A method for image motion deblurring of intelligent reconnaissance equipment

Abstract

The present invention relates to the field of image processing technology, and proposes an image motion deblurring method for intelligent reconnaissance equipment, comprising the following steps: collecting a target image by intelligent reconnaissance equipment, sending the target image to an edge computing device for motion deblurring processing, and outputting a generated image; the edge computing device is equipped with a trained semi-supervised generative adversarial network consisting of a generator and a discriminator; wherein the generator includes a feature pyramid network, which is used to perform motion deblurring processing on the input image and output the generated image; the discriminator includes a discriminator network based on a PatchGAN network, which is used to judge whether the generated clear image is true or false; when the discriminator output is true, the generated image output by the generator is output as the image motion deblurring result; when the discriminator output is false, the parameters of the generator are updated according to a preset loss function and the generated image is re-output.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more particularly to an image motion deblurring method for intelligent reconnaissance equipment. Background Art

[0002] As a highly integrated, technologically advanced, and complex high-tech optoelectronic device, the optoelectronic pod utilizes optoelectronic sensors in various wavelength bands, including visible light, infrared, laser, and millimeter waves, along with a precise, stable platform and high-speed onboard control system, to search, track, identify, and measure targets on land, sea, and in the air. However, due to various factors, such as motion blur, severe weather conditions like haze and sandstorms, and data transmission errors, the images collected by the optoelectronic pod can degrade. This degradation can severely impact image clarity and usability.

[0003] Currently, traditional image motion blur algorithms primarily calculate the blur kernel of a degraded image and then attempt to infer the clear image. However, due to the inability to fully restore the complete blur kernel of the degraded image, traditional algorithms often produce unsatisfactory restoration results. With the advancement of deep learning, a growing number of researchers are turning to convolutional neural networks to extract feature information from degraded images in the hope of achieving better image restoration. However, deep learning algorithms require large datasets of real-world data sets for training, and real datasets based on blurred images are difficult to obtain. As a result, existing datasets are primarily composed of algorithmically generated blurred images, which directly affects the quality of image restoration. Summary of the Invention

[0004] In order to overcome the defects of poor image restoration effect and low real-time performance in the above-mentioned prior art, the present invention provides an image motion deblurring method for intelligent reconnaissance equipment.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] A method for deblurring image motion of intelligent reconnaissance equipment comprises the following steps:

[0007] Capturing target images through intelligent reconnaissance equipment, sending the target images to edge computing devices for motion deblurring, and outputting generated images;

[0008] The edge computing device is equipped with a trained semi-supervised generative adversarial network consisting of a generator and a discriminator;

[0009] The generator includes a feature pyramid network for performing motion deblurring on an input image and outputting a generated image;

[0010] The discriminator includes a discriminator network based on the PatchGAN network, which is used to judge whether the generated clear image is true or false; when the discriminator output is true, the generated image output by the generator is output as the image motion deblurring result; when the discriminator output is false, the parameters of the generator are updated according to a preset loss function and the generated image is re-output.

[0011] Furthermore, the present invention also proposes a device comprising a memory and a processor, wherein the memory stores computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, the processor performs all or part of the steps of the image motion deblurring method proposed in the present invention.

[0012] Furthermore, the present invention also proposes a storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement all or part of the steps of the image motion deblurring method proposed in the present invention.

[0013] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0014] The present invention aims to solve the problem of local motion blur such as blurring and smearing of objects when the object moves due to long exposure time during image acquisition for optoelectronic pods. A new model algorithm is established, which adopts a semi-supervised generative adversarial network to generate clear images through self-supervision. The feature pyramid network and dense network are integrated into the model, which greatly improves the detailed texture of the image. At the same time, improvements are made based on the PatchGAN network to increase its receptive field to improve the network's discrimination effect, indirectly improving the algorithm's image restoration capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The figure is a flow chart of a method for image motion deblurring according to an embodiment of the present invention.

[0016] Figure 2 The figure is an architectural diagram of a generator according to one embodiment of the present invention.

