Crop identification method based on remote sensing image of unmanned aerial vehicle

By introducing the MobileNetV2 model with attention mechanism and the CA attention mechanism in the DeepLab V3+ model, the improved model improves the accuracy of crop recognition in the remote sensing image of drone, solves the problem of insufficient recognition accuracy in the existing technology, and realizes Higher identification accuracy and computing resource efficiency.

CN118918471AActive Publication Date: 2024-11-08CHENGDU UNIVERSITY OF TECHNOLOGY +1

Patent Information

Application Number
CN202410962535.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-08
Estimated Expiration
2044-07-18

AI Technical Summary

Technical Problem

The existing drone remote sensing image crop identification methods have deviations in accurately identifying crop structures and extracting local information, and the recognition accuracy is insufficient, making it difficult to meet the needs of agricultural production management and food security.

Method used

The improved DeepLab V3+ model is adopted, and the accuracy and accuracy of crop recognition are improved by replacing the MobileNetV2 model with attention mechanism with the backbone structure in the DeepLab V3+ model, combined with the CA attention mechanism.

Benefits of technology

Compared with the original DeepLab V3+ model, the improved model has significantly improved crop recognition accuracy, with MIoU increasing by 1.04%, MPA increasing by 2.37%, Recall and Precision increasing by 0.05%, F1-Score increasing by 1%, corn and walnuts The classification accuracy of crops such as soybeans has also been significantly improved, and the complex structure of crops can be extracted more accurately.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918471A_ABST
    Figure CN118918471A_ABST
Patent Text Reader

Abstract

The invention discloses a crop recognition method based on an unmanned aerial vehicle remote sensing image, and relates to the technical field of crop recognition, a trained crop recognition model is designed, an improved DeepLab V3 + model is adopted, the improved DeepLab V3 + model is obtained by replacing a trunk structure in the DeepLab V3 + model with a MobileNet V2 model with an attention mechanism, and the DeepLab V3 + model is obtained by improving the DeepLab V3 + model. After the unmanned aerial vehicle remote sensing image obtained by shooting the crops is obtained, the trained crop recognition model is directly utilized to recognize the unmanned aerial vehicle remote sensing image to obtain the crop recognition result, and compared with a DeepLab V3 + model, the improved DeepLab V3 + model is adopted to recognize the crops, so that the recognition precision is higher.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of crop identification, and in particular to a crop identification method based on unmanned aerial vehicle remote sensing images. Background Art

[0002] Crop spatial distribution information is an important basic data for agricultural production management, crop yield estimation, and agricultural land structure adjustment. Traditional field survey methods have long cycles and high costs, resulting in high uncertainty and time lag in the spatial distribution information of crops. Timely and accurate grasp of crop spatial distribution information is of great significance for adjusting crop planting structure, formulating agricultural production layout plans, promoting efficient use of water and soil resources, and ensuring food security.

[0003] The rapid development of drones has enabled them to be widely used in agriculture, surveying engineering, urban and rural construction, ecological environment and other fields. As an emerging technology, drone remote sensing technology has gradually become one of the mainstream ways to obtain low-altitude remote sensing images, effectively making up for the shortcomings of aerospace remote sensing. At present, drone remote sensing technology is often used to obtain drone remote sensing images of crops, and drone remote sensing images are identified in order to identify crops. In recent years, in the research field of drone remote sensing image classification and recognition, the application of convolutional neural networks (CNNs) based on deep learning has gradually increased, and the research results obtained have also performed well. Convolutional neural networks use convolutional layers and fully connected layers as key structural modules. Due to the partial connection and shared parameter structure of the convolutional layer, it can automatically extract features and allow deeper feature extraction, with better robustness advantages. It can effectively avoid the problems of insufficient extracted features and difficult reconstruction of diverse data in manual feature extraction. The fully connected layer is used as a module to complete the classification task. Image classification is realized based on the features extracted by the convolutional layer, which usually has good results. Many researchers in the field of remote sensing have found that convolutional neural networks can automatically extract features while maintaining high image classification and recognition accuracy, and have begun to study the application of convolutional neural networks in remote sensing image classification. Today, deep learning has attracted widespread attention from experts because of its low requirements for feature extraction, no need for expert participation, reduced human intervention, and more comprehensive feature extraction. For crop identification, the recognition accuracy of the DeepLab V3+ model needs to be improved. It is not enough to accurately grasp the crop structure in a certain area, and there is still deviation in the local extracted information. Summary of the invention

