A high-speed unmanned aerial vehicle image motion deblurring identification method and system based on a DeblurGAN network

Abstract

The application discloses a kind of based on DeblurGAN network's high-speed unmanned plane image motion blur identification method, system, the method includes collecting the motion clear image and motion blur image of high-speed unmanned plane, motion clear image is artificially simulated, and the simulated motion blur image is obtained;Using motion clear image and simulated motion blur image preprocessed DeblurGAN network model is trained, and motion blur network is obtained, and motion blur image is input into motion blur network, and the final motion blur after removing image is obtained;Using adaptive edge extraction weight technology is trained and optimized to edge convolutional neural network model, and the optimized edge convolutional neural network model is obtained, and the final motion blur after removing image is input into the model, and the classification result of unmanned plane is obtained.The application effectively solves the motion blur problem in high-speed unmanned plane image, improves the recognition and tracking ability of unmanned plane.

Description

Technical Field

[0001] This invention relates to the fields of image processing and machine learning technology, specifically to a method and system for de-motion blurring of high-speed UAV images based on a DeblurGAN network. Background Technology

[0002] With the rapid development of drone technology, the use of LFS-UAVs (Low Fast and Small Unmanned Aerial Vehicles) has become increasingly common and its application scope is expanding. However, the widespread use of drones has also brought a series of safety hazards and management challenges. Improper use of drones may pose a threat to public safety, and may even lead to unauthorized entry into "no-fly" zones, collisions with birds, aircraft, and other aerial vehicles, causing serious safety accidents, disrupting air traffic control, and endangering air traffic safety.

[0003] Currently, drone identification and tracking primarily rely on technologies such as radar and radio signal detection. However, these methods have limitations in practical applications. For example, radar equipment is expensive and has blind spots when detecting at close range; radio signal detection can be affected by environmental interference, impacting identification accuracy. Therefore, drone identification technology based on video images has become an important research direction. However, in high-speed motion scenarios, drone images are easily affected by motion blur, leading to image quality degradation and consequently affecting drone identification and tracking. Traditional deblurring methods often perform poorly when dealing with high-speed motion blur, failing to meet practical application requirements. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for de-motion blurring of high-speed UAV images based on the DeblurGAN network. By utilizing advanced image processing and machine learning technologies, the motion blur problem in high-speed UAV images can be effectively solved, thereby improving the recognition and tracking capabilities of UAVs.

[0005] The present invention adopts the following technical solution:

[0006] A method for motion blur removal in high-speed UAV images based on a DeblurGAN network includes the following steps:

[0007] S1. Collect clear motion images and blurred motion images of high-speed UAVs. Based on the blurred motion images, artificially simulate the clear motion images to obtain simulated blurred motion images. The clear motion images and simulated blurred motion images constitute a dataset of motion images.

[0008] S2. Preprocess the DeblurGAN network model using the clear motion image. Train the preprocessed DeblurGAN network model using the dataset in step S1 to obtain the deblurred network. After background ablation and normalization processing of the motion-blurred image in step S1, input it into the deblurred network to obtain the final deblurred image. Evaluate the image using YOLOv5s.

[0009] S3. Construct an edge convolutional neural network model using deep learning and computer image processing techniques; based on the final de-motion blurred image, train and optimize the model using adaptive edge extraction weights to obtain the optimized edge convolutional neural network model.

[0010] S4. The final de-motion blurred image is input into the optimized edge convolutional neural network model. After feature extraction, feature fusion and classification processing, the classification result of the UAV is obtained, and the UAV is recognized.

[0011] Furthermore, in step S1, the dataset constituting the motion-blurred image includes the following:

[0012] S101. Acquire clear and blurred motion images of high-speed UAVs under different background, lighting, and weather conditions, and perform random rotation, scaling, and cropping transformations on the clear motion images.

