CN109614973A - Rice seedling and Weeds at seedling image, semantic dividing method, system, equipment and medium - Google Patents
Rice seedling and Weeds at seedling image, semantic dividing method, system, equipment and medium Download PDFInfo
- Publication number
- CN109614973A CN109614973A CN201811395683.XA CN201811395683A CN109614973A CN 109614973 A CN109614973 A CN 109614973A CN 201811395683 A CN201811395683 A CN 201811395683A CN 109614973 A CN109614973 A CN 109614973A
- Authority
- CN
- China
- Prior art keywords
- seedling
- image
- weeds
- rice
- rice seedling
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of rice seedling and Weeds at seedling image, semantic dividing method, system, equipment and media, which comprises obtains the color catalog image of rice seedling and Weeds at seedling；Generate exemplar image corresponding with color catalog image；Color catalog image and its corresponding exemplar image are divided into training sample and test sample；Pretreatment and data amplification are carried out to all samples, form training dataset and test data set；Construct rice seedling and Weeds at seedling image, semantic parted pattern based on full convolutional neural networks；Pixel using rice seedling and the color image of Weeds at seedling image, semantic parted pattern rice seedling to be split and Weeds at seedling is classified, and rice seedling and Weeds at seedling segmented image are exported, and realizes the semantic segmentation of rice seedling and Weeds at seedling image.The present invention can learn from sample and extract to obtain the feature of strong robustness, realize the semantic segmentation of rice seedling and Weeds at seedling image.
Description
Technical field
The present invention relates to a kind of rice seedling and Weeds at seedling image, semantic dividing method, especially a kind of rice seedling and
Weeds at seedling image, semantic dividing method, system, computer equipment and storage medium, belong to image procossing and depth learning technology
Field.
Background technique
Precisely spraying for pesticide can be under the premise of not influencing weeds control effect, and effectively save 40~60% pesticide is used
Amount.Rice seedling and Weeds at seedling identification are the foundations that herbicide sprays object and drug variety selection, are that weeds in paddy field is accurate
Therefore how the basis of prevention and control management quick and precisely carries out rice seedling and Weeds at seedling automatic identification is of great significance.
Due to often relying on hand-designed feature in previous rice seedling and Weeds at seedling identification process, need to enrich
Professional knowledge and devote a tremendous amount of time.The quality of feature largely will also rely on experience and fortune, often whole
The test and adjusting work of a algorithm all concentrate on this, need to have been manually done, cause high effort for.In contrast, in recent years by wide
An important insight in the deep learning theory of general concern is exactly first that the Feature Descriptor of hand design is calculated as vision
Step, often prematurely loses useful information, and study is to character representation relevant to task directly from image, than setting by hand
It is more efficient to count feature.
It is disclosed application No. is 201710102806.5 Chinese invention patent application and a kind of network is stacked based on depth
Weed images recognition methods.This includes: to be collected to training image based on the weed images recognition methods that depth stacks network
And pretreatment；It constructs and depth is trained to stack network model；Depth of the test sample input after training is stacked into network mould
Type carries out the automatic identification of weed images.But depth stacks the image data that network architecture is directed to large sample, has instruction
Practice the disadvantages of time is long and robustness is not strong.
Summary of the invention
The first purpose of this invention is the defect in order to solve the above-mentioned prior art, provides a kind of rice seedling and seedling
Phase weed images semantic segmentation method, this method can learn from sample and extract to obtain the feature of strong robustness, realize water
The semantic segmentation of rice sprouts and Weeds at seedling image.
Second object of the present invention is to provide a kind of rice seedling and Weeds at seedling image, semantic segmenting system.
Third object of the present invention is to provide a kind of computer equipment.
Fourth object of the present invention is to provide a kind of storage medium.
The first purpose of this invention can be reached by adopting the following technical scheme that:
A kind of rice seedling and Weeds at seedling image, semantic dividing method, which comprises
Obtain the color catalog image of rice seedling and Weeds at seedling；
Generate exemplar image corresponding with color catalog image；Wherein, the exemplar image has rice seedling
Seedling, Weeds at seedling and the corresponding pixel type label of background；
Color catalog image and its corresponding exemplar image are divided into training sample and test sample；
Pretreatment and data amplification are carried out to all samples, form training dataset and test data set；
By the full convolutional neural networks of training dataset training, rice seedling and seedling based on full convolutional neural networks are constructed
Phase weed images semantic segmentation model；
Using full convolutional neural networks training pattern to the pixel of the color image of rice seedling to be split and Weeds at seedling
Classify, export rice seedling and Weeds at seedling segmented image, realizes the semantic segmentation of rice seedling and Weeds at seedling image.
Further, described pair of all samples carry out pretreatment and data amplification, form training dataset and test data
Collection, specifically includes:
By the rescaling of color catalog image and its corresponding exemplar image to preset value；
By image [- 180 °, 180 °] interior Random-Rotation, along X, Y-axis [- 10 °, 10 °] interior random offset method
The color catalog image and its corresponding exemplar image are expanded；
At random using 80% color catalog image and its corresponding exemplar image as training dataset, residue 20%
Color catalog image and its corresponding exemplar image pattern as test data set.
Further, described pair of all samples carry out pretreatment and data amplification, form training dataset and test data
Collection, further includes:
By calculating rice seedling, Weeds at seedling and background pixel sum, the method that punishment weight is added makes picture
Plain type reaches balance.