[0004] The purpose of this application is to provide a crop recognition method based on UAV remote sensing images, using an improved DeepLab V3+ model to identify crops with high recognition accuracy.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] The present application provides a method for identifying crops based on UAV remote sensing images, and the method for identifying crops based on UAV remote sensing images includes:

[0007] Obtaining drone remote sensing images of crops;

[0008] The UAV remote sensing image is used as input, and the trained crop recognition model is used for recognition to obtain a crop recognition result; the trained crop recognition model adopts the improved DeepLab V3+ model, and the improved DeepLab V3+ model is obtained by replacing the backbone structure in the DeepLab V3+ model with the MobileNetV2 model with an attention mechanism to improve the DeepLabV3+ model; the crop recognition result includes the crop type at each location in the UAV remote sensing image.

[0009] Optionally, the MobileNetV2 model with an attention mechanism includes: a MobileNet V2 model and an attention mechanism connected in sequence, and the attention mechanism is a CA attention mechanism.

[0010] Optionally, the MobileNetV2 model includes an input layer, a residual block layer, and a classification layer connected in sequence;

[0011] The input layer includes a first convolutional block;

[0012] The residual block layer includes a second convolution block, a plurality of third convolution blocks and a fourth convolution block connected in sequence;

[0013] The classification layer includes a fifth convolutional block.

[0014] Optionally, the first convolution block includes a convolution layer, a batch normalization layer and an activation function layer connected in sequence;

[0015] The second convolution block includes a convolution layer, a batch normalization layer, an activation function layer, a convolution layer and a batch normalization layer connected in sequence;

[0016] The third convolution block includes a convolution layer, a batch normalization layer, an activation function layer, a convolution layer, a batch normalization layer, an activation function layer, a convolution layer and a batch normalization layer connected in sequence;

[0017] The fourth convolution block includes a batch normalization layer, an activation function layer, a convolution layer, a batch normalization layer, an activation function layer, a convolution layer and a batch normalization layer connected in sequence;

[0018] The fifth convolution block includes a convolution layer, a batch normalization layer, and an activation function layer connected in sequence.

[0019] Optionally, before taking the UAV remote sensing image as input and using a trained crop recognition model to perform recognition to obtain a crop recognition result, the crop recognition method based on UAV remote sensing image further includes:

[0020] Acquire a data set; the data set includes a plurality of sample UAV remote sensing images and a label image corresponding to each of the sample UAV remote sensing images, wherein the label image is obtained by marking the crop type at each location in the sample UAV remote sensing image;

[0021] Constructing an initial crop recognition model; the initial crop recognition model adopts an improved DeepLab V3+ model;

[0022] The initial crop recognition model is trained using the data set to obtain a trained crop recognition model.

[0023] Optionally, obtain a data set, including:

[0024] Obtain original drone remote sensing images;

[0025] The original UAV remote sensing image is cropped multiple times to obtain multiple sample UAV remote sensing images;

[0026] For each of the sample UAV remote sensing images, the sample UAV remote sensing image is annotated to obtain a label image corresponding to the sample UAV remote sensing image; and all the sample UAV remote sensing images and the label image corresponding to each of the sample UAV remote sensing images are combined into a data set.