[0013] S102. Based on image processing technology, a computer script is used to randomly select the transformed, clear motion image from step S101 as the foreground image. The appropriate blur kernel size is selected based on the height of the foreground image. The specific selection process is as follows:

[0014] When the height of the foreground image is less than or equal to 15 pixels, the blur kernel size is randomly selected between 2 and 4; when the height of the foreground image is greater than 15 pixels and less than or equal to 25 pixels, the blur kernel size is randomly selected between 2 and 8 pixels; when the height of the foreground image is greater than 25 pixels and less than or equal to 150 pixels, the blur kernel size is randomly selected between 5 and 20 pixels; when the height of the foreground image is greater than 150 pixels, the blur kernel size is randomly selected between 5 and 50 pixels.

[0015] S103. Randomly overlay the image corresponding to the blur kernel onto the background image in a vertical, horizontal, or mixed vertical and horizontal manner to obtain a simulated motion-blurred image.

[0016] Furthermore, in step S2, the image obtained after motion blur removal includes the following:

[0017] S201. After background ablation of the motion-clear image in step S1, the foreground is obtained. The foreground and the original background are normalized and then input into the input port of the DeblurGAN network model to obtain the preprocessed DeblurGAN network model.

[0018] S202. Select a set number of images from the dataset in step S1 as the training set. Input the training set into the preprocessed DeblurGAN network model. Extract the feature map of the image in the conv3*3 layer using the VGG-19 model. Calculate the difference between the feature maps of the simulated motion-blurred image and the motion-clear image. Based on this difference, iteratively optimize the generator using content loss and adversarial loss until the set maximum number of iterations is reached. The optimized generator and the discriminator constitute the de-motion-blurring network.

[0019] S203. Perform background ablation on the acquired motion-blurred image to obtain the foreground moving object, moving target, and background. Normalize the moving target and background. Use a de-blurring network to process the processed motion-blurred image to obtain a de-blurred image. Perform brightness, contrast, and color balance processing on the image to obtain the final de-blurred image.

[0020] Furthermore, in step S3, constructing the optimized edge convolutional neural network model includes the following:

[0021] S301, the edge convolutional neural network model includes an input layer, three convolutional layers, two fully connected layers and an output layer connected in sequence, wherein the input layer is designed with a shape of 64*64*4.

[0022] S302. Based on image processing technology, the weights of the final de-motion-blurred image are extracted using two different scale levels of edge detection, and a 3x3 Gaussian kernel is used for smoothing. The Canny function in the OpenCV open-source library is used for edge detection. Then, a 7x7 Gaussian kernel is used for smoothing, and the Canny function is used for edge detection. The results of the two edge detections are then weighted and summed.

[0023] S303. The non-maximum suppression and double threshold methods in the OpenCV open-source library are called to process the weighted summation edge detection results to obtain the final edge abstract information.

[0024] S304. Use edge detection methods to perform edge detection on the drone in the final de-motion blurred image, and use the final edge abstract information to train and optimize the edge convolutional neural network model to obtain the optimized edge convolutional neural network model.

[0025] Furthermore, in step S4, the classification results of the drones include the following:

[0026] The final de-blurred image is divided into training and testing sets in a 4:1 ratio. The training set is input into the optimized edge convolutional neural network model for iterative training until the maximum number of iterations is reached, resulting in the trained model. The testing set is resized, converted to a tensor, and normalized. The processed testing set and the final edge abstraction information are then input into the trained model. The processed testing set passes through the first three channels of the model's input layer, and the final edge abstraction information passes through the fourth channel. In the first convolutional layer, 19 19x19 filters are used for convolution to extract primary features. The system firstly outputs 19 first feature maps. When these first feature maps pass through the second convolutional layer, they undergo convolution with 22 15x15 filters, outputting 22 second feature maps. When these second feature maps pass through the third convolutional layer, they undergo convolution with 8 9x9 filters, outputting 8 third feature maps. The third feature maps are then non-linearly transformed using the ReLU activation function and input into the first fully connected layer for linear transformation, outputting a set number of nodes. These nodes undergo linear transformation again in the second fully connected layer, outputting the final node. Stochastic gradient descent is used as the optimizer, with learning rate and momentum parameters set, and loss and backpropagation performed to obtain the classification results for the UAV.

