinate for the location information of true frame or default boundary frame;Cy represents central point y
Axial coordinate;W represents the width of frame;H represents the height of frame), p indicates p-th of classification:
In formula,
Confidence level loses LconfFor multi-class softmax loss function, such as formula 8.Wherein,Indicate i-th of default frame, the
J true frames correspond to the prediction probability of classification p, and calculation formula is as follows:
The method of the present invention can in soybean plant strain image beanpod and stalk accurately detect with positioning, to blocking and again
It is folded that interference is waited to have strong anti-interference ability, improve the accuracy rate of soybean plant strain stem pod detection.Wherein, the sample image
It is as shown in Figure 3 to identify result.
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention
Protection scope within.
Claims (7)
1. a kind of soybean plant strain stem pod recognition methods of convolutional network based on SSD, which comprises the following steps:
S1. the position apart from blue background cloth 120cm is fixed on by Canon's 5D Mark II camera and carries out single plant soybean sample
Image Acquisition obtains soybean plant strain image library;
S2. sample image all in image library described in traversal step S1 carries out the hand of beanpod, stalk to every sample image
The beanpod tip not being blocked is labeled as beanpod class by work mark, and stalk exposed part marks stalk class, obtains original image
Library, and image library is divided into training set, verifying collection, test set without duplicate;
S3. random image enhancing and data amplification are carried out for the training set image marked described in step S2.Using certainly
Adaptive histogram equalizationization carries out image enhancement;Using the random adjustment in RGB color channel in certain threshold value and horizontal, vertical
Mirror image switch and Random-Rotation, translation carry out data amplification, and are intercepted between two parties to the image after rotation, translation, pass through
Treated, and image abandons mark if target is beyond boundary, obtains enhancing amplification training set;
S4. SSD convolutional network is constructed, with the characteristic pattern of different levels, carries out multiple scale detecting;
S5. by training sample in step S2, S3 be transported to SSD convolutional neural networks carry out pre-training and iteration pre-training obtain it is pre-
Model after training, while determining the learning parameter in the SSD convolutional neural networks;
S6. the test set is transported in the trained SSD convolutional neural networks and carries out identification test, and identification is tied
Output recognition result of classification results of the confidence level greater than 40% as test sample in fruit.
2. the soybean plant strain stem pod recognition methods of the convolutional network according to claim 1 based on SSD, which is characterized in that
In step S2, the manual mark of beanpod, stalk is carried out to every sample image, the beanpod tip not being blocked is labeled as beanpod
Class, stalk exposed part mark stalk class.
3. the soybean plant strain stem pod recognition methods of the convolutional network according to claim 1 based on SSD, which is characterized in that
The training set image marked described in step S3 carries out random image enhancing and data amplification.It is equal using self-adapting histogram
Weighing apparatusization carries out image enhancement;Using the random adjustment in RGB color channel in certain threshold value and horizontal, vertical mirror overturning and with
Machine rotation, translation carry out data amplification, and are intercepted between two parties to the image after rotation, translation, image after processing
The method of mark is abandoned if target is beyond boundary.
4. the soybean plant strain stem pod recognition methods of the convolutional network according to claim 1 based on SSD, which is characterized in that
SSD model described in step S4 increases by one layer of fused layer and four convolution layer buildings in VGG-16 network, instructs in step S4
The establishment step for practicing model is as follows:
It S41. is input, the characteristic pattern obtained in convolutional layer to image convolution operation with soybean plant strain sample image;
S42. to increasing Add4_3 layers in VGG-16 network, Add4_3 is by two characteristic pattern fusions of Maxpool3 and Conv4_2
(Add) it after, is activated through ReLU, and by constituting after Batch Normalization (BN) normalization, Add4_3 is as Conv4_3
The input of layer, Conv4_3 layer in network, Fc7 layers, Conv8_2~Conv11_2 layers of characteristic pattern carries out with 3 × 3 convolution kernels
Convolution realizes the confidence level of output category and the location information of output regression respectively;
S43. all export structures are merged, handles to obtain testing result by non-maxima suppression.