Further, the rice seedling and Weeds at seedling image, semantic parted pattern include the coding based on SegNet
Device-decoder architecture and pixel classifications network；
The building process of the coder-decoder structure based on SegNet are as follows: encoder is by 13 layers before VGG16
Convolutional network is constituted, wherein there is multiple coding units, extracts feature by convolution, keeps the corresponding decoding of each coding unit single
Member, the decoder that multiple decoding units are constituted use deconvolution and up-sampling, to be filled in the content lacked during pondization；
The building process of the pixel classifications network are as follows: last in the coder-decoder structure based on SegNet
One convolutional layer output rice seedling, Weeds at seedling and the corresponding pixel type of background, the coder-decoder based on SegNet
It is that each pixel generates class probability that structure, which finally adds a softmax classifier layer, completes the other classification of image pixel-class.
Further, the coder-decoder structure includes:
Each coding unit includes at least two convolution units and a pond layer；Wherein, each convolution unit includes volume
Lamination, batch normalization layer and Relu layers, the characteristic pattern that the pond layer is used to that convolutional layer to be made to export is reduced into original 1/2；
Increase an index function in the layer of pond, opposite position of the weight selected with the maximum pond of preservation in filter
It sets；Up-sampling is the inverse process in pond in a decoder, by index function image is become larger 2 times in up-sampling.
Second object of the present invention can be reached by adopting the following technical scheme that:
A kind of rice seedling and Weeds at seedling image, semantic segmenting system, the system comprises:
Module is obtained, for obtaining the color catalog image of rice seedling and Weeds at seedling；
Generation module, for generating exemplar image corresponding with color catalog image；Wherein, the exemplar figure
As having rice seedling, Weeds at seedling and the corresponding pixel type label of background；
Sample division module, for color catalog image and its corresponding exemplar image to be divided into training sample and survey
Sample sheet；
Data set forms module, for carrying out pretreatment and data amplification to all samples, forms training dataset and survey
Try data set；
Module is constructed, for by the full convolutional neural networks of training dataset training, building to be based on full convolutional neural networks
Rice seedling and Weeds at seedling image, semantic parted pattern；
Semantic segmentation module, for utilizing full convolutional neural networks training pattern to rice seedling to be split and Weeds at seedling
The pixel of color image classify, export rice seedling and Weeds at seedling segmented image, realize that rice seedling and seedling stage are miscellaneous
The semantic segmentation of sketch picture.
Further, the data set forms module, specifically includes:
Adjustment unit, for by the rescaling of color catalog image and its corresponding exemplar image to preset value；
Amplification unit, for by image [- 180 °, 180 °] interior Random-Rotation, along X, Y-axis in [- 10 °, 10 °] with
The method of machine offset expands the color catalog image and its corresponding exemplar image；
It is randomly formed unit, at random using 80% color catalog image and its corresponding exemplar image as instruction
Practice data set, the color catalog image and its corresponding exemplar image pattern of residue 20% are as test data set.
Further, the data set forms module, further includes:
Computing unit, for power of punishment to be added by calculating rice seedling, Weeds at seedling and background pixel sum
The method of weight makes pixel type reach balance.
Further, in the building module, rice seedling and Weeds at seedling image, semantic parted pattern include being based on
The coder-decoder structure and pixel classifications network of SegNet；
The building process of the coder-decoder structure based on SegNet are as follows: encoder is by 13 layers before VGG16
Convolutional network is constituted, wherein there is multiple coding units, extracts feature by convolution, keeps the corresponding decoding of each coding unit single
Member, the decoder that multiple decoding units are constituted use deconvolution and up-sampling, to be filled in the content lacked during pondization；
The building process of the pixel classifications network are as follows: last in the coder-decoder structure based on SegNet
One convolutional layer output rice seedling, Weeds at seedling and the corresponding pixel type of background, the coder-decoder based on SegNet
It is that each pixel generates class probability that structure, which finally adds a softmax classifier layer, completes the other classification of image pixel-class.
Further, the coder-decoder structure includes:
Each coding unit includes at least two convolution units and a pond layer；Wherein, each convolution unit includes volume
Lamination, batch normalization layer and Relu layers, the characteristic pattern that the pond layer is used to that convolutional layer to be made to export is reduced into original 1/2；
Increase an index function in the layer of pond, opposite position of the weight selected with the maximum pond of preservation in filter
It sets；Up-sampling is the inverse process in pond in a decoder, by index function image is become larger 2 times in up-sampling.
Third object of the present invention can be reached by adopting the following technical scheme that:
A kind of computer equipment, including processor and for the memory of storage processor executable program, the place
When managing the program of device execution memory storage, above-mentioned rice seedling and Weeds at seedling image, semantic dividing method are realized.
Fourth object of the present invention can be reached by adopting the following technical scheme that:
A kind of storage medium is stored with program, when described program is executed by processor, realizes above-mentioned rice seedling and seedling
Phase weed images semantic segmentation method.
The present invention have compared with the existing technology it is following the utility model has the advantages that
The present invention is by acquisition rice seedling and the color catalog image of Weeds at seedling, and generates and color catalog image
Corresponding exemplar image, obtains training dataset, and by the full convolutional neural networks of training dataset training, building is based on complete
The rice seedling and Weeds at seedling image, semantic parted pattern of convolutional neural networks, by test, the rice seedling and seedling stage are miscellaneous
Sketch can be used in the semantic segmentation of rice seedling and Weeds at seedling image as semantic segmentation model, ensure that the accurate journey of segmentation
Degree and integrality, and segmentation effect is good.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
The structure shown according to these attached drawings obtains other attached drawings.
Fig. 1 is the flow chart of the rice seedling and Weeds at seedling image, semantic dividing method of the embodiment of the present invention 1.
Fig. 2 is the rice seedling of the embodiment of the present invention 1 and the sample image of Weeds at seedling.
Fig. 3 is the rice seedling and the corresponding exemplar image of Weeds at seedling sample image of the embodiment of the present invention 1.
