Sunflower coverage acquisition method, system, device and medium

Abstract

The present application relates to a sunflower coverage acquisition method, system, device and medium, the method comprising: collecting a plurality of sunflower image data, and preprocessing the same to generate a data set for model training; training based on the data set to obtain a target network model, wherein the target network model comprises a color fusion module for fusing color features and an edge enhancement module for segmenting the edge of a sunflower disc; after obtaining a to-be-identified sunflower picture, performing color feature fusion processing on the to-be-identified sunflower picture using the color fusion module to obtain fused features, and calculating the spatial relationship of the fused features using the edge enhancement module to obtain an output result; determining the sunflower coverage corresponding to the to-be-identified sunflower picture based on the output result. The present application can accurately acquire the coverage of sunflowers.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a sunflower coverage acquisition method, system, device and medium. BACKGROUND

[0002] Sunflower is a widely planted ornamental crop with multiple biological and ecological functions. Sunflower is a high-efficiency photosynthesis plant with strong drought resistance, salt tolerance and acid tolerance, and can grow in adverse environments. Its long flowering period not only provides important pollen and food sources for various insects and birds, but also becomes a beautiful garden and wild landscape. In addition, sunflower seeds contain high levels of protein and oil, as well as rich minerals and vitamins, which have wide applications in the food industry.

[0003] Vegetation coverage of crops refers to the coverage texture formed by crops on the soil surface, and is an important indicator for describing the distribution of ground vegetation. The acquisition and evaluation of crop coverage are very important for agricultural production. Crop vegetation coverage reflects the changes in crop growth conditions and yield. Studies have shown that vegetation coverage is closely related to crop growth conditions, yield and quality. The coverage of crops has a high correlation with the yield of crops, and the yield of crops can be estimated by extracting the coverage of crops. Therefore, how to accurately acquire the coverage of crops has become a technical problem that needs to be solved in agriculture.

[0004] With the development of computer technology, image segmentation technology has been widely used in agriculture. In the prior art, when acquiring sunflower coverage, a traditional image segmentation method can be used to segment crops to acquire coverage. However, when acquiring sunflower coverage, the traditional image segmentation method cannot segment the sunflower disc well, and thus cannot accurately acquire the coverage of sunflower. SUMMARY

[0005] The present application aims to provide a sunflower coverage acquisition method, system, device and medium, which can accurately acquire the coverage of sunflower.

[0006] To achieve the above-mentioned purpose, in a first aspect, the present application provides a sunflower coverage acquisition method, which comprises:

[0007] Collecting a plurality of sunflower image data and preprocessing the same to generate a data set for model training;

[0008] Training based on the data set to obtain a target network model, wherein the target network model comprises a color fusion module for fusing color features and an edge enhancement module for segmenting the edge of the sunflower disc.

[0009] After obtaining the sunflower picture to be identified, the color fusion module is used for color feature fusion processing on the sunflower picture to be identified to obtain a fusion feature, and the edge enhancement module is used for calculating the spatial relationship of the fusion feature to obtain an output result.

[0010] Based on the output result, the sunflower coverage degree corresponding to the sunflower picture to be identified is determined.

[0011] Preferably, the color fusion module is used for color feature fusion processing on the sunflower picture to be identified to obtain a fusion feature, including:

[0012] Obtaining the RGB picture feature corresponding to the sunflower picture to be identified and the vegetation index picture feature calculated from the RGB picture feature;

[0013] The RGB picture feature and the vegetation index picture feature are fused to obtain a fusion feature.

[0014] Preferably, the RGB picture feature and the vegetation index picture feature are fused to obtain a fusion feature, specifically including:

[0015] Setting X represents the RGB picture feature and Y represents the vegetation index picture feature; performing global average pooling Fgp on the feature Y to obtain a feature U, and passing through two fully connected layers Ffc to obtain a feature S;

[0016] The feature S is activated by a softmax function, a weight value in a specified interval is obtained for each channel of the feature S, and the weight value is point multiplied with the feature X to obtain a fusion feature Z after fusion, and the calculation formula is:

[0017] Z=X*softmax(Ffc(Ffc(Fgp(Y)))).