[0027] Furthermore, this invention also proposes a high-speed UAV image de-motion blur recognition system based on a DeblurGAN network, comprising:

[0028] The dataset construction module is used to collect clear motion images and blurred motion images of high-speed UAVs. Based on the blurred motion images, the clear motion images are artificially simulated to obtain simulated blurred motion images. The clear motion images and the simulated blurred motion images constitute the dataset of motion images.

[0029] The image deblurring module is used to preprocess the DeblurGAN network model using the motion-clear image. The preprocessed DeblurGAN network model is trained using the dataset in step S1 to obtain the deblurring network. The motion-blurred image in step S1 is processed by background ablation and normalization and then input into the deblurring network to obtain the final deblurred image. The image is then evaluated using YOLOv5s.

[0030] The model optimization module is used to construct an edge convolutional neural network model using deep learning and computer image processing techniques. Based on the final de-motion blurred image, the model is trained and optimized using adaptive edge extraction weights to obtain the optimized edge convolutional neural network model.

[0031] The recognition module is used to input the final de-motion-blurred image into the optimized edge convolutional neural network model. After feature extraction, feature fusion and classification processing, the classification result of the UAV is obtained, thus realizing the recognition of the UAV.

[0032] Furthermore, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the high-speed UAV image de-motion blur recognition method based on the DeblurGAN network.

[0033] Furthermore, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the aforementioned method for de-motion blur recognition of high-speed UAV images based on a DeblurGAN network.

[0034] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0035] 1. This invention simplifies the image processing workflow and improves processing efficiency and accuracy.

[0036] 2. This invention can effectively handle different types of fuzziness, including local fuzziness and global fuzziness, and has better adaptability and robustness.

[0037] 3. This invention addresses the problem of motion blur caused by high-speed drones during flight by proposing a method of first removing the blur before classification and recognition. This motion blur removal significantly improves the recognition rate of drone images.

[0038] 4. In this invention, edge information is incorporated into the UAV detection and classification network for fusion. Edge information plays an important guiding role in the details and structure of the image, which can help the network better capture and restore the key features in the image, thereby improving the accuracy and reliability of recognition.

[0039] 5. This invention is not only applicable to the identification of high-speed drones, but can also be extended to the identification of blurred images of other high-speed moving objects, and has wide applicability and flexibility. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the overall implementation of the present invention.

[0041] Figure 2 These are optical image samples collected in embodiments of the present invention.

[0042] Figure 3 This is the preprocessed DeblurGAN network model of this invention.

[0043] Figure 4 This is a diagram illustrating the recognition effects of clear images, blurred images, and restored images in optical images in an embodiment of the present invention.

[0044] Figure 5 This is an edge detection result diagram of an embodiment of the present invention.

[0045] Figure 6 This is a diagram illustrating the image input process of the input layer in the optimized edge convolutional neural network model of this invention.

[0046] Figure 7 This is a diagram showing the restoration effect of the optical image in an embodiment of the present invention.

[0047] Figure 8 This is a comparison chart of drone images and classification accuracy after deblurring using different networks in this embodiment of the invention. Detailed Implementation

[0048] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0049] To achieve the above objectives, this invention proposes a method for de-motion blur recognition of high-speed UAV images based on a DeblurGAN network, such as... Figure 1 As shown, the specific steps are as follows:

[0050] S1. Acquire clear motion images of a high-speed drone (racing drone) during low-speed flight, images with out-of-focus lenses, and motion-blurred images caused by the high speed of the high-speed drone (racing drone). Based on the motion-blurred images of the high-speed drone, artificially simulate the clear motion images to obtain simulated motion-blurred images. The clear motion images and the simulated motion-blurred images constitute a dataset of motion images. Specific content includes:

[0051] S101. Collect clear and blurred images of high-speed drones under different background, lighting, and weather conditions. Perform random rotation, scaling, and cropping on the clear images to simulate shooting effects at different angles and distances for data augmentation.