Fig. 4 is the rice seedling and Weeds at seedling image, semantic parted pattern training schematic diagram of the embodiment of the present invention 1.
Fig. 5 is the rice seedling and Weeds at seedling image, semantic parted pattern structure chart of the embodiment of the present invention 1.
Fig. 6 is that the rice seedling of the embodiment of the present invention 1 and Weeds at seedling image, semantic divide schematic diagram.
Fig. 7 is the rice seedling of the embodiment of the present invention 2 and the structural block diagram of Weeds at seedling image, semantic segmenting system.
Fig. 8 is that the data set of the embodiment of the present invention 2 forms the structural block diagram of module.
Fig. 9 is the structural block diagram of the computer equipment of the embodiment of the present invention 3.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments, based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Embodiment 1:
As shown in Figure 1, present embodiments providing a kind of rice seedling and Weeds at seedling image, semantic dividing method, this method
The following steps are included:
S1, the color catalog image for obtaining rice seedling and Weeds at seedling.
The present embodiment first obtains the color catalog image (i.e. RGB sample image) of rice seedling and Weeds at seedling, can be with
It is obtained by acquisition, such as shoots the color catalog image of rice seedling and Weeds at seedling by camera under field conditions (factors),
It can also be obtained from database lookup, such as in advance in databases water storage rice sprouts and the color catalog figure of Weeds at seedling
Picture, rice seedling and the color catalog image of Weeds at seedling are searched for from database can be obtained.
S2, generation exemplar image corresponding with color catalog image；Wherein, the exemplar image has rice
The label of rice shoot, Weeds at seedling and the corresponding pixel type of background.
After the color catalog image for obtaining rice seedling and Weeds at seedling, can by being manually labeled, specifically:
User inputs annotation command, is then responding to the annotation command, carries out water to the color catalog image of rice seedling and Weeds at seedling
The mark of rice sprouts, Weeds at seedling and the corresponding pixel type of background, to generate label sample corresponding with color catalog image
This image, i.e. the exemplar image have rice seedling, Weeds at seedling and the corresponding pixel type label of background.
Fig. 2 shows rice seedling and the sample image of Weeds at seedling, Fig. 3 shows corresponding exemplar image.
S3, color catalog image and its corresponding exemplar image are divided into training sample and test sample.
S4, all samples are carried out with pretreatment and data amplification, forms training dataset and test data set.
Specifically, step S4 includes:
S401, by the rescaling of color catalog image and its corresponding exemplar image to preset value.
In the present embodiment, preset value is 256 × 228, color catalog image and its corresponding exemplar image just
Beginning scale is 1024 × 912, and in order to facilitate subsequent processing, scale is transformed to 256 × 228 by 1024 × 912.
S402, by image [- 180 °, 180 °] interior Random-Rotation, along X, Y-axis in [- 10 °, 10 °] interior random offset
Method expands the color catalog image and its corresponding exemplar image；
S403, at random using 80% color catalog image and its corresponding exemplar image as training dataset, remain
Remaining 20% color catalog image and its corresponding exemplar image pattern as test data set.
In order to solve the problems, such as that pixel type is unbalanced, the step S4 of the present embodiment further include:
S404, by calculating rice seedling, Weeds at seedling and background pixel sum, method that punishment weight is added
Pixel type is set to reach balance, shown in punishment weight coefficient calculating process such as formula (1)~(3)；
Wherein, w0Weight, w for rice seedling pixel1For the weight and w of Weeds at seedling pixel2For the power of background pixel
Weight, Ni0For the number of rice seedling pixel in i-th image, Ni1For the number of Weeds at seedling pixel in i-th image, Ni2For
The number of background pixel, N in i-th imageiFor the summation of all pixels in i-th image.Experiment calculation result such as 1 institute of table
Show.
1 image pattern amount of pixels of table and punishment weight coefficient
S5, pass through the full convolutional neural networks of training dataset training, rice seedling of the building based on full convolutional neural networks
And Weeds at seedling image, semantic parted pattern.
By the full convolutional neural networks of training dataset training as shown in figure 4, rice seedling and Weeds at seedling image, semantic
Parted pattern includes coder-decoder structure and pixel classifications network based on SegNet, as shown in Figure 5.
The building process of coder-decoder structure based on SegNet are as follows: encoder is by ten three-layer coils product before VGG16
Network is constituted, wherein there are five coding units, first coding unit includes the first layer and second layer convolutional network of VGG16,
Second coding unit includes the third layer of VGG16 and the 4th layer of convolutional network, third coding unit include the 5th of VGG16
Layer, layer 6 and layer 7 convolutional network, the 4th coding unit include the 8th layer, the 9th layer and the tenth layer convolution of VGG16
Network, the 5th coding unit include eleventh floor, Floor 12 and the tenth three-layer coil the product network of VGG16, are mentioned by convolution
Feature is taken, makes the corresponding decoding unit of each coding unit, the decoder that five decoding units are constituted uses deconvolution and upper
Sampling, to be filled in the content lacked during pondization.
The building process of pixel classifications network are as follows: the last one in the coder-decoder structure based on SegNet
Convolutional layer exports rice seedling, Weeds at seedling and the corresponding pixel type of background, the coder-decoder structure based on SegNet
It is finally that each pixel generates class probability plus a softmax classifier layer, completes the other classification of image pixel-class.
Further, the coder-decoder structure includes:
Each coding unit includes that (i.e. pooling layers, 3 × 3 windows walk at least two convolution units and pond layer
Into 2 and maximum pond), wherein all there are two convolution units for first coding unit and second coding unit, and third is compiled
All there are three convolution units for code unit, the 4th coding unit and the 5th coding unit；Wherein, each convolution unit includes volume
Lamination, batch normalization layer and Relu (Rectified linear unit corrects linear unit) layer, the specific function of pond layer
It can be that the characteristic pattern for exporting convolutional layer is reduced into original 1/2, design parameter is as shown in table 2 below.