[0027] Optionally, before using the data set to train the initial crop recognition model, the crop recognition method based on UAV remote sensing images further includes: performing data enhancement on the data set to obtain an enhanced data set, and using the enhanced data set as a new data set;

[0028] The data set is enhanced, specifically including:

[0029] Image enhancement is performed on each of the sample UAV remote sensing images and each of the label images in the data set respectively; the image enhancement includes image translation and image rotation.

[0030] Optionally, when the initial crop recognition model is trained using the data set, the optimizer adopts a stochastic gradient descent optimizer.

[0031] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0032] The present application provides a method for crop recognition based on UAV remote sensing images. A trained crop recognition model is designed and trained using an improved DeepLabV3+ model. The improved DeepLabV3+ model is obtained by replacing the backbone structure in the DeepLab V3+ model with a MobileNetV2 model with an attention mechanism to improve the DeepLab V3+ model. After obtaining the UAV remote sensing images obtained by shooting crops, the UAV remote sensing images are directly used as input, and the trained crop recognition model is used for recognition to obtain crop recognition results. The crop recognition results include the types of crops at various locations in the UAV remote sensing images. Simulations have shown that, compared with the DeepLab V3+ model, the improved DeepLab V3+ model is used to identify crops with higher recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 A diagram of the application environment of a method for crop identification based on drone remote sensing images provided in Example 1 of the present application.

[0035] Figure 2 A flowchart of a method for crop identification based on UAV remote sensing images provided in Example 1 of the present application.

[0036] Figure 3 A schematic diagram of the technical route of a method for crop identification based on drone remote sensing images provided in Example 1 of the present application.

[0037] Figure 4 A schematic diagram of the format conversion process provided in Example 1 of the present application.

[0038] Figure 5 A comparison diagram of the UAV remote sensing image A provided in Example 1 of the present application; Figure 5 (a) is the UAV remote sensing image A; Figure 5 (b) The crop recognition results obtained by using the improved DeepLab V3+ model to identify UAV remote sensing image A; Figure 5 (c) The crop recognition result obtained by using the DeepLab V3+ model to identify the UAV remote sensing image A.

[0039] Figure 6A comparison diagram of the UAV remote sensing image B provided in Example 1 of the present application; Figure 6 (a) is the UAV remote sensing image B; Figure 6 (b) The crop recognition results obtained by using the improved DeepLab V3+ model to identify UAV remote sensing image B; Figure 6 (c) The crop recognition result obtained by using the DeepLab V3+ model to identify the UAV remote sensing image B.

[0040] Figure 7 A comparison diagram of the UAV remote sensing image C provided in Example 1 of the present application; Figure 7 (a) is the UAV remote sensing image C; Figure 7 (b) The crop recognition results obtained by using the improved DeepLab V3+ model to identify the UAV remote sensing image C; Figure 7 (c) The crop recognition result obtained by using the DeepLab V3+ model to identify the UAV remote sensing image C. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0042] Example 1

[0043] The crop identification method based on UAV remote sensing images provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be set up separately, integrated on the server, or placed on the cloud or other servers. The terminal can send the unmanned aerial vehicle remote sensing image to be processed to the server. After the server receives the unmanned aerial vehicle remote sensing image to be processed, the server uses the unmanned aerial vehicle remote sensing image as input for the unmanned aerial vehicle remote sensing image to be processed, and uses the trained crop recognition model to perform recognition to obtain the crop recognition result. The server can feed back the obtained crop recognition result for the unmanned aerial vehicle remote sensing image to the terminal.

[0044] In addition, in some embodiments, the crop identification method based on UAV remote sensing images can also be implemented independently by a server or a terminal. For example, the terminal can directly identify the UAV remote sensing images to be processed, or the server can obtain the UAV remote sensing images to be processed from the data storage system and identify the UAV remote sensing images to be processed.