[0052] S102. Based on image processing technology, a computer script is used to randomly select the transformed, clear motion image from step S101 as the foreground image. An appropriate blur kernel size is selected based on the height of the foreground image. The specific selection process is as follows:

[0053] When the height of the foreground image is less than or equal to 15 pixels, the blur kernel size is randomly selected between 2 and 4; when the height of the foreground image is greater than 15 and less than 25 pixels, the blur kernel size is randomly selected between 2 and 8; when the height of the foreground image is greater than 25 and less than 150 pixels, the blur kernel size is randomly selected between 5 and 20; when the height of the foreground image is greater than 150 pixels, the blur kernel size is randomly selected between 5 and 50.

[0054] S103. Randomly overlay the image corresponding to the blur kernel onto the background image in a vertical, horizontal, or mixed manner (horizontal and vertical superposition) to achieve artificially synthesized motion blur effects of different degrees, and obtain simulated motion blur images.

[0055] In this embodiment, the motion image dataset contains 500 sets of sharp motion images and simulated motion-blurred images. For example... Figure 2 As shown, Figure 2 (a) is the optical image in the dataset without a blur kernel. Figure 2 (b) is Figure 2 (a) corresponds to the optical image with the added blur kernel. Figure 2 (c) and Figure 2 (d) are all drones randomly selected from the original image.

[0056] S2. Preprocess the DeblurGAN network model using clear motion images. Train the preprocessed DeblurGAN network model using the dataset from step S1 to obtain a deblurred network. Perform background ablation and normalization on the motion-blurred images acquired in step S1, then input them into the deblurred network to obtain the final deblurred image. Evaluate the image using YOLOv5. The specific steps are as follows:

[0057] This invention uses the DeblurGAN network model as the foundation for image deblurring. Specifically, the source code of the DeblurGAN network model can be downloaded from the [KupynOrest] / [DeblurGAN] repository on GitHub. This model is based on the deblurring algorithm proposed by Goroshin et al. in 2018. This invention fine-tunes this network model; specifically, as follows... Figure 3As shown, since the network performs downsampling on high-resolution images, in order to preserve the blurred details of the drone, the foreground is obtained by background ablation of the clear motion image in step S1. The foreground and the original background are normalized and then input into the input port of the DeblurGAN network model to avoid the loss of details caused by downsampling, thus obtaining the preprocessed DeblurGAN network model.

[0058] 400 images from the dataset in step S1 are selected as the training set. The training set is input into the preprocessed DeblurGAN network model. The VGG-19 model is used to extract the feature maps of the images at the conv3*3 layer, and the difference between the feature maps of the simulated motion-blurred image and the motion-clear image is calculated. Based on this difference, the generator is iteratively optimized using content loss and adversarial loss until it stops after 200 rounds, resulting in an optimized generator. The optimized generator and discriminator constitute the de-blurring network. Background ablation is performed on the acquired motion-blurred image to obtain the foreground moving object, moving target, and background. The moving target and background are normalized, and the de-blurring network is used to process the processed motion-blurred image to obtain the de-blurred image. Brightness, contrast, and color balance are processed on the image to further improve the visual effect of the image, resulting in the final de-blurred image. Compared with the input blurred image, this image has clearer edges and details.

[0059] Using YOLOv5s, UAV target detection was performed on 100 sets of simulated motion-blurred images, clear images, and the final de-blurred images. The ratio of correctly identified targets to the total number of targets was calculated. Experiments show that the DeblurGAN network model can significantly improve the target recognition rate of blurred images. Specifically: the recognition rate of blurred images was between 50% and 75%, with an average of 65%; the recognition rate of clear images was between 90% and 95%, with an average of 92%; and the recognition rate of the de-blurred images was between 85% and 95%, with an average of 88%. In high-speed UAV deblurring, it significantly improved the recognition rate of blurred images, with an average improvement of 23%, and the recognition rate of the de-blurred images was close to that of clear images, differing by only 4%. Figure 4 As shown, Figure 4 (a) shows the test results of the original image of the UAV. Figure 4 (b) shows the test results of the UAV optical image to be restored. Figure 4 (c) shows the test results of the network after restoring the optical image. It can be seen that after the blurred image is restored by the network, the drone recognition rate will be greatly improved, even higher than that of the original image.