[0018] Preferably, the edge enhancement module is used for calculating the spatial relationship of the fusion feature to obtain an output result, specifically including:

[0019] Setting X represents the input feature; the input feature X is respectively subjected to global maximum pooling Fgmp and global average pooling Fgap to obtain two features A and B, the two features A and B are pixel by pixel added to obtain a preliminary fusion feature C, the feature C is subjected to convolution to obtain a feature D, the feature D is activated by an activation function to assign a weight value to each pixel, and finally the weight value is multiplied with the initial feature to obtain an output feature E, and the calculation formula is:

[0020] E=X*(conv(Fgap(X)+Fgmp(X))).

[0021] Preferably, the process of fusing the RGB image features and the vegetation index image features to obtain the fused features specifically includes:

[0022] The RGB image features are convolved to obtain feature A1, and the vegetation index image features are convolved to obtain feature B1. Feature A1 and feature B1 are fused together for the first time to obtain fused feature C1. Feature C1 is downsampled to obtain feature D1. Feature D1 is convolved to obtain feature A2. Feature B1 is downsampled to obtain feature E1. Feature E1 is convolved to obtain feature B2. Feature A2 and feature B2 are fused together for the second time to obtain fused feature C2. In this way, multiple fused features are obtained after at least two fusion processes. The final fused feature C4 is downsampled and convolved to obtain fused feature C5.

[0023] Preferably, the step of using the edge enhancement module to calculate the spatial relationship of the fused features to obtain the output result specifically includes:

[0024] After performing spatial relationship calculations on the fused feature C5 using the edge enhancement module, feature F1 is obtained. Feature F1 and feature C4 are then combined through channel stacking to obtain feature G1. Feature G1 is then convolved to obtain feature C6. After performing spatial relationship calculations on feature C6 using the edge enhancement module, feature F2 is obtained. Feature F2 and feature C3 are then combined through channel stacking to obtain feature G2. Feature G2 is then convolved to obtain feature C7. In this way, after at least two consecutive operations by the edge enhancement module, feature C9 is obtained. Finally, feature C9 is convolved to obtain the final output result.

[0025] Preferably, determining the sunflower coverage corresponding to the sunflower image to be identified based on the output result specifically includes:

[0026] Based on the output results, a segmented image of the sunflower head is obtained. The segmented image is then binarized, and the pixel ratio of the sunflower head is calculated to obtain the sunflower coverage.

[0027] Secondly, the present invention provides a sunflower coverage acquisition system, the sunflower coverage acquisition system comprising:

[0028] The acquisition module is used to acquire a number of sunflower image data and preprocess them to generate a dataset for model training.

[0029] a training module configured to train a target network model based on the dataset, wherein the target network model comprises a color fusion module configured to fuse color features and an edge enhancement module configured to segment edges of a sunflower disc;

[0030] a calculating module configured to, after obtaining a to-be-identified sunflower picture, perform color feature fusion processing on the to-be-identified sunflower picture by using the color fusion module to obtain fused features, and perform calculation on spatial relationships of the fused features by using the edge enhancement module to obtain an output result;

[0031] a determining module configured to determine a sunflower coverage degree corresponding to the to-be-identified sunflower picture based on the output result.

[0032] In a third aspect, the present application provides an electronic device, which comprises a memory and a processor, and the memory stores a computer program capable of running on the processor, and the computer program is executed by the processor to implement the steps of the sunflower coverage degree acquisition method.

[0033] In a fourth aspect, the present application provides a computer readable storage medium, which stores a sunflower coverage degree acquisition system, and the sunflower coverage degree acquisition system can be executed by at least one processor to make the at least one processor execute the steps of the sunflower coverage degree acquisition method.

[0034] The beneficial effects of the above-mentioned embodiments are as follows: the present application trains a target network model based on deep learning by collecting a plurality of sunflower image data, wherein the target network model comprises a color fusion module configured to fuse color features and an edge enhancement module configured to segment edges of a sunflower disc; after obtaining a to-be-identified sunflower picture, the color fusion module is used to perform color feature fusion processing on the to-be-identified sunflower picture to obtain fused features, and the edge enhancement module is used to calculate spatial relationships of the fused features to obtain an output result; and the sunflower coverage degree corresponding to the to-be-identified sunflower picture is determined based on the output result. Since the color features of the sunflower picture can be fused by using the trained target network model, and the edge features of the sunflower disc are enhanced and segmented, the edge part of the sunflower disc can be more effectively segmented, and the sunflower coverage degree can be accurately acquired. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 Method flow of the sunflower coverage degree acquisition method embodiment provided by the present application Figure 1 ;