[0045] The terminal may be, but is not limited to, various desktop computers, laptops, smart phones, tablet computers, IoT devices, and portable wearable devices. The server may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0046] like Figure 2 As shown, this embodiment provides a method for crop identification based on drone remote sensing images. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 Taking the server in as an example, the crop recognition method based on drone remote sensing images includes:

[0047] S1: Obtain drone remote sensing images of crops.

[0048] S2: Using the UAV remote sensing image as input, using the trained crop recognition model to perform recognition and obtain a crop recognition result; the trained crop recognition model adopts an improved DeepLab V3+ model, and the improved DeepLab V3+ model is obtained by replacing the backbone structure in the DeepLabV3+ model with a MobileNetV2 model with an attention mechanism to improve the DeepLab V3+ model; the crop recognition result includes the crop type at each location in the UAV remote sensing image.

[0049] By implementing the above-mentioned steps S1 to S2, this embodiment improves the existing DeepLab V3+ model, specifically replaces the backbone structure of the DeepLab V3+ model, and adds an attention mechanism to achieve the expected effect. The improved DeepLabV3+ model is used to identify UAV remote sensing images, which can improve the recognition accuracy and more accurately extract the complex structure of crops, providing strong support for future agricultural structural adjustment work and an important basis for agricultural development.

[0050] The DeepLab V3+ model includes an encoder and a decoder.

[0051] The encoder includes a backbone structure and an atrous convolution structure (ASPP). The backbone structure can be Xception. The features output by the middle convolution layer of the backbone structure are fed into the decoder, and the features output by the last convolution layer of the backbone structure are fed into the atrous convolution structure. The atrous convolution structure includes a 1*1 convolution block connected in parallel, three 3*3 atrous convolution blocks with different expansion rates (the expansion rates can be 6, 12, and 18, respectively), and a global average pooling block. The five modules connected in parallel are then connected to a concat layer and a 1*1 convolution block.

[0052] The decoder includes a 1*1 convolution block, a concat layer, a 3*3 convolution block and an upsampling layer connected in sequence. The other input end of the concat layer is also connected to another upsampling layer. The features output by the middle convolution layer of the backbone structure are subjected to channel dimension reduction through a 1*1 convolution block, and the features output by the hole convolution structure are subjected to interpolation upsampling through an upsampling layer. The features obtained by channel dimension reduction and the features obtained by interpolation upsampling are concatenated through a concat layer, and then convolved through a 3*3 convolution block in sequence, and interpolated upsampling is performed through an upsampling layer.

[0053] This embodiment uses a MobileNetV2 model with an attention mechanism to replace the backbone structure in the DeepLab V3+ model to improve the DeepLabV3+ model and obtain an improved DeepLab V3+ model. The MobileNetV2 model with an attention mechanism includes: a MobileNet V2 model and an attention mechanism connected in sequence, and the attention mechanism can be a CA attention mechanism (Coordinate attention).

[0054] The MobileNetV2 model consists of an input layer, a residual block layer, and a classification layer connected sequentially.

[0055] The input layer includes a first convolution block, which includes a 3*3 two-dimensional convolution layer, a two-dimensional batch normalization layer and a ReLU6 activation function layer connected in sequence.

[0056] The residual block layer includes a second convolution block, several third convolution blocks and a fourth convolution block connected in sequence. The second convolution block includes a 3*3 two-dimensional convolution layer, a two-dimensional batch normalization layer, a ReLU6 activation function layer, a 1*1 two-dimensional convolution layer and a two-dimensional batch normalization layer connected in sequence. The third convolution block includes a 1*1 two-dimensional convolution layer, a two-dimensional batch normalization layer, a ReLU6 activation function layer, a 3*3 two-dimensional convolution layer, a two-dimensional batch normalization layer, a ReLU6 activation function layer, a 1*1 two-dimensional convolution layer and a two-dimensional batch normalization layer connected in sequence, and the number of the third convolution blocks can be 15. The fourth convolution block includes a two-dimensional batch normalization layer, a ReLU6 activation function layer, a 3*3 two-dimensional convolution layer, a two-dimensional batch normalization layer, a ReLU6 activation function layer, a 1*1 two-dimensional convolution layer and a two-dimensional batch normalization layer connected in sequence.