2 SegNet-VGG16 decoder internal structural parameters table of table
Increase index (index) function in the layer of pond, is filtered with saving the weight that maximum pond is selected 3 × 3
Relative position in device；Up-sampling is the inverse process in pond in a decoder, makes image in up-sampling by index function
Become larger 2 times, design parameter is as shown in table 3 below.
3 SegNet-VGG16 decoder internal structure parameter list of table
As described above, using 20% color catalog image and its corresponding exemplar image pattern as test data
Collection, tests rice seedling and Weeds at seedling image, semantic parted pattern, pixel accuracy rate PA (Pixel Accuracy)
The ratio that the correct total pixel of pixel Zhan is marked for semantic analogy in image, as shown in formula (4):
IoU (Intersection over Union, equal direct ratio) is the gauge of semantic segmentation, and what is calculated is picture
The label true value (ground truth) of element and the union of pixel predictors (predicted segmentation), the IoU
Value is calculated by each classification, is then averaged, as shown in formula (5):
Wherein, k indicates classification, total k+1 classification (comprising background), k=2 in this experiment.I indicates true classification, and j is indicated
Predict classification.piiIt indicates really, is true classification pixel quantity identical with prediction category result, and pijIt indicates vacation just, is i
Classification is the pixel quantity of j classification, p by false judgmentjiThen it is expressed as false negative, i.e. piiIt is judicious to represent semantic classes
Pixel quantity, pijAnd pjiRepresent the pixel quantity of semantic classes misjudgment.
F value is the balance index for comprehensively considering recall rate and precision, takes into account the precise degrees and integrality of segmentation, and F value is got over
Height, segmentation effect is better, as shown in formula (6):
Wherein, A is the pixel class set divided by full convolution partitioning algorithm, including background pixel (v=0), water
Rice sprouts pixel (v=1) and Weeds at seedling pixel (v=2), B are the true tag set of respective pixel collection, including background pixel
(v=0), rice seedling pixel (v=1) and Weeds at seedling pixel (v=2).I, j are pixel index, and m is that image is high, and n is image
Width, vi,jFor the gray value of the i-th column jth row pixel.Test result is as follows shown in table 4, it is seen that the rice seedling and seedling of the present embodiment
Phase weed images semantic segmentation model can be used in the semantic segmentation of rice seedling and Weeds at seedling image.
4 SegNet-VGG16 model test results of table
| Pixel accuracy rate | IoU | F value | |
| Rice seedling | 0.94792 | 0.51843 | 0.6972 |
| Weeds at seedling | 0.90606 | 0.53446 | 0.64007 |
| Background | 0.89914 | 0.89586 | 0.78987 |
S6, using rice seedling and Weeds at seedling image, semantic parted pattern to rice seedling to be split and Weeds at seedling
The pixel of color image is classified, and rice seedling and Weeds at seedling segmented image are exported, and realizes rice seedling and Weeds at seedling
The semantic segmentation of image.
It will be understood by those skilled in the art that journey can be passed through by implementing the method for the above embodiments
Sequence is completed to instruct relevant hardware, and corresponding program can store in computer readable storage medium.
It should be noted that this is not although describing the method operation of above-described embodiment in the accompanying drawings with particular order
It is required that hint must execute these operations in this particular order, could be real or have to carry out shown in whole operation
Existing desired result.On the contrary, the step of describing can change and execute sequence.Additionally or alternatively, it is convenient to omit certain steps,
Multiple steps are merged into a step to execute, and/or a step is decomposed into execution of multiple steps.
Embodiment 2:
As shown in fig. 7, present embodiments providing a kind of rice seedling and Weeds at seedling image, semantic segmenting system, the system
Module, building module and semantic segmentation module are formed including obtaining module, generation module, sample division module, data set, it is each
The concrete function of module is as follows:
The acquisition module, for obtaining the color catalog image of rice seedling and Weeds at seedling.
The generation module, for generating exemplar image corresponding with color catalog image；Wherein, the label sample
Label of this image with rice seedling, Weeds at seedling and the corresponding pixel type of background.
The sample division module, for color catalog image and its corresponding exemplar image to be divided into training sample
And test sample.
The data set forms module, for carrying out pretreatment and data amplification to all samples, forms training dataset
And test data set；The data set forms module as shown in figure 8, specifically including:
Adjustment unit, for by the rescaling of color catalog image and its corresponding exemplar image to preset value.
Amplification unit, for by image [- 180 °, 180 °] interior Random-Rotation, along X, Y-axis in [- 10 °, 10 °] with
The method of machine offset expands the color catalog image and its corresponding exemplar image.
It is randomly formed unit, at random using 80% color catalog image and its corresponding exemplar image as instruction
Practice data set, the color catalog image and its corresponding exemplar image pattern of residue 20% are as test data set.
Computing unit, for power of punishment to be added by calculating rice seedling, Weeds at seedling and background pixel sum
The method of weight makes pixel type reach balance.
The building module, for by the full convolutional neural networks of training dataset training, building to be based on full convolutional Neural
The rice seedling and Weeds at seedling image, semantic parted pattern of network；Wherein, rice seedling and the segmentation of Weeds at seedling image, semantic
Model includes coder-decoder structure and pixel classifications network based on SegNet；
The building process of the coder-decoder structure based on SegNet are as follows: encoder is by 13 layers before VGG16
Convolutional network is constituted, wherein there is multiple coding units, extracts feature by convolution, keeps the corresponding decoding of each coding unit single
Member, the decoder that multiple decoding units are constituted use deconvolution and up-sampling, to be filled in the content lacked during pondization；
The building process of the pixel classifications network are as follows: last in the coder-decoder structure based on SegNet
One convolutional layer output rice seedling, Weeds at seedling and the corresponding pixel type of background, the coder-decoder based on SegNet
It is that each pixel generates class probability that structure, which finally adds a softmax classifier layer, completes the other classification of image pixel-class.