5. the soybean plant strain stem pod recognition methods of the convolutional network according to claim 4 based on SSD, which is characterized in that
When the confidence level of output category, each frame generates the confidence level of two classifications;When the location information of output regression, each
Four coordinate values (x, y, w, h) of generation of frame.
6. the soybean plant strain stem pod recognition methods of the convolutional network according to claim 4 based on SSD, which is characterized in that
Characteristic pattern carries out operation as follows in step S41:
Step 1: the characteristic pattern that Conv4_3 layers export being divided into 76 × 38 units, uses four kinds of default boundaries on each unit
Frame carries out convolution algorithm using the convolution kernel that size is 3 × 3 on each default boundary frame, exports four elements of frame, respectively
It is to export transverse and longitudinal coordinate x, the y in the upper left corner of frame and wide w, the high h of frame recurrence the exported frame of layer and the object in frame
Body is belonging respectively to the confidence level of beanpod and stalk;
Step 2: according to identical method in step S411 successively in Fc7 layers, the Conv8_2~Conv11_2 layers of characteristic pattern exported
On calculated;Wherein, each layer characteristic pattern is respectively divided into 38 × 19,20 × 10,10 × 5,6 × 3,1 × 1 units, each
Default boundary frame number used in unit is respectively 6,6,6,4,4.
7. the soybean plant strain stem pod recognition methods of the convolutional network according to claim 1 based on SSD, which is characterized in that
The training error of model is less than 15% after pre-training in step S4, and the average value of test error is less than 20%.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811540821.9A CN109684967A (en) | 2018-12-17 | 2018-12-17 | A kind of soybean plant strain stem pod recognition methods based on SSD convolutional network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811540821.9A CN109684967A (en) | 2018-12-17 | 2018-12-17 | A kind of soybean plant strain stem pod recognition methods based on SSD convolutional network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109684967A true CN109684967A (en) | 2019-04-26 |
Family
ID=66187871
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811540821.9A Pending CN109684967A (en) | 2018-12-17 | 2018-12-17 | A kind of soybean plant strain stem pod recognition methods based on SSD convolutional network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109684967A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110100774A (en) * | 2019-05-08 | 2019-08-09 | 安徽大学 | River crab male and female recognition methods based on convolutional neural networks |
CN110110702A (en) * | 2019-05-20 | 2019-08-09 | 哈尔滨理工大学 | It is a kind of that algorithm is evaded based on the unmanned plane for improving ssd target detection network |
CN110443778A (en) * | 2019-06-25 | 2019-11-12 | 浙江工业大学 | A method of detection industrial goods random defect |
CN110602411A (en) * | 2019-08-07 | 2019-12-20 | 深圳市华付信息技术有限公司 | Method for improving quality of face image in backlight environment |
CN110781870A (en) * | 2019-11-29 | 2020-02-11 | 东北农业大学 | Milk cow rumination behavior identification method based on SSD convolutional neural network |
CN110839366A (en) * | 2019-10-21 | 2020-02-28 | 中国科学院东北地理与农业生态研究所 | Soybean plant seed tester and phenotype data acquisition and identification method |
CN111126402A (en) * | 2019-11-04 | 2020-05-08 | 北京海益同展信息科技有限公司 | Image processing method and device, electronic equipment and storage medium |
CN111597868A (en) * | 2020-01-08 | 2020-08-28 | 浙江大学 | SSD-based substation disconnecting switch state analysis method |
CN111652012A (en) * | 2020-05-11 | 2020-09-11 | 中山大学 | Curved surface QR code positioning method based on SSD network model |
CN111833310A (en) * | 2020-06-17 | 2020-10-27 | 桂林理工大学 | Surface defect classification method based on neural network architecture search |
CN112232263A (en) * | 2020-10-28 | 2021-01-15 | 中国计量大学 | Tomato identification method based on deep learning |