[0036] Figure 2Method flow in the sunflower coverage acquisition method provided by the present application Figure 2 ;

[0037] Figure 3 Sunflower image after cropping in the sunflower coverage acquisition method provided by the present application

[0038] Figure 4 Structure diagram of the unet-color-edge network in the sunflower coverage acquisition method provided by the present application

[0039] Figure 5 Structure diagram of the unet-color module in the sunflower coverage acquisition method provided by the present application

[0040] Figure 6 Structure diagram of the unet-edge module in the sunflower coverage acquisition method provided by the present application

[0041] Figure 7 Comparison diagram of segmentation effects of different networks in the sunflower coverage acquisition method provided by the present application

[0042] Figure 8 Principle block diagram of the sunflower coverage acquisition system provided by the present application

[0043] Figure 9 Principle block diagram of the electronic device provided by the present application DETAILED DESCRIPTION

[0044] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0045] It should be noted that the description of "first", "second" and the like in the present application is only for the purpose of description and should not be understood as indicating or implying the relative importance of the technical features indicated or the number of the technical features indicated. Therefore, the features with "first" and "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the realization of those of ordinary skill in the art, and when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist and is not within the scope of protection required by the present application.

[0046] The application provides a sunflower coverage acquisition method, a system, an electronic device and a storage medium, which are described in detail below.

[0047] Referring to Figure 1 A method flowchart of an embodiment of the sunflower coverage acquisition method provided by the application is provided. The method comprises the following steps:

[0048] S1, a plurality of sunflower image data are collected and preprocessed to generate a data set for model training;

[0049] S2, a target network model is obtained based on the data set, wherein the target network model comprises a color fusion module for fusing color features and an edge enhancement module for segmenting the edge of a sunflower disc;

[0050] S3, after obtaining a to-be-identified sunflower picture, the color fusion module is used to perform color feature fusion processing on the to-be-identified sunflower picture to obtain fusion features, and the edge enhancement module is used to calculate the spatial relationship of the fusion features to obtain an output result;

[0051] S4, the sunflower coverage corresponding to the to-be-identified sunflower picture is determined based on the output result.

[0052] The application trains a target network model based on deep learning by collecting a plurality of sunflower image data, wherein the target network model comprises a color fusion module for fusing color features and an edge enhancement module for segmenting the edge of a sunflower disc. After obtaining a to-be-identified sunflower picture, the color fusion module is used to perform color feature fusion processing on the to-be-identified sunflower picture to obtain fusion features, and the edge enhancement module is used to calculate the spatial relationship of the fusion features to obtain an output result. The sunflower coverage corresponding to the to-be-identified sunflower picture is determined based on the output result. Since the trained target network model can be used to fuse the color features of the sunflower picture and enhance the features of the edge of the sunflower disc, the edge part of the sunflower disc can be more effectively segmented, and the coverage of the sunflower can be accurately acquired.

[0053] Further, in some embodiments, the color fusion module is used to perform color feature fusion processing on the to-be-identified sunflower picture to obtain fusion features, comprising:

[0054] obtaining the RGB picture features corresponding to the to-be-identified sunflower picture and the vegetation index picture features calculated from the RGB picture features;

[0055] Fuse the RGB picture features and the vegetation index picture features to obtain fused features.

[0056] In some embodiments, the RGB picture features and the vegetation index picture features are fused to obtain fused features, specifically including:

[0057] Let X represent the RGB picture features and Y represent the vegetation index picture features; the features Y are globally averaged and pooled to obtain features U, and two fully connected layers Ffc are used to obtain features S;

[0058] The features S are activated by a softmax activation function to obtain a weight value in a specified interval for each channel of the features S, and the weight value is point multiplied with the features X to obtain fused features Z after fusion, and the calculation formula is:

[0059] Z = X * softmax (Ffc (Ffc (Fgp (Y)))).

[0060] In some embodiments, the spatial relationship of the fused features is calculated using the edge enhancement module to obtain an output result, specifically including:

[0061] Let X represent the input features; the input features X are respectively subjected to global maximum pooling Fgmp and global average pooling Fgap to obtain two features A and B, the two features A and B are pixel by pixel added to obtain preliminary fused features C, and the features C are subjected to convolution to obtain features D, the features D are activated by an activation function to assign a weight value to each pixel, and finally the weight value is multiplied with the initial features to obtain output features E, and the calculation formula is:

[0062] E = X * (conv (Fgap (X) + Fgmp (X))).