[0057] The classification layer is used to map the number of channels to a higher dimension through a 1*1 convolution operation, and then adjust the feature representation through batch normalization and ReLU6 activation function to better adapt to subsequent model learning and classification tasks. The classification layer includes a fifth convolution block, which includes a 1*1 two-dimensional convolution layer, a two-dimensional batch normalization layer, and a ReLU6 activation function layer connected in sequence.

[0058] In this embodiment, the MobileNetV2 model introduces a new network structure that flips the standard residual structure. The structure uses a lightweight 1*1 convolution to expand the input, followed by a 3*3 depth-separable convolution, and finally a 1*1 convolution for compression. This design helps to reduce computational costs while maintaining feature expression capabilities. The CA attention mechanism encodes precise location information into the neural network to model channel relationships and long-term dependencies. It consists of two steps: coordinate information embedding and coordinate attention generation. In the coordinate information embedding step, the location information is embedded into the input feature map. In the coordinate attention generation step, the attention map is generated using the location information and applied to the input feature map to emphasize the representation of interest.

[0059] In this embodiment, the features output by the last two-dimensional convolutional layer of the last third convolutional block of the residual block layer are sent to the decoder as the features output by the middle convolutional layer of the backbone structure, and the features output by the two-dimensional convolutional layer of the fifth convolutional layer of the classification layer are sent to the dilated convolutional structure as the features output by the last convolutional layer of the backbone structure. Finally, two features are obtained from the MobileNetV2 model with an attention mechanism. The shallow features (i.e., the features output by the middle convolutional layer of the backbone structure) directly enter the decoder, and the deep features (i.e., the features output by the last convolutional layer of the backbone structure) enter the dilated convolutional structure.

[0060] Before using the UAV remote sensing image as input and using the trained crop recognition model to identify the crop, and obtaining the crop recognition result, Figure 3 As shown, the crop identification method based on drone remote sensing images of this embodiment also includes:

[0061] (1) Obtaining a data set, where the data set includes a plurality of sample UAV remote sensing images and a label image corresponding to each sample UAV remote sensing image. The label image is obtained by annotating the crop type at each location in the sample UAV remote sensing image.

[0062] Get the dataset, including:

[0063] (1.1) Obtain original UAV remote sensing images.

[0064] Use UAV remote sensing technology to photograph crops and obtain original UAV remote sensing images.

[0065] (1.2) The original UAV remote sensing image is cropped multiple times to obtain multiple sample UAV remote sensing images.

[0066] For the original UAV remote sensing image, this embodiment uses a cropping tool to crop it multiple times, sets the cropping size to 512*512, and a total of 370 images (i.e., sample UAV remote sensing images) are cropped, and the size of each sample UAV remote sensing image is 512*512. This embodiment uses the PIL function to crop the original UAV remote sensing image. First, the original UAV remote sensing image must be referenced. Specifically, the open(file) method of Image can be used to return the opened original UAV remote sensing image, and then the crop((x0, y0, x1, y1)) method can be used to crop the original UAV remote sensing image. x0, y0, x1, y1 are used to define the cropping range, where x0 is the x coordinate of the upper left corner position point of the original UAV remote sensing image, y0 is the y coordinate of the upper left corner position point of the original UAV remote sensing image, x1 is the x coordinate of the upper left corner position point after 512 pixels are translated to the right, and y1 is the y coordinate of the upper left corner position point after 512 pixels are translated downward.

[0067] (1.3) For each sample UAV remote sensing image, the sample UAV remote sensing image is labeled to obtain the label image corresponding to the sample UAV remote sensing image, and all the sample UAV remote sensing images and the label image corresponding to each sample UAV remote sensing image are combined into a data set.