[0060] S3. Construct an edge convolutional neural network model using deep learning and computer image processing techniques; based on the final de-motion-blurred image, train and optimize the model using adaptive edge extraction weights to obtain the optimized edge convolutional neural network model; the specific content is as follows:

[0061] S301. This invention designs a convolutional neural network specifically for UAV classification based on the PyTorch neural network library. When building the network model, the necessary data processing and model building libraries are first imported; then, based on the APIs or functions provided in the PyTorch library, the convolutional neural network model is built according to the defined model structure. The edge convolutional neural network model includes an input layer, three convolutional layers, two fully connected layers, and an output layer connected in sequence, wherein the input layer is designed with a shape of 64*64*4.

[0062] S302, such as Figure 5 As shown, based on image processing techniques, two different scale levels are used to extract the weights of the final de-motion-blurred image, and a small Gaussian kernel (3x3) is used for smoothing. Edge detection is performed using the Canny function from the OpenCV open-source library to capture detailed features (propellers, wings, etc.). Figure 5 (a) demonstrates that small-scale edge extraction can effectively capture detailed features such as propellers. A larger Gaussian kernel (7x7) is then used for smoothing, followed by Canny edge detection to capture the main contours. Figure 5 (b) demonstrates that large-scale edge extraction can clearly extract the main outline of the UAV. Finally, the edge results at small and large scales are weighted and summed.

[0063] S303. The OpenCV open-source library's non-maximum suppression and double thresholding methods are used to improve the accuracy and reliability of the edge results after weighted summation, obtaining the final edge abstraction information, such as... Figure 5 As shown in (c).

[0064] S304. Use edge detection methods to perform edge detection on the drone in the final de-motion blurred image, and use the edge detection results to train and optimize the edge convolutional neural network model to obtain the optimized edge convolutional neural network model.

[0065] S4. The final de-motion-blurred image is input into the optimized edge convolutional neural network model. After feature extraction, feature fusion, and classification processing, the classification result of the drone is obtained, realizing drone recognition; the specific content is as follows:

[0066] The final de-blurred image is divided into training and test sets in a 4:1 ratio. The training set is input into the optimized edge convolutional neural network model, with 100 training iterations and a sample size of 64. The forward propagation and loss function use the same parameters as in step S301. After 100 iterations, the test set is processed through predefined transformations, including resizing (64*64), converting to a tensor, and standardization. The processed test set and the final edge abstraction information are then input into the trained and optimized edge convolutional neural network model, as follows: Figure 6 As shown, the processed test set passes through the first three channels of the model's input layer, and the final edge abstraction information passes through the fourth channel of the model's input layer. In the first convolutional layer, 19 19*19 filters are used for convolution to extract primary features, outputting 19 46*46 first feature maps. In the second convolutional layer, 22 15*15 filters are used for convolution to extract more complex features, outputting 22 32*32 second feature maps. In the third convolutional layer, 8 9*9 filters are used for convolution. The product operation extracts higher-level features, outputting eight 24*24 third feature maps. After a non-linear transformation of the third feature maps using the ReLU activation function, they are input into the first fully connected layer, flattened into 4608 nodes, and then linearly transformed to output 80 nodes. When these nodes pass through the second fully connected layer, a linear transformation is performed, outputting 3 nodes. Stochastic gradient descent is used as the optimizer, with a learning rate of 0.01 and a momentum parameter of 0.5. The loss and backpropagation of the network are calculated to obtain the classification results of the drones, namely drones, birds, and other background (negative samples).

[0067] like Figure 7 As shown, Figure 7 (a) is the original image from the drone. Figure 7 (b) is the optical image of the UAV to be restored. Figure 7 (c) shows the effect of the trained network on the restoration of optical images. The shape of the UAV is restored relatively completely and the ghosting is eliminated to a certain extent.

[0068] like Figure 8 The figure shows a comparison of the accuracy of the optimized edge convolutional neural network model and the convolutional neural network without edge information in classifying deblurred drone images, as well as the accuracy of the convolutional neural network in classifying blurred drone images. The experimental results show that the classification results output by the optimized edge convolutional neural network model on the dataset produced in this invention can reach an accuracy of 90% or more, and the classification accuracy of drones is improved by about 2 to 3 percentage points compared with ordinary convolutional neural networks.