[0063] In some embodiments, the RGB picture features and the vegetation index picture features are fused to obtain fused features, specifically including:

[0064] The RGB picture features are subjected to convolution to obtain features A1, the vegetation index picture features are subjected to convolution to obtain features B1, and the features A1 and the features B1 are subjected to first fusion processing to obtain fused features C1; the features C1 are subjected to downsampling operation to obtain features D1, the features D1 are subjected to convolution to obtain features A2, the features B1 are subjected to downsampling to obtain features E1, the features E1 are subjected to convolution to obtain features B2, and the features A2 and the features B2 are subjected to second fusion processing to obtain fused features C2; in this way, after at least twice fusion processing, a plurality of fused features are obtained, and the finally obtained fused features C4 are subjected to downsampling and convolution to obtain fused features C5.

[0065] In some embodiments, the spatial relationship of the fused features is calculated by the edge enhancement module to obtain an output result, specifically including:

[0066] After the spatial relationship of the fused feature C5 is calculated by the edge enhancement module, a feature F1 is obtained. After the channel superposition operation of the feature F1 and the feature C4, a feature G1 is obtained. After the convolution of the feature G1, a feature C6 is obtained. After the spatial relationship of the feature C6 is calculated by the edge enhancement module, a feature F2 is obtained. After the channel superposition operation of the feature F2 and the feature C3, a feature G2 is obtained. After the convolution of the feature G2, a feature C7 is obtained. In this way, after at least two consecutive operations of the edge enhancement module, a feature C9 is obtained, and after the convolution of the feature C9, a final output result is obtained.

[0067] Preferably, the output result is used to determine the sunflower coverage corresponding to the sunflower picture to be identified, specifically including:

[0068] According to the output result, a segmented sunflower picture is obtained. The segmented sunflower picture is binarized, and the pixel ratio of the sunflower disc is calculated to obtain the coverage of the sunflower.

[0069] Specifically, referring to Figure 2 In an alternative embodiment, the present application comprises the following steps:

[0070] Step 1: Data acquisition. In the present application, when a plurality of sunflower image data are collected and preprocessed to generate a data set for model training, the data that can be collected include but are not limited to: data 1, the unmanned aerial vehicle used for data collection is DJI Spirit Phantom 4Pro V2.0, equipped with a 2x107 pixel camera, and the size of a single image captured is 5472x3648. An automatic planning aerial photography mode is adopted, the heading overlap rate and the lateral overlap rate are set to 75%, the flight height is 10m, and the photographing is performed in an equidistant interval manner, the flight speed is 1.9m / s, and about 10min is required to complete the data collection of the study area. And data 2, using a search engine or a python crawler to search for sunflower disc data sets on the Internet.

[0071] Step 2: Construction of data set. The unmanned aerial vehicle collects sunflower RGB image data of different flower periods and different varieties. The acquired data is preprocessed to obtain training data for the network model. First, the unmanned aerial vehicle aerial image of the study area is spliced to obtain the orthographic image of the field; then, the orthographic image is cropped according to the actual ground size to obtain the image of each small area.

[0072] When the image data of each cell is obtained, the sunflower disc in the picture is labeled using the labeling tool labelme, and then the json file obtained by labeling is processed to convert it into a mask picture required for semantic segmentation.

[0073] Considering that the obtained image data is less, it may not meet the feature extraction of model training. In order to solve this problem, the original cell picture marked is cropped in this study, and the size of the cropped picture is 512x512 pixels. The cropped picture is shown in Figure 3 Subsequently, data augmentation is performed on the cropped picture to obtain the data set required for the final network model training.

[0074] When the data samples required for network training are obtained, the data set is allocated as a training set, a validation set and a test set in a ratio of 8:1:1.