[0068] Label the cropped sample drone remote sensing images. Specifically, labelme software can be used to label. Select representative crop types, such as rice, corn, peanuts, soybeans, etc., as needed. Label the locations with crops (types are the representative crop types selected above) in the sample drone remote sensing images, and mark the crop types of the locations. Label the locations without crops in the sample drone remote sensing images as background. After labeling the sample drone remote sensing images, the labeling file obtained is a json file. At this time, the labeled json file needs to be further converted into a png image format to obtain a labeled image.

[0069] like Figure 4 As shown, the process of converting a json file to a png picture format includes: first extracting the json file information, then extracting the connected domain, then filling the connected domain, and finally writing to png until the end.

[0070] (2) Construct an initial crop recognition model, which uses the improved DeepLab V3+ model.

[0071] The structures of the initial crop recognition model and the trained crop recognition model are exactly the same, only the network parameters are different.

[0072] (3) Use the data set to train the initial crop recognition model to obtain a trained crop recognition model.

[0073] Before using the data set to train the initial crop recognition model, the crop recognition method based on UAV remote sensing images in this embodiment further includes: performing data enhancement on the data set to obtain an enhanced data set, and using the enhanced data set as a new data set.

[0074] Among them, data enhancement is performed on the data set, specifically including: image enhancement is performed on each sample drone remote sensing image and each label image in the data set, and image enhancement includes image translation and image rotation.

[0075] When performing image enhancement on sample drone remote sensing images and labeled images, the image enhancement methods used include:

[0076] (1) Image translation

[0077] The images (sample drone remote sensing images and label images) are translated. The translation formula is as follows:

[0078]

[0079] In formula (1), x′ and y′ are the x-coordinate and y-coordinate of the pixel point A in the translated image, respectively; Δx and Δy are the customized translation amounts in the x-direction and y-direction, respectively; and x and y are the x-coordinate and y-coordinate of the pixel point A in the original image, respectively.

[0080] The above formula (1) can also be expressed as:

[0081]

[0082] (2) Image rotation

[0083] The images (sample drone remote sensing images and label images) are rotated. The rotation formula is as follows:

[0084]

[0085] In formula (3), x′ and y′ are the x-coordinate and y-coordinate of the pixel point A in the rotated image, respectively; θ is the custom rotation angle, cosθ and sinθ are the custom rotation amounts in the x-direction and y-direction, respectively; x and y are the x-coordinate and y-coordinate of the pixel point A in the original image, respectively.

[0086] The above formula (3) can also be expressed as:

[0087]

[0088] The enhanced data set (including sample drone remote sensing images and label images) is loaded into the improved DeepLabV3+ model for training. During the training process, the GPU type used in this embodiment is NVIDIAGeForce RTX3090, CUDA (a parallel computing framework for completing complex parallel computing) version is 11.3, the code used is based on the torch framework in Python, and the training parameters are set as follows: SGD (Stochastic Gradient Descent) optimizer is selected as the optimizer, which can adaptively adjust the learning rate during the training process, the initial learning rate is set to 0.001, batch size (batch size, refers to the number of samples selected for one training) is 10, downsample_factor (downsampling multiples) is 8, the iteration cycle is 80 times, and num_workers (the number of threads for loading data) is 4. In this embodiment, when the initial crop recognition model is trained using the data set, the optimizer uses a stochastic gradient descent optimizer.

[0089] This embodiment aims at the agricultural remote sensing field and proposes a crop identification method. By adopting this complete set of technical processes of crop identification methods, practical problems in scientific research and production are solved. The specific process includes: preprocessing the original UAV remote sensing image, specifically cropping the original UAV remote sensing image, setting the size to 512*512, obtaining a sample UAV remote sensing image, and further labeling the sample UAV remote sensing image using labelme software. During the labeling process, representative crop types are selected to obtain a data set, and then data enhancement operations are performed on the data set. The data enhancement operations include basic operations such as translation and rotation to obtain an enhanced data set. The enhanced data set is divided into a training set and a verification set in a ratio of 7:3 for training and verification models to obtain a trained crop identification model. The trained crop identification model can be used to identify UAV remote sensing images to obtain crop identification results.