Further, the coder-decoder structure includes:
Each coding unit includes at least two convolution units and a pond layer；Wherein, each convolution unit includes volume
Lamination, batch normalization layer and Relu layers, the characteristic pattern that the pond layer is used to that convolutional layer to be made to export is reduced into original 1/2；
Increase an index function in the layer of pond, opposite position of the weight selected with the maximum pond of preservation in filter
It sets；Up-sampling is the inverse process in pond in a decoder, by index function image is become larger 2 times in up-sampling.
The semantic segmentation module, for utilizing rice seedling and Weeds at seedling image, semantic parted pattern to water to be split
The pixel of the color image of rice sprouts and Weeds at seedling is classified, and rice seedling and Weeds at seedling segmented image are exported, and is realized
The semantic segmentation of rice seedling and Weeds at seedling image.
It should be noted that system provided by the above embodiment is only illustrated with the division of above-mentioned each functional module
Illustrate, in practical applications, can according to need and be completed by different functional modules above-mentioned function distribution, i.e., by internal junction
Structure is divided into different functional modules, to complete all or part of the functions described above.
Embodiment 3:
A kind of computer equipment is present embodiments provided, which can be computer, as shown in figure 9, including
Processor, memory, input unit, display and the network interface connected by system bus.Wherein, processor is for providing
It calculates and control ability, memory includes non-volatile memory medium and built-in storage, which is stored with
Operating system, computer program and database, the built-in storage are operating system and computer in non-volatile memory medium
The operation of program provides environment, when computer program is executed by processor, realizes that the rice seedling of above-described embodiment 1 and seedling stage are miscellaneous
Sketch is as follows as semantic segmentation method:
Obtain the color catalog image of rice seedling and Weeds at seedling；
Generate exemplar image corresponding with color catalog image；Wherein, the exemplar image has rice seedling
Seedling, Weeds at seedling and the corresponding pixel type label of background；
Color catalog image and its corresponding exemplar image are divided into training sample and test sample；
Pretreatment and data amplification are carried out to all samples, form training dataset and test data set；
By the full convolutional neural networks of training dataset training, rice seedling and seedling based on full convolutional neural networks are constructed
Phase weed images semantic segmentation model；
Using rice seedling and Weeds at seedling image, semantic parted pattern to the coloured silk of rice seedling to be split and Weeds at seedling
The pixel of chromatic graph picture is classified, and rice seedling and Weeds at seedling segmented image are exported, and realizes rice seedling and Weeds at seedling figure
The semantic segmentation of picture.
It is appreciated that the computer equipment of the present embodiment can also be server, mobile terminal etc..
Embodiment 4:
A kind of storage medium is present embodiments provided, which is stored with one or more programs, described program quilt
When processor executes, the rice seedling and Weeds at seedling image, semantic dividing method of above-described embodiment 1 are realized, as follows:
Obtain the color catalog image of rice seedling and Weeds at seedling；
Generate exemplar image corresponding with color catalog image；Wherein, the exemplar image has rice seedling
Seedling, Weeds at seedling and the corresponding pixel type label of background；
Color catalog image and its corresponding exemplar image are divided into training sample and test sample；
Pretreatment and data amplification are carried out to all samples, form training dataset and test data set；
By the full convolutional neural networks of training dataset training, rice seedling and seedling based on full convolutional neural networks are constructed
Phase weed images semantic segmentation model；
Using rice seedling and Weeds at seedling image, semantic parted pattern to the coloured silk of rice seedling to be split and Weeds at seedling
The pixel of chromatic graph picture is classified, and rice seedling and Weeds at seedling segmented image are exported, and realizes rice seedling and Weeds at seedling figure
The semantic segmentation of picture.
The storage medium of the present embodiment can be disk, CD, computer storage, read-only memory (ROM, Read-
Only Memory), random access memory (RAM, Random Access Memory), USB flash disk, the media such as mobile hard disk.
In conclusion the present invention passes through acquisition rice seedling and the color catalog image of Weeds at seedling, and generation and coloured silk
The corresponding exemplar image of colo(u)r atlas image, obtains training dataset, trains full convolutional neural networks by training dataset,
Rice seedling and Weeds at seedling image, semantic parted pattern based on full convolutional neural networks are constructed, by test, the rice seedling
Seedling and Weeds at seedling image, semantic parted pattern can be used in the semantic segmentation of rice seedling and Weeds at seedling image, ensure that point
The precise degrees and integrality cut, and segmentation effect is good.
The above, only the invention patent preferred embodiment, but the scope of protection of the patent of the present invention is not limited to
This, anyone skilled in the art is in the range disclosed in the invention patent, according to the present invention the skill of patent
Art scheme and its inventive concept are subject to equivalent substitution or change, belong to the scope of protection of the patent of the present invention.
Claims (10)
1. a kind of rice seedling and Weeds at seedling image, semantic dividing method, which is characterized in that the described method includes:
Obtain the color catalog image of rice seedling and Weeds at seedling；
Generate exemplar image corresponding with color catalog image；Wherein, the exemplar image has rice seedling, seedling
Phase weeds and the corresponding pixel type label of background；
Color catalog image and its corresponding exemplar image are divided into training sample and test sample；
Pretreatment and data amplification are carried out to all samples, form training dataset and test data set；
By the full convolutional neural networks of training dataset training, rice seedling and seedling stage of the building based on full convolutional neural networks are miscellaneous
Sketch is as semantic segmentation model；
Using rice seedling and Weeds at seedling image, semantic parted pattern to the cromogram of rice seedling to be split and Weeds at seedling
The pixel of picture is classified, and rice seedling and Weeds at seedling segmented image are exported, and realizes rice seedling and Weeds at seedling image
Semantic segmentation.