[0075] Step 3: segmentation method based on Unet-Color-Edge network. The deep learning semantic segmentation network used in this application is Unet-Color-Edge network. The network is a U-shaped structure, as shown in Figure 4 The network is roughly divided into two parts, the left encoder part and the right decoder part. It includes:

[0076] (1) Unet-Color module fusing color feature prior. The left encoder structure of the network contains the unet-color module fusing color features. The function of this module is to fuse two different features. In this application, the two features are an RGB picture and a vegetation index picture calculated from the RGB picture. In this application, the specific operation of this module is shown in Figure 5

[0077] X represents the feature obtained from the RGB picture, and Y represents the feature obtained from the vegetation index picture. It is assumed that the size of X and Y is HxWxC. First, Y is globally averaged pooled (Fgp) to obtain a feature U with a size of 1x1xC, and then two fully connected layers Ffc are used to obtain a feature S. The feature S is activated by a softmax function to obtain a weight value for each channel of the feature S in the interval of 0-1. Finally, the weight value is multiplied by the feature X to obtain the fused feature Z, which has the same size as X and Y. The specific formula is:

[0078] Z=X*softmax(Ffc(Ffc(Fgp(Y)))).

[0079] ​(2) The network module of Unet-Edge with edge enhancement. In order to better segment the edge of the sunflower disc, the present application proposes a unet-edge module, which can make the network pay more attention to the spatial relationship of features, generate a spatial attention map using the spatial relationship, and the specific operation is as shown in Figure 6

[0080] The size of the input feature X is HxWxC. The input feature is respectively subjected to global maximum pooling (Fgmp) and global average pooling (Fgap) to obtain two features A and B, and the size of the features is HxWx1. Then, the two features are added pixel by pixel to obtain a preliminary fusion feature C, and the size of the feature C is HxWx2. The feature C is subjected to a single-channel convolution to obtain a feature D, and the size of the feature D is HxWx1. Finally, D is activated by an activation function to assign a weight value to each pixel. Finally, the weight value is multiplied by the initial feature to obtain an output feature E. The specific formula is:

[0081] E=X*(conv(Fgap(X)+Fgmp(X)))

[0082] (3) The multi-module fusion strategy of Unet-Color-Edge. The present application proposes two modules, namely the Unet-color module for fusing color features and the Unet-edge module for edge attention. The specific fusion strategy of the network is as shown in Figure 4

[0083] The size of the RGB picture input into the network is 512x512x3. After the first convolution module, a feature A1 is obtained, and the size of the feature A1 is 512x512x64. The first convolution module includes two 3x3x64 convolution kernels and a ReLU activation function. In order to ensure that the picture size is the same before and after convolution, padding operation needs to be performed on the picture.

[0084] The present application obtains a single-channel vegetation index map by calculating the vegetation index of the RGB picture, then copies the vegetation index map three times to obtain a three-channel vegetation index map with a size of 512x512x3, and then sends it into the network to obtain a feature B1 with a size of 512x512x64 after the first convolution module. The feature A1 and the feature B1 are sent into the first unet-color module to perform the first fusion of the color features provided by the RGB picture and the vegetation index, and an output feature C1 with a size of 512x512x64 is obtained.

[0085] ​​Feature C1 is obtained by downsampling operation, and the size is 256x256x64. Feature D1 is sent to the second convolution module to obtain feature A2 with the size of 256x256x128. The second convolution module includes two 3x3x128 convolutions and a Relu activation function. Feature B1 obtained from the vegetation index is downsampled to obtain feature E1 with the size of 256x256x64. Feature E1 is sent to the second convolution module to obtain feature B2 with the size of 256x256x128. Feature A2 and feature B2 are sent to the second unet-color module to perform the second fusion of the RGB picture and the color feature provided by the vegetation index, and output feature C2 with the size of 256x256x128 is obtained.

[0086] Similarly, the two branches are fused four times to obtain feature C4 with the size of 64x64x512. Feature C4 is downsampled and sent to the fifth convolution module to obtain feature C5 with the size of 32x32x512.

[0087] The above is the operation of the encoder part, Figure 4 The right side is the operation of the decoder, which is as follows:

[0088] Feature C5 is obtained by the first unet-edge module, and feature F1 with the same size is obtained. Feature F1 and feature C4 of the encoder are skip-connected, specifically channel superposition, to obtain feature G1 with the size of 64x64x1024. Feature G1 is sent to the sixth convolution module to obtain feature C6 with the size of 64x64x512.

[0089] After four consecutive operations, feature C9 with the size of 512x512x64 is obtained. Feature C9 is finally sent to a 1x1 convolution to obtain the output result of the network.

[0090] Step 4: Sunflower coverage calculation. The dataset is sent to the improved network for training to obtain a trained model. Then, the model is used to predict the sunflower picture to obtain the segmentation picture of the sunflower disc. The obtained segmentation picture is binarized, and the pixel ratio of the flower is calculated to obtain the coverage of the sunflower.