CN117975172A (en) * | 2024-03-29 | 2024-05-03 | 安徽农业大学 | Method and system for constructing and training whole pod recognition model |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2012123626A (en) * | 2010-12-08 | 2012-06-28 | Toyota Central R&D Labs Inc | Object detector and program |
US20170084067A1 (en) * | 2015-09-23 | 2017-03-23 | Samsung Electronics Co., Ltd. | Electronic device for processing image and method for controlling thereof |
CN107315999A (en) * | 2017-06-01 | 2017-11-03 | 范衠 | A kind of tobacco plant recognition methods based on depth convolutional neural networks |
CN107578050A (en) * | 2017-09-13 | 2018-01-12 | 浙江理工大学 | The automatic classifying identification method of rice basal part of stem On Planthopperss and its worm state |
CN108133186A (en) * | 2017-12-21 | 2018-06-08 | 东北林业大学 | A kind of plant leaf identification method based on deep learning |
CN108288075A (en) * | 2018-02-02 | 2018-07-17 | 沈阳工业大学 | A kind of lightweight small target detecting method improving SSD |
CN108564065A (en) * | 2018-04-28 | 2018-09-21 | 广东电网有限责任公司 | A kind of cable tunnel open fire recognition methods based on SSD |
CN108592799A (en) * | 2018-05-02 | 2018-09-28 | 东北农业大学 | A kind of soybean kernel and beanpod image collecting device |
CN108647652A (en) * | 2018-05-14 | 2018-10-12 | 北京工业大学 | A kind of cotton development stage automatic identifying method based on image classification and target detection |
-
2018
- 2018-12-17 CN CN201811540821.9A patent/CN109684967A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2012123626A (en) * | 2010-12-08 | 2012-06-28 | Toyota Central R&D Labs Inc | Object detector and program |
US20170084067A1 (en) * | 2015-09-23 | 2017-03-23 | Samsung Electronics Co., Ltd. | Electronic device for processing image and method for controlling thereof |
CN107315999A (en) * | 2017-06-01 | 2017-11-03 | 范衠 | A kind of tobacco plant recognition methods based on depth convolutional neural networks |
CN107578050A (en) * | 2017-09-13 | 2018-01-12 | 浙江理工大学 | The automatic classifying identification method of rice basal part of stem On Planthopperss and its worm state |
CN108133186A (en) * | 2017-12-21 | 2018-06-08 | 东北林业大学 | A kind of plant leaf identification method based on deep learning |
CN108288075A (en) * | 2018-02-02 | 2018-07-17 | 沈阳工业大学 | A kind of lightweight small target detecting method improving SSD |
CN108564065A (en) * | 2018-04-28 | 2018-09-21 | 广东电网有限责任公司 | A kind of cable tunnel open fire recognition methods based on SSD |
CN108592799A (en) * | 2018-05-02 | 2018-09-28 | 东北农业大学 | A kind of soybean kernel and beanpod image collecting device |
CN108647652A (en) * | 2018-05-14 | 2018-10-12 | 北京工业大学 | A kind of cotton development stage automatic identifying method based on image classification and target detection |
Non-Patent Citations (4)
Title |
---|
W.D.HANSON: "Modified Seed Maturation Rates and Seed Yield Potentials in Soybean", 《CROP PHYSIOLOGY & METABOLISM》 * |
张晗: "苹果果实生长信息远程测定方法与技术研究", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》 * |
赵庆北: "改进的SSD的目标检测研究", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》 * |
高艳霞等: "基于粒子群算法和神经网络的大豆识别研究", 《信息自动化》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110100774A (en) * | 2019-05-08 | 2019-08-09 | 安徽大学 | River crab male and female recognition methods based on convolutional neural networks |
CN110110702A (en) * | 2019-05-20 | 2019-08-09 | 哈尔滨理工大学 | It is a kind of that algorithm is evaded based on the unmanned plane for improving ssd target detection network |
CN110443778B (en) * | 2019-06-25 | 2021-10-15 | 浙江工业大学 | Method for detecting irregular defects of industrial products |
CN110443778A (en) * | 2019-06-25 | 2019-11-12 | 浙江工业大学 | A method of detection industrial goods random defect |
CN110602411A (en) * | 2019-08-07 | 2019-12-20 | 深圳市华付信息技术有限公司 | Method for improving quality of face image in backlight environment |
CN110839366A (en) * | 2019-10-21 | 2020-02-28 | 中国科学院东北地理与农业生态研究所 | Soybean plant seed tester and phenotype data acquisition and identification method |