2. rice seedling according to claim 1 and Weeds at seedling image, semantic dividing method, which is characterized in that described right
All samples carry out pretreatment and data amplification, form training dataset and test data set, specifically include:
By the rescaling of color catalog image and its corresponding exemplar image to preset value；
By image [- 180 °, 180 °] interior Random-Rotation, along X, Y-axis [- 10 °, 10 °] interior random offset method to institute
It states color catalog image and its corresponding exemplar image is expanded；
At random using 80% color catalog image and its corresponding exemplar image as training dataset, the coloured silk of residue 20%
Colo(u)r atlas image and its corresponding exemplar image pattern are as test data set.
3. rice seedling according to claim 2 and Weeds at seedling image, semantic dividing method, which is characterized in that described right
All samples carry out pretreatment and data amplification, form training dataset and test data set, further includes:
By calculating rice seedling, Weeds at seedling and background pixel sum, the method that punishment weight is added makes pixel kind
Class reaches balance.
4. rice seedling according to claim 1-3 and Weeds at seedling image, semantic dividing method, feature exist
In, the rice seedling and Weeds at seedling image, semantic parted pattern include coder-decoder structure based on SegNet and
Pixel classifications network；
The building process of the coder-decoder structure based on SegNet are as follows: encoder is by ten three-layer coils product before VGG16
Network is constituted, wherein there is multiple coding units, extracts feature by convolution, makes the corresponding decoding unit of each coding unit,
The decoder that multiple decoding units are constituted uses deconvolution and up-sampling, to be filled in the content lacked during pondization；
The building process of the pixel classifications network are as follows: the last one in the coder-decoder structure based on SegNet
Convolutional layer exports rice seedling, Weeds at seedling and the corresponding pixel type of background, the coder-decoder structure based on SegNet
It is finally that each pixel generates class probability plus a softmax classifier layer, completes the other classification of image pixel-class.
5. rice seedling according to claim 4 and Weeds at seedling image, semantic dividing method, which is characterized in that the volume
Code device-decoder architecture include:
Each coding unit includes at least two convolution units and a pond layer；Wherein, each convolution unit include convolutional layer,
Normalization layer and Relu layers are criticized, the characteristic pattern that the pond layer is used to that convolutional layer to be made to export is reduced into original 1/2；
Increase an index function in the layer of pond, relative position of the weight selected with the maximum pond of preservation in filter；
Up-sampling is the inverse process in pond in a decoder, by index function image is become larger 2 times in up-sampling.
6. a kind of rice seedling and Weeds at seedling image, semantic segmenting system, which is characterized in that the system comprises:
Module is obtained, for obtaining the color catalog image of rice seedling and Weeds at seedling；
Generation module, for generating exemplar image corresponding with color catalog image；Wherein, the exemplar picture strip
There are rice seedling, Weeds at seedling and the corresponding pixel type label of background；
Sample division module, for color catalog image and its corresponding exemplar image to be divided into training sample and test specimens
This；
Data set forms module, for carrying out pretreatment and data amplification to all samples, forms training dataset and test number
According to collection；
Module is constructed, for constructing the water based on full convolutional neural networks by the full convolutional neural networks of training dataset training
Rice sprouts and Weeds at seedling image, semantic parted pattern；
Semantic segmentation module pre-processes for the color image to rice seedling to be split and Weeds at seedling, utilizes rice
Rice shoot and Weeds at seedling image, semantic parted pattern carry out the pixel of the color image of rice seedling to be split and Weeds at seedling
Classification exports rice seedling and Weeds at seedling segmented image, realizes the semantic segmentation of rice seedling and Weeds at seedling image.
7. rice seedling according to claim 6 and Weeds at seedling image, semantic segmenting system, which is characterized in that the number
Module is formed according to collection, is specifically included:
Adjustment unit, for by the rescaling of color catalog image and its corresponding exemplar image to preset value；
Amplification unit, for by image [- 180 °, 180 °] interior Random-Rotation, along X, Y-axis in [- 10 °, 10 °] it is random partially
The method of shifting expands the color catalog image and its corresponding exemplar image；
It is randomly formed unit, at random using 80% color catalog image and its corresponding exemplar image as training number
According to collection, the color catalog image and its corresponding exemplar image pattern of residue 20% are as test data set.
8. rice seedling according to claim 7 and Weeds at seedling image, semantic segmenting system, which is characterized in that the number
Module is formed according to collection, further includes:
Computing unit, for punishment weight to be added by calculating rice seedling, Weeds at seedling and background pixel sum
Method makes pixel type reach balance.
9. a kind of computer equipment, including processor and for the memory of storage processor executable program, feature exists
In, when the processor executes the program of memory storage, the described in any item rice seedlings of realization claim 1-5 and seedling stage
Weed images semantic segmentation method.