[0091] With the development of computer technology, image segmentation technology has a wide range of applications in agriculture. At present, there are many segmentation methods applied to crop segmentation, which can be roughly divided into two categories: traditional image segmentation methods and deep learning segmentation methods. The common methods based on traditional learning are based on color, shape and texture. For flower images collected by remote sensing platform, color threshold and clustering method are widely used. The deep learning segmentation method mainly refers to using classic semantic segmentation network to segment crops. The commonly used semantic segmentation network includes unet, deeplabv3+ and other classic networks.

[0092] Compared with other classic semantic segmentation methods, the sunflower disc segmentation method of the unet-color-edge network proposed in the application has obvious advantages in image segmentation quality and algorithm performance indicators, and can effectively segment the edge part of the flower disc.

[0093] Further, in order to prove the effectiveness of the proposed network, the inventors of the application compared the segmentation effect of the unet-color-edge network with the classic semantic segmentation network unet and deeplabv3+.

[0094] The application sends the data set into the unet and deeplabv3+ networks for segmentation, and compares the segmentation results with the segmentation results of the unet-color-edge network, and the specific visualization results are as shown in Figure 7 Figure 7 It can be seen that the segmentation effect of the unet-color-egde proposed in the application is the best, which can well segment the edge of the sunflower disc. The second is the unet network, which can basically segment the sunflower disc well. In the segmentation map of the deepabv3+ network, there are many noise points.

[0095] The specific segmentation accuracy comparison is shown in Table 1 below:

[0096] Miou (%) Map (%) Mrecall (%) Mprecision (%) Unet-color-edge 90.4 93.9 93.4 95.4 Unet 88.1 92.4 91.8 93.1 Deeplabv3+ 85.4 91.7 90.8 90.7

[0097] Table 1

[0098] From Table 1, it can be seen that the unet-color-edge network proposed in the application has a significantly higher accuracy index than the classic semantic segmentation network unet and deeplabv3+. Therefore, compared with the prior art, the application can more accurately obtain the coverage of sunflower.

[0099] The sunflower disc segmentation method based on color prior and edge constraint proposed in the application can be applied in the following aspects:

[0100] ​By segmenting the sunflower disc, the coverage of the sunflower disc can be obtained, and the coverage can be linearly correlated with the number of sunflower discs and the yield of sunflowers. By calculating the correlation, a model for estimating the number of sunflower discs and the yield of sunflowers from the coverage of the sunflower disc can be obtained.

[0101] In many regions, there are cases of growing sunflowers and zucchini together in the same field. Using the segmentation network proposed in the present application, the sunflower disc can be effectively segmented in the farmland where the two symbiotic plants are planted, and the coverage of the sunflower can be more accurately obtained.

[0102] It should be understood that the size of the serial number of each step in the above embodiments does not mean the order of execution, and the execution order of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0103] The present embodiment also provides a sunflower coverage obtaining system, which corresponds one-to-one to the sunflower coverage obtaining method in the above embodiments. As shown in the figure, the sunflower coverage obtaining system comprises: Figure 8

[0104] The acquisition module 20 is configured to acquire a plurality of sunflower image data and pre-process the same to generate a data set for model training;

[0105] The training module 21 is configured to train a target network model based on the data set, wherein the target network model comprises a color fusion module for fusing color features and an edge enhancement module for segmenting the edges of sunflower discs;

[0106] The calculation module 22 is configured to, after obtaining a to-be-identified sunflower picture, perform color feature fusion processing on the to-be-identified sunflower picture by using the color fusion module to obtain fused features, and calculate the spatial relationship of the fused features by using the edge enhancement module to obtain an output result;

[0107] The determination module 23 is configured to determine the sunflower coverage corresponding to the to-be-identified sunflower picture based on the output result.

[0108] For specific limitations of each module of the sunflower coverage obtaining system, please refer to the limitations of the sunflower coverage obtaining method described above, which will not be repeated here. Each module in the above sunflower coverage obtaining system can be realized by software, hardware and their combination. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above modules. ​

[0109] Referring to Figure 9 is a schematic diagram of an operating environment of a preferred embodiment of the sunflower coverage acquisition system 10 of the present application.