[0090] In terms of the model, this embodiment improves the DeepLab V3+ model and uses the improved DeepLab V3+ model to identify crops. The recognition results of the improved DeepLab V3+ model and the original DeepLab V3+ model are further compared and analyzed to evaluate the accuracy improvement of the improved DeepLab V3+ model. Figure 5-Figure 7 As shown, Figure 5 (a) Figure 6 (a) and Figure 7 (a) are all UAV remote sensing images. Figure 5 (b) Figure 6 (b) and Figure 7 (b) are the crop recognition results obtained by using the improved DeepLabV3+ model to identify UAV remote sensing images. Figure 5 (c) Figure 6 (c) and Figure 7 (c) are all crop recognition results obtained by using the DeepLab V3+ model to recognize UAV remote sensing images. It can be seen that compared with the existing DeepLab V3+ model, the recognition accuracy of this embodiment is higher, and it can complete the accurate recognition of UAV remote sensing images, and the computer resource occupation is lower. The advantages are as follows:

[0091] (1) High recognition accuracy: The improved DeepLab V3+ model with the backbone structure of MobileNetV2 model and CA attention mechanism is superior to the original DeepLabV3+ model in all aspects. The simulation results show that when the backbone structure is MobileNetV2 model with CA attention mechanism, compared with the original DeepLab V3+ model, the MIoU (mean intersection over union) of the improved DeepLabV3+ model increased by 1.04%, MPA (mean pixel accuracy) increased by 2.37%, Recall increased by 2.37%, Precision increased by 0.05%, F1-Score increased by 1%, the classification accuracy of corn increased by 0.09%, the classification accuracy of walnut increased by 0.51%, the classification accuracy of soybean increased by 1.75%, and the classification accuracy of ginger increased by 3.50%.

[0092] (2) Accurate recognition of UAV remote sensing images: The core of this embodiment is to accurately extract and identify the planting structure of crops. It is obvious that the improved DeepLab V3+ model has clearer boundaries, a wider and more accurate extraction range in crop extraction compared to the original DeepLab V3+ model.

[0093] (3) Low computer resource usage: Similar to the MobileNet V1 model, the MobileNet V2 model also uses two hyperparameters, width multiplier and resolution multiplier, to adjust the model size and computational complexity, allowing users to make trade-offs based on device performance and application scenarios, balancing the accuracy and speed of the model. Compared with the MobileNetV1 model, the MobileNetV2 model further improves performance while maintaining lightweight, low latency, and low computational resource usage.

[0094] The embodiment of the present application also provides an application scenario, which applies the above-mentioned crop identification method based on drone remote sensing images. Specifically, the crop identification method based on drone remote sensing images provided in this embodiment can be applied in the crop identification scenario. The crop identification scenario includes an image acquisition link and an image recognition link. The drone remote sensing image is acquired through the image acquisition link, and the drone remote sensing image is recognized through the image recognition link to obtain the crop identification result. The crop identification method based on drone remote sensing images provided in this embodiment belongs to the image recognition link.

[0095] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for crop identification based on UAV remote sensing images, characterized in that: The crop identification method based on UAV remote sensing images includes: Obtaining drone remote sensing images of crops; The UAV remote sensing image is used as input, and the trained crop recognition model is used for recognition to obtain a crop recognition result; the trained crop recognition model adopts the improved DeepLab V3+ model, and the improved DeepLab V3+ model is obtained by replacing the backbone structure in the DeepLab V3+ model with the MobileNetV2 model with an attention mechanism to improve the DeepLabV3+ model; the crop recognition result includes the crop type at each location in the UAV remote sensing image.