10. a kind of storage medium, is stored with program, which is characterized in that when described program is executed by processor, realize claim
The described in any item rice seedlings of 1-5 and Weeds at seedling image, semantic dividing method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811395683.XA CN109614973A (en) | 2018-11-22 | 2018-11-22 | Rice seedling and Weeds at seedling image, semantic dividing method, system, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811395683.XA CN109614973A (en) | 2018-11-22 | 2018-11-22 | Rice seedling and Weeds at seedling image, semantic dividing method, system, equipment and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109614973A true CN109614973A (en) | 2019-04-12 |
Family
ID=66004836
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811395683.XA Pending CN109614973A (en) | 2018-11-22 | 2018-11-22 | Rice seedling and Weeds at seedling image, semantic dividing method, system, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109614973A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059772A (en) * | 2019-05-14 | 2019-07-26 | 温州大学 | Remote sensing images semantic segmentation method based on migration VGG network |
| CN110197491A (en) * | 2019-05-17 | 2019-09-03 | 上海联影智能医疗科技有限公司 | Image partition method, device, equipment and storage medium |
| CN110321905A (en) * | 2019-07-11 | 2019-10-11 | 广东工业大学 | Abnormal area detection method, system and associated component based on semantic segmentation |
| CN110969182A (en) * | 2019-05-17 | 2020-04-07 | 丰疆智能科技股份有限公司 | Convolutional neural network construction method and system based on farmland image |
| CN111368843A (en) * | 2020-03-06 | 2020-07-03 | 电子科技大学 | Method for extracting lake on ice based on semantic segmentation |
| CN111612797A (en) * | 2020-03-03 | 2020-09-01 | 江苏大学 | Rice image information processing system |
| CN111639575A (en) * | 2020-05-25 | 2020-09-08 | 广东石油化工学院 | Weeding tilling depth adjusting method, weeding tilling depth adjusting device, weeding tilling depth adjusting system, computer equipment and storage medium |
| CN111724371A (en) * | 2020-06-19 | 2020-09-29 | 联想(北京)有限公司 | Data processing method and device and electronic equipment |
| CN112116595A (en) * | 2020-10-27 | 2020-12-22 | 河北农业大学 | End-to-end automatic plant root system characteristic segmentation system |
| CN113280820A (en) * | 2021-06-09 | 2021-08-20 | 华南农业大学 | Orchard visual navigation path extraction method and system based on neural network |
| CN113610040A (en) * | 2021-08-16 | 2021-11-05 | 华南农业大学 | Paddy field weed density real-time statistical method based on improved BiSeNetV2 segmentation network |
| CN113610035A (en) * | 2021-08-16 | 2021-11-05 | 华南农业大学 | Rice tillering stage weed segmentation and identification method based on improved coding and decoding network |
| CN113597874A (en) * | 2021-09-29 | 2021-11-05 | 农业农村部南京农业机械化研究所 | Weeding robot and weeding path planning method, device and medium thereof |
| CN113807143A (en) * | 2020-06-12 | 2021-12-17 | 广州极飞科技股份有限公司 | Crop connected domain identification method and device and operation system |
| CN114913544A (en) * | 2022-04-28 | 2022-08-16 | 华南农业大学 | Shrimp larvae counting method and system based on semantic segmentation, cloud server and medium |
| CN117496353A (en) * | 2023-11-13 | 2024-02-02 | 安徽农业大学 | Rice seedling weed stem center distinguishing and positioning method based on two-stage segmentation model |
| US12141985B2 (en) | 2021-09-29 | 2024-11-12 | Nanjing Institute Of Agricultural Mechanization, Ministry Of Agriculture And Rural Affairs | Weeding robot and method and apparatus for planning weeding path thereof, and medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108564587A (en) * | 2018-03-07 | 2018-09-21 | 浙江大学 | A kind of a wide range of remote sensing image semantic segmentation method based on full convolutional neural networks |
| CN108647568A (en) * | 2018-03-30 | 2018-10-12 | 电子科技大学 | Grassland degeneration extraction method based on full convolutional neural networks |
-
2018
- 2018-11-22 CN CN201811395683.XA patent/CN109614973A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108564587A (en) * | 2018-03-07 | 2018-09-21 | 浙江大学 | A kind of a wide range of remote sensing image semantic segmentation method based on full convolutional neural networks |
| CN108647568A (en) * | 2018-03-30 | 2018-10-12 | 电子科技大学 | Grassland degeneration extraction method based on full convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| EVAN SHELHAMER, JONATHAN LONG, AND TREVOR DARRELL, MEMBER: "Fully Convolutional Networks for Semantic Segmentation", 《 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 轩永仓: "基于全卷积神经网络的大田复杂场景图像的语义分割研究", 《信息科技辑》 * |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059772A (en) * | 2019-05-14 | 2019-07-26 | 温州大学 | Remote sensing images semantic segmentation method based on migration VGG network |
| CN110059772B (en) * | 2019-05-14 | 2021-04-30 | 温州大学 | Remote sensing image semantic segmentation method based on multi-scale decoding network |
| CN110969182A (en) * | 2019-05-17 | 2020-04-07 | 丰疆智能科技股份有限公司 | Convolutional neural network construction method and system based on farmland image |
| CN110197491A (en) * | 2019-05-17 | 2019-09-03 | 上海联影智能医疗科技有限公司 | Image partition method, device, equipment and storage medium |
| CN110197491B (en) * | 2019-05-17 | 2021-08-17 | 上海联影智能医疗科技有限公司 | Image segmentation method, device, equipment and storage medium |
| CN110321905A (en) * | 2019-07-11 | 2019-10-11 | 广东工业大学 | Abnormal area detection method, system and associated component based on semantic segmentation |
| CN111612797A (en) * | 2020-03-03 | 2020-09-01 | 江苏大学 | Rice image information processing system |
| CN111612797B (en) * | 2020-03-03 | 2021-05-25 | 江苏大学 | Rice image information processing system |
| CN111368843A (en) * | 2020-03-06 | 2020-07-03 | 电子科技大学 | Method for extracting lake on ice based on semantic segmentation |
| CN111368843B (en) * | 2020-03-06 | 2022-06-10 | 电子科技大学 | Method for extracting lake on ice based on semantic segmentation |