[0110] In this embodiment, the sunflower coverage acquisition system 10 is installed and operated in an electronic device 1. The electronic device 1 is a device capable of automatically performing numerical calculation and / or information processing according to pre-set or stored instructions. The electronic device 1 can be a computer, a single network server, a server group composed of multiple network servers, or a cloud composed of a large number of hosts or network servers based on cloud computing. Cloud computing is a kind of distributed computing, which is a super virtual computer composed of a group of loosely coupled computer clusters. (The electronic device 1 can be a server, a smart phone, a tablet computer, a portable computer, a desktop computer, etc. Terminal equipment with storage and operation functions. In an embodiment, when the electronic device 1 is a server, the server can be one or more of a rack-mounted server, a blade server, a tower server, or a cabinet server, etc.)

[0111] In this embodiment, the electronic device 1 can include, but is not limited to, a memory 11, a processor 12, and a network interface 13 that can be connected to each other through a system bus, and the memory 11 stores the sunflower coverage acquisition system 10 that can be run on the processor 12. It should be noted that, Figure 1 Only the electronic device 1 with components 11-13 is shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0112] The memory 11 includes an internal memory and at least one type of readable storage medium. The internal memory provides a cache for the operation of the electronic device 1; the readable storage medium can be a nonvolatile storage medium such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the readable storage medium can be an internal storage unit of the electronic device 1, such as a hard disk of the electronic device 1; in other embodiments, the nonvolatile storage medium can also be an external storage device of the electronic device 1, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. In this embodiment, the readable storage medium of the memory 11 is generally used to store an operating system and various application software installed in the electronic device 1, such as the sunflower coverage acquisition system 10 in an embodiment of the present application. In addition, the memory 11 can also be used to temporarily store various data that have been output or will be output.

[0113] The processor 12 can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 12 is generally used to control the overall operation of the electronic device 1, such as performing control and processing related to data interaction or communication with other devices, etc. In this embodiment, the processor 12 is used to run program codes or process data stored in the memory 11, such as the sunflower coverage acquisition system 10, etc.

[0114] The network interface 13 can include a wireless network interface or a wired network interface, and the network interface 13 is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0115] The sunflower coverage acquisition system 10 includes at least one computer readable instruction stored in the memory 11, and the at least one computer readable instruction can be executed by the processor 12 to implement various embodiments of the present application.

[0116] In addition, the present application also provides a computer readable storage medium storing a sunflower coverage acquisition system, and the sunflower coverage acquisition system can be executed by at least one processor to enable the at least one processor to perform various steps in the above embodiments.

[0117] The computer readable storage medium embodiment of the present application is basically the same as the above-mentioned electronic device 1 and method embodiments, and will not be described here.

[0118] It should be noted that in this paper, the term "include", "contain" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or device. Without more limitations, the element defined by the sentence "includes a…" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0119] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment method can be realized by means of software and the necessary general hardware platform, of course, it can also be realized by hardware, but in many cases the former is the better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for making a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) execute the method described in each embodiment of the present application.

[0120] The preferred embodiments of the present application are described above with reference to the accompanying drawings, which do not limit the scope of the present application. The above-mentioned embodiment serial numbers of the present application are only for description, not representing the advantages and disadvantages of the embodiments. In addition, although the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in an order different from that here.

[0121] Those skilled in the art can have various modification schemes to realize the present application without departing from the scope and essence of the present application, such as using the features of one embodiment in another embodiment to obtain another embodiment. Any modification, equivalent replacement and improvement within the technical concept of the present application shall be within the scope of the present application.