[0069] This invention also proposes a high-speed UAV image de-motion blur recognition system based on a DeblurGAN network, including a dataset construction module, an image de-blurring processing module, a model optimization module, a recognition module, and a computer program that can run on a processor. It should be noted that each module in the above system corresponds to a specific step of the method provided in this invention embodiment, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention embodiment.

[0070] This invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. It should be noted that when the processor executes the computer program, it corresponds to the specific steps of the method provided in this invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention.

[0071] This invention also proposes a computer-readable storage medium storing a computer program. It should be noted that when the computer program is executed by a processor, it corresponds to the specific steps of the method provided in this invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention.

[0072] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A high-speed unmanned aerial vehicle image motion deblurring recognition method based on a DeblurGAN network, characterized by, The application relates to a method for recognizing a high-speed unmanned aerial vehicle (UAV) and a device thereof. The method comprises the following steps: S1, collecting a motion clear image and a motion blur image of a high-speed unmanned aerial vehicle, artificially simulating the motion clear image based on the motion blur image to obtain a simulated motion blur image, and forming a data set of the motion image by the motion clear image and the simulated motion blur image; S2, preprocessing a DeblurGAN network model by using the motion clear image, training the preprocessed DeblurGAN network model by using the data set in step S1 to obtain a motion deblurring network, inputting the motion blur image in step S1 into the motion deblurring network after background ablation and normalization processing to obtain a final deblurred image, and evaluating the image by using Yolo5s; The motion deblurring network comprises the following steps: selecting a certain number of images in the data set in step S1 as a training set, inputting the training set into the preprocessed DeblurGAN network model, extracting a feature map of the image at a conv3*3 layer by using a VGG-19 model, calculating the difference between the simulated motion blur image and the motion clear image feature map, and iteratively optimizing a generator based on the difference by using a content loss and an adversarial loss until a maximum iteration number is reached to stop, thereby obtaining an optimized generator, and the optimized generator and a discriminator constitute the motion deblurring network; S3, constructing an edge convolutional neural network model by using deep learning and computer image processing technology; training and optimizing the model by using an adaptive edge extraction weight technology based on the final deblurred image to obtain an optimized edge convolutional neural network model; and the method comprises the following steps: S301, the edge convolutional neural network model comprises an input layer, three convolutional layers, two fully connected layers and an output layer which are sequentially connected, wherein the design shape of the input layer is 64*64*4; S302, based on the image processing technology, the weight of the final deblurred image is extracted by using two different scale levels, and the weight is smoothed by using a 3x3 Gaussian kernel; the edge is detected by using a Canny function in an Opencv open source library; the weight is smoothed by using a 7x7 Gaussian kernel, and the edge is detected by using the Canny function; and the edge detection results after the two times of edge detection are weighted and summed; S303, the edge detection results after the weighting and summing are processed by using a non-maximum suppression and a double-threshold method in the Opencv open source library to obtain final edge abstract information; S304, the UAV in the final deblurred image is edge-detected by using an edge detection method, and the edge convolutional neural network model is trained and optimized by using the final edge abstract information to obtain an optimized edge convolutional neural network model; 2. The method of claim 1, wherein the DeblurGAN network-based high-speed UAV image motion deblurring recognition method is characterized in that, S4, the final deblurred image is input into the optimized edge convolutional neural network model, feature extraction, feature fusion and classification processing are carried out, and a classification result of the UAV is obtained, thereby realizing the recognition of the UAV. In step S1, the data set of the motion blur image comprises the following contents: S101, collect high-speed unmanned aerial vehicle motion clear images and motion blur images under different backgrounds, lighting conditions and weather conditions, and perform random rotation, scaling and cropping transformation on the motion clear images; S102, based on image processing technology, randomly select a motion clear image transformed in step S101 as a foreground image using a computer script, select a corresponding blur kernel size according to the height of the foreground image, and the specific selection process is as follows: When the height of the foreground image is less than or equal to 15, the blur kernel size is randomly selected between 2 and 4; when the height of the foreground image is greater than 15 and less than or equal to 25, the blur kernel size is randomly selected between 2 and 8; when the height of the foreground image is greater than 25 and less than or equal to 150, the blur kernel size is randomly selected between 5 and 20; when the height of the foreground image is greater than 150, the blur kernel size is randomly selected between 5 and 50; S103, randomly cover the image corresponding to the blur kernel to the background image in a vertical, horizontal or mixed manner to obtain a simulated motion blur image.