2. The crop identification method based on UAV remote sensing images according to claim 1 is characterized in that: The MobileNetV2 model with an attention mechanism includes: a MobileNetV2 model and an attention mechanism connected in sequence, wherein the attention mechanism is a CA attention mechanism.

3. The crop identification method based on UAV remote sensing images according to claim 2 is characterized in that: The MobileNetV2 model consists of an input layer, a residual block layer, and a classification layer connected sequentially; The input layer includes a first convolutional block; The residual block layer includes a second convolution block, a plurality of third convolution blocks and a fourth convolution block connected in sequence; The classification layer includes a fifth convolutional block.

4. The crop identification method based on UAV remote sensing images according to claim 3 is characterized in that: The first convolution block includes a convolution layer, a batch normalization layer, and an activation function layer connected in sequence; The second convolution block includes a convolution layer, a batch normalization layer, an activation function layer, a convolution layer and a batch normalization layer connected in sequence; The third convolution block includes a convolution layer, a batch normalization layer, an activation function layer, a convolution layer, a batch normalization layer, an activation function layer, a convolution layer and a batch normalization layer connected in sequence; The fourth convolution block includes a batch normalization layer, an activation function layer, a convolution layer, a batch normalization layer, an activation function layer, a convolution layer and a batch normalization layer connected in sequence; The fifth convolution block includes a convolution layer, a batch normalization layer, and an activation function layer connected in sequence.

5. The crop identification method based on UAV remote sensing images according to claim 1 is characterized in that: Before using the UAV remote sensing image as input and using the trained crop recognition model to perform recognition to obtain the crop recognition result, the crop recognition method based on UAV remote sensing image further includes: Acquire a data set; the data set includes a plurality of sample UAV remote sensing images and a label image corresponding to each of the sample UAV remote sensing images, wherein the label image is obtained by marking the crop type at each location in the sample UAV remote sensing image; Constructing an initial crop recognition model; the initial crop recognition model adopts an improved DeepLab V3+ model; The initial crop recognition model is trained using the data set to obtain a trained crop recognition model.

6. The crop identification method based on UAV remote sensing images according to claim 5 is characterized in that: Get the dataset, including: Obtain original drone remote sensing images; The original UAV remote sensing image is cropped multiple times to obtain multiple sample UAV remote sensing images; For each of the sample UAV remote sensing images, the sample UAV remote sensing image is annotated to obtain a label image corresponding to the sample UAV remote sensing image; and all the sample UAV remote sensing images and the label image corresponding to each of the sample UAV remote sensing images are combined into a data set.

7. The crop identification method based on UAV remote sensing images according to claim 5 is characterized in that: Before using the data set to train the initial crop recognition model, the crop recognition method based on UAV remote sensing images further includes: performing data enhancement on the data set to obtain an enhanced data set, and using the enhanced data set as a new data set; The data set is enhanced, specifically including: Image enhancement is performed on each of the sample UAV remote sensing images and each of the label images in the data set respectively; the image enhancement includes image translation and image rotation.

8. The crop identification method based on UAV remote sensing images according to claim 5 is characterized in that: When the initial crop recognition model is trained using the data set, the optimizer adopts a stochastic gradient descent optimizer.

Citation Information

Patent Citations

  • Crop remote sensing image semantic segmentation method based on parallel attention mechanism

    CN113887517A

  • Crop planting information extraction method based on convolutional neural network

    CN115331104A

  • Field structure remote sensing image multi-classification identification method

    CN116310876A

  • High-efficiency and high-precision remote sensing image semantic segmentation method and application

    CN116863347A

  • Crop remote sensing recognition model and crop remote sensing recognition method

    CN117197648A

Cited By

  • Operation environment key element identification method

    CN119851021A

  • A method for identifying key elements of the work environment

    CN119851021B

  • Crop planting area identification method and system based on cross-time-phase feature coupling

    CN120147869A