Claims

1. A method for obtaining sunflower coverage, characterized in that, The sunflower coverage acquisition method comprises: Collecting a plurality of sunflower image data and pre-processing the same to generate a data set for model training; Training based on the data set to obtain a target network model, wherein the target network model comprises a color fusion module for fusing color features and an edge enhancement module for segmenting the edge of a sunflower disc; After obtaining a to-be-identified sunflower picture, performing color feature fusion processing on the to-be-identified sunflower picture by using the color fusion module to obtain fused features, and calculating the spatial relationship of the fused features by using the edge enhancement module to obtain an output result; Determining the sunflower coverage corresponding to the to-be-identified sunflower picture based on the output result; The color feature fusion processing on the to-be-identified sunflower picture by using the color fusion module to obtain fused features comprises: Obtaining an RGB picture feature corresponding to the to-be-identified sunflower picture and a vegetation index picture feature calculated from the RGB picture feature; Fusing the RGB picture feature and the vegetation index picture feature to obtain fused features; The calculation of the spatial relationship of the fused features by using the edge enhancement module to obtain an output result comprises: Setting X to represent an input feature; obtaining two features A and B by respectively passing the input feature X through global maximum pooling Fgmp and global average pooling Fgap; adding the two features A and B pixel by pixel to obtain a preliminary fused feature C, and passing the feature C through convolution to obtain a feature D, activating the feature D through an activation function, assigning a weight value to each pixel, and finally multiplying the weight value with the initial feature to obtain an output feature E, and the calculation formula is: E=X*(conv(Fgap(X)+Fgmp(X))); The fusing of the RGB picture feature and the vegetation index picture feature to obtain fused features comprises: Passing the RGB picture feature through convolution to obtain a feature A1, passing the vegetation index picture feature through convolution to obtain a feature B1, performing first fusion processing on the feature A1 and the feature B1 to obtain a fused feature C1; performing down-sampling operation on the feature C1 to obtain a feature D1, passing the feature D1 through convolution to obtain a feature A2, passing the feature B1 through down-sampling to obtain a feature E1, passing the feature E1 through convolution to obtain a feature B2, performing second fusion processing on the feature A2 and the feature B2 to obtain a fused feature C2; and in this way, a plurality of fused features are obtained after at least twice fusion processing, and a fused feature C5 is obtained by passing the finally obtained fused feature C4 through down-sampling and convolution; The calculation of the spatial relationship of the fused features by using the edge enhancement module to obtain an output result comprises: The edge enhancement module is used for spatial relationship calculation on the fusion feature C5, feature F1 is obtained, feature G1 is obtained after channel superposition operation of feature F1 and feature C4, and feature C6 is obtained after convolution of feature G1; the edge enhancement module is used for spatial relationship calculation on feature C6, feature F2 is obtained, feature G2 is obtained after channel superposition operation of feature F2 and feature C3, and feature C7 is obtained after convolution of feature G2; in this way, after at least two times of operation of the edge enhancement module, feature C9 is obtained, and the final output result is obtained after convolution of feature C9.

2. The sunflower coverage acquisition method according to claim 1, characterized in that, The RGB picture feature and the vegetation index picture feature are fused to obtain a fusion feature, and the fusion feature is obtained by performing the following steps: X represents the RGB picture feature, and Y represents the vegetation index picture feature; feature Y is globally averaged by a global average pooling layer Fgp to obtain feature U, and two fully connected layers Ffc are used to obtain feature S; The feature S is activated by a softmax function, a weight value in a specified interval is obtained for each channel of the feature S, and the weight value is multiplied by the feature X to obtain the fusion feature Z after fusion, and the calculation formula is: Z=X*softmax(Ffc(Ffc(Fgp(Y)))).

3. The sunflower coverage acquisition method according to claim 1, characterized in that, The output result is used to determine the sunflower coverage corresponding to the sunflower picture to be recognized, and the sunflower coverage corresponding to the sunflower picture to be recognized is determined by performing the following steps: A segmented picture of a sunflower disc is obtained according to the output result, the segmented picture is binarized, and the pixel ratio of the sunflower disc is calculated to obtain the sunflower coverage.

4. A sunflower coverage acquisition system for implementing the method according to any one of claims 1 to 3, characterized in that, The sunflower coverage acquisition system comprises: A collection module is configured to collect a plurality of sunflower image data and pre-process the sunflower image data to generate a data set for model training; A training module is configured to train a target network model based on the data set, wherein the target network model comprises a color fusion module for fusing color features and an edge enhancement module for segmenting edges of a sunflower disc; A calculation module is configured to fuse color features of a sunflower picture to be recognized by using the color fusion module to obtain a fusion feature, and calculate a spatial relationship of the fusion feature by using the edge enhancement module to obtain an output result; A determination module is configured to determine a sunflower coverage corresponding to the sunflower picture to be recognized based on the output result.

5. An electronic device, comprising: The electronic device comprises a memory and a processor, and a computer program is stored on the memory and can run on the processor, and the computer program is executed by the processor to implement the steps of the sunflower coverage acquisition method according to any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer program is stored on the computer readable storage medium and is executed by the processor to implement the steps of the sunflower coverage acquisition method according to any one of claims 1 to 3.

Images