[0017] Figure 3 FIG. 1 is an architecture diagram of a discriminator according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0019] The terms used in this invention are for the purpose of describing specific embodiments only and are not intended to limit the invention. The singular forms "a," "the," and "the" used in this invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0020] It should be understood that although the terms "first," "second," "third," etc. may be used in the present invention to describe various information, such information should not be limited to these terms. These terms are merely used to distinguish information of the same type from one another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present invention. Depending on the context, the term "if" as used herein may be interpreted as "when," "when," or "in response to determining."

[0021] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. Example

[0022] This embodiment proposes an image motion deblurring method for intelligent reconnaissance equipment, such as Figure 1 FIG. 1 is a flow chart of the image motion deblurring method according to the present embodiment.

[0023] The image motion deblurring method proposed in this embodiment includes the following steps:

[0024] Capturing target images through intelligent reconnaissance equipment, sending the target images to edge computing devices for motion deblurring, and outputting generated images;

[0025] The edge computing device is equipped with a trained semi-supervised generative adversarial network consisting of a generator and a discriminator;

[0026] The generator includes a feature pyramid network for performing motion deblurring on an input image and outputting a generated image;

[0027] The discriminator includes a discriminator network based on the PatchGAN network, which is used to judge whether the generated clear image is true or false; when the discriminator output is true, the generated image output by the generator is output as the image motion deblurring result; when the discriminator output is false, the parameters of the generator are updated according to a preset loss function and the generated image is re-output.

[0028] In this embodiment, when collecting images from an optoelectronic pod, due to the long exposure time, when the object moves, it is easy to cause local motion blur such as blurring and smearing of the photographed object. Therefore, a new model algorithm is established, which uses a semi-supervised generative adversarial network to generate clear images through self-supervision. The feature pyramid network and dense network are integrated into the model, which greatly improves the detailed texture of the image. At the same time, improvements are made based on the PatchGAN network to increase its receptive field to improve the network's discrimination effect, indirectly improving the algorithm's image restoration capability.

[0029] In an optional embodiment, the edge computing device uses a Jetson AGX Orin edge computing device.

[0030] As an example, the Jetson AGX Orin edge computing device is a GPU-accelerated computing platform with a computing power of about 200 TOPS. The device integrates a powerful GPU with excellent parallel computing capabilities, which enables it to handle multiple image processing tasks simultaneously and speed up processing. Secondly, Jetson AGX Orin supports a variety of mainstream programming languages, such as C / C++ and Python, and is compatible with a wide range of open source libraries and tools, which provides a wealth of tools and resources for the development and implementation of image processing algorithms. Thirdly, it has rich software development tools and documentation support, providing a flexible software programming environment and architecture. Users can freely customize and optimize image processing algorithms according to their needs to obtain better performance and results. Finally, the device is suitable for power-constrained pod environments. This enables this embodiment to reduce energy consumption while ensuring high performance.

[0031] In an optional embodiment, the generator includes a feature extraction module and a splicing module based on a feature pyramid network structure.

[0032] The feature extraction module includes a 7×7 convolutional layer and four dense convolutional connection modules connected in sequence, which are used to extract and output first feature maps of five different scales;

[0033] The feature extraction module also includes a channel splicing unit for upsampling the first feature maps of each level to the same size and then adding them through an add layer to generate four second feature maps of different scales;

[0034] The second feature map output by the feature extraction module is input into the splicing module after being upsampled by 8 times, 4 times, 2 times and 1 times respectively, for upsampling the second feature map to the same size and then performing a dimensional addition operation to obtain a third feature map;

[0035] Inputting the first feature map of the first scale output by the 7×7 convolutional layer into a splicing module for upsampling the third feature map to the same size and then performing a dimension addition operation on the third feature map and the first feature map to obtain a fourth feature map;

[0036] The input image is input to the stitching module, which is used to upsample the fourth feature map to the same size and then perform a feature value addition operation with the input image to obtain a generated image.

[0037] For example, Figure 2 The figure shows the architecture diagram of the generator of this embodiment.

[0038] Exemplarily, in the feature pyramid network structure of the generator, when a single 256×256 three-channel image is input, the 7×7 convolutional layer and the four dense convolutional connection modules