CN110839366B (en) * | 2019-10-21 | 2024-07-09 | 中国科学院东北地理与农业生态研究所 | Soybean plant seed tester and phenotype data acquisition and identification method |
CN111126402A (en) * | 2019-11-04 | 2020-05-08 | 北京海益同展信息科技有限公司 | Image processing method and device, electronic equipment and storage medium |
CN111126402B (en) * | 2019-11-04 | 2023-11-03 | 京东科技信息技术有限公司 | Image processing method and device, electronic equipment and storage medium |
CN110781870A (en) * | 2019-11-29 | 2020-02-11 | 东北农业大学 | Milk cow rumination behavior identification method based on SSD convolutional neural network |
CN111597868A (en) * | 2020-01-08 | 2020-08-28 | 浙江大学 | SSD-based substation disconnecting switch state analysis method |
CN111652012A (en) * | 2020-05-11 | 2020-09-11 | 中山大学 | Curved surface QR code positioning method based on SSD network model |
CN111833310B (en) * | 2020-06-17 | 2022-05-06 | 桂林理工大学 | Surface defect classification method based on neural network architecture search |
CN111833310A (en) * | 2020-06-17 | 2020-10-27 | 桂林理工大学 | Surface defect classification method based on neural network architecture search |
CN112232263A (en) * | 2020-10-28 | 2021-01-15 | 中国计量大学 | Tomato identification method based on deep learning |
CN112232263B (en) * | 2020-10-28 | 2024-03-19 | 中国计量大学 | Tomato identification method based on deep learning |
CN117975172A (en) * | 2024-03-29 | 2024-05-03 | 安徽农业大学 | Method and system for constructing and training whole pod recognition model |
CN117975172B (en) * | 2024-03-29 | 2024-07-09 | 安徽农业大学 | Method and system for constructing and training whole pod recognition model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109684967A (en) | A kind of soybean plant strain stem pod recognition methods based on SSD convolutional network | |
CN109800736B (en) | Road extraction method based on remote sensing image and deep learning | |
CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
CN107392091B (en) | Agricultural artificial intelligence crop detection method, mobile terminal and computer readable medium | |
CN107016405B (en) | A kind of pest image classification method based on classification prediction convolutional neural networks | |
CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
CN110717903A (en) | Method for detecting crop diseases by using computer vision technology | |
CN104103033B (en) | View synthesis method | |
CN107844797A (en) | A kind of method of the milking sow posture automatic identification based on depth image | |
CN109886153A (en) | A kind of real-time face detection method based on depth convolutional neural networks | |
CN109948501A (en) | The detection method of personnel and safety cap in a kind of monitor video | |
CN109712128A (en) | Feature point detecting method, device, computer equipment and storage medium | |
CN107871316A (en) | A kind of X-ray hand bone interest region extraction method based on deep neural network | |
CN108932712A (en) | A kind of rotor windings quality detecting system and method | |
CN109377441A (en) | Tongue with privacy protection function is as acquisition method and system | |
CN107256398A (en) | The milk cow individual discrimination method of feature based fusion | |
Hu et al. | A rapid, low-cost deep learning system to classify squid species and evaluate freshness based on digital images | |
Bondre et al. | Review on leaf diseases detection using deep learning | |
Xu et al. | Improved residual network for automatic classification grading of lettuce freshness | |
Brar et al. | A smart approach to coconut leaf spot disease classification using computer vision and deep learning technique | |
CN117423134A (en) | Human body target detection and analysis multitasking cooperative network and training method thereof | |
CN116778482A (en) | Embryo image blastomere target detection method, computer equipment and storage medium | |
CN106650824A (en) | Moving object classification method based on support vector machine | |
CN115862003A (en) | Lightweight YOLOv 5-based in-vivo apple target detection and classification method | |
Hui et al. | Rice Disease Identification and Classification Based on MATLAB and Android Platform |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190426 |