| CN111639575A (en) * | 2020-05-25 | 2020-09-08 | 广东石油化工学院 | Weeding tilling depth adjusting method, weeding tilling depth adjusting device, weeding tilling depth adjusting system, computer equipment and storage medium |
| CN113807143A (en) * | 2020-06-12 | 2021-12-17 | 广州极飞科技股份有限公司 | Crop connected domain identification method and device and operation system |
| CN113807143B (en) * | 2020-06-12 | 2024-11-12 | 广州极飞科技股份有限公司 | Identification method, device and operation system of crop connected domain |
| CN111724371B (en) * | 2020-06-19 | 2023-05-23 | 联想(北京)有限公司 | Data processing method and device and electronic equipment |
| CN111724371A (en) * | 2020-06-19 | 2020-09-29 | 联想(北京)有限公司 | Data processing method and device and electronic equipment |
| CN112116595A (en) * | 2020-10-27 | 2020-12-22 | 河北农业大学 | End-to-end automatic plant root system characteristic segmentation system |
| CN113280820B (en) * | 2021-06-09 | 2022-11-29 | 华南农业大学 | Orchard visual navigation path extraction method and system based on neural network |
| CN113280820A (en) * | 2021-06-09 | 2021-08-20 | 华南农业大学 | Orchard visual navigation path extraction method and system based on neural network |
| CN113610035B (en) * | 2021-08-16 | 2023-10-10 | 华南农业大学 | Rice tillering stage weed segmentation and identification method based on improved coding and decoding network |
| CN113610040A (en) * | 2021-08-16 | 2021-11-05 | 华南农业大学 | Paddy field weed density real-time statistical method based on improved BiSeNetV2 segmentation network |
| CN113610040B (en) * | 2021-08-16 | 2023-06-27 | 华南农业大学 | Paddy field weed density real-time statistical method based on improved BiSeNetV2 segmentation network |
| CN113610035A (en) * | 2021-08-16 | 2021-11-05 | 华南农业大学 | Rice tillering stage weed segmentation and identification method based on improved coding and decoding network |
| WO2023050783A1 (en) * | 2021-09-29 | 2023-04-06 | 农业农村部南京农业机械化研究所 | Weeding robot and method and apparatus for planning weeding path thereof, and medium |
| AU2022256171B2 (en) * | 2021-09-29 | 2024-02-29 | Nanjing Institute Of Agricultural Mechanization, Ministry Of Agriculture And Rural Affairs | Weeding robot and method, apparatus for planning weeding path for the same and medium |
| US12141985B2 (en) | 2021-09-29 | 2024-11-12 | Nanjing Institute Of Agricultural Mechanization, Ministry Of Agriculture And Rural Affairs | Weeding robot and method and apparatus for planning weeding path thereof, and medium |
| CN113597874A (en) * | 2021-09-29 | 2021-11-05 | 农业农村部南京农业机械化研究所 | Weeding robot and weeding path planning method, device and medium thereof |
| CN114913544A (en) * | 2022-04-28 | 2022-08-16 | 华南农业大学 | Shrimp larvae counting method and system based on semantic segmentation, cloud server and medium |
| CN114913544B (en) * | 2022-04-28 | 2024-09-13 | 华南农业大学 | Shrimp larva counting method and system based on semantic segmentation, cloud server and medium |
| CN117496353A (en) * | 2023-11-13 | 2024-02-02 | 安徽农业大学 | Rice seedling weed stem center distinguishing and positioning method based on two-stage segmentation model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109614973A (en) | Rice seedling and Weeds at seedling image, semantic dividing method, system, equipment and medium | |
| CN110136170B (en) | Remote sensing image building change detection method based on convolutional neural network | |
| CN103258214B (en) | Based on the Classifying Method in Remote Sensing Image of image block Active Learning | |
| CN112906720B (en) | Multi-label image identification method based on graph attention network | |
| CN111079602A (en) | Vehicle fine granularity identification method and device based on multi-scale regional feature constraint | |
| CN113033520B (en) | Tree nematode disease wood identification method and system based on deep learning | |
| CN107679250A (en) | A kind of multitask layered image search method based on depth own coding convolutional neural networks | |
| CN107564025A (en) | A kind of power equipment infrared image semantic segmentation method based on deep neural network | |
| CN107730000A (en) | Refrigerant leakage detection method and device for air conditioner | |
| CN110619059B (en) | Building marking method based on transfer learning | |
| CN106227851A (en) | Based on the image search method searched for by depth of seam division that degree of depth convolutional neural networks is end-to-end | |
| CN111931505A (en) | Cross-language entity alignment method based on subgraph embedding | |
| CN107358142A (en) | Polarimetric SAR Image semisupervised classification method based on random forest composition | |
| CN103020265B (en) | The method and system of image retrieval | |
| CN105809146A (en) | Image scene recognition method and device | |
| CN112949740B (en) | Small sample image classification method based on multilevel measurement | |
| CN104268140B (en) | Image search method based on weight self study hypergraph and multivariate information fusion | |
| CN113761259A (en) | Image processing method and device and computer equipment | |
| CN113255915A (en) | Knowledge distillation method, device, equipment and medium based on structured instance graph | |
| CN112464766A (en) | Farmland automatic identification method and system | |
| CN110765882A (en) | Video tag determination method, device, server and storage medium | |
| CN110807485B (en) | Method for fusing two-classification semantic segmentation maps into multi-classification semantic map based on high-resolution remote sensing image | |
| CN113111716A (en) | Remote sensing image semi-automatic labeling method and device based on deep learning | |
| CN112131261A (en) | Community query method and device based on community network and computer equipment | |
| CN110070026A (en) | Video image danger Situation Awareness method, system, device based on pattern-recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190412 |
|
| RJ01 | Rejection of invention patent application after publication |