3. The method of claim 1, wherein the DeblurGAN network-based high-speed UAV image motion deblurring recognition method is characterized in that, In step S2, the image after motion deblurring includes the following contents: S201, after background ablation of the motion clear image in step S1, the foreground is obtained, the foreground and the original background are normalized and input into the input port of the DeblurGAN network model to obtain the pretreated DeblurGAN network model; S202, ablate the background of the collected motion blur image to obtain the foreground motion object, the motion target and the background, normalize the motion target and the background, process the processed motion blur image using a motion deblurring network to obtain an image after motion deblurring, and process the image in brightness, contrast and color balance to obtain a final image after motion deblurring.

4. The method of claim 1, wherein the DeblurGAN network-based high-speed UAV image motion deblurring recognition method is characterized in that, In step S4, the classification result of the unmanned aerial vehicle includes the following contents: The final motion deblurred image is divided according to a ratio of 4:1 to obtain a training set and a test set; the training set is input into the optimized edge convolutional neural network model for training iteration until a set maximum iteration number is reached to stop, and a trained model is obtained; the test set is processed by resizing, converting into Tensor and standardizing, and the processed test set and the final edge abstract information are input into the trained model, wherein the processed test set passes through the first three channels of the input layer of the model, and the final edge abstract information passes through the fourth channel of the input layer of the model; when passing through the first convolutional layer, 19 filters of 19*19 are used for convolution operation to extract primary features and output 19 first feature maps; when the first feature maps pass through the second convolutional layer, 22 filters of 15*15 are used for convolution operation to output 22 second feature maps; when the second feature maps pass through the third convolutional layer, 8 filters of 9*9 are used for convolution operation to output 8 third feature maps; after the third feature maps are subjected to nonlinear transformation by using a ReLU activation function, the third feature maps are input into the first fully connected layer for linear transformation to output a set number of nodes; the nodes are subjected to linear transformation by the second fully connected layer to output final nodes. The random gradient descent is used as an optimizer, learning rate and momentum parameters are set, and loss calculation and network back propagation are performed to obtain the classification result of the unmanned aerial vehicle.

5. The system for identifying the high-speed unmanned aerial vehicle image deblurring method based on the DeblurGAN network of claim 1, wherein The method comprises the following steps: A data set construction module is configured to collect motion clear images and motion blur images of high-speed unmanned aerial vehicles, artificially simulate the motion clear images based on the motion blur images to obtain simulated motion blur images, and construct a data set of the motion images from the motion clear images and the simulated motion blur images. An image deblurring processing module is configured to preprocess a DeblurGAN network model using the motion clear images, train the preprocessed DeblurGAN network model using the data set in step S1 to obtain a motion deblurring network, input the motion blur images in step S1 into the motion deblurring network after background ablation and normalization processing, obtain a final motion deblurred image, and evaluate the image using Yolo5s. A model optimization module is configured to construct an edge convolutional neural network model using deep learning and computer image processing technology, train and optimize the model based on the final motion deblurred image using adaptive edge extraction weight technology, and obtain an optimized edge convolutional neural network model. An identification module is configured to input the final motion deblurred image into the optimized edge convolutional neural network model, perform feature extraction, feature fusion and classification processing, and obtain a classification result of the unmanned aerial vehicle to realize identification of the unmanned aerial vehicle.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method for identifying high-speed unmanned aerial vehicle image motion deblurring based on the DeblurGAN network in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, the computer-readable storage medium being characterized by, The computer program is run by a processor to perform the method for identifying high-speed unmanned aerial vehicle image motion blur based on the DeblurGAN network according to any one of claims 1 to 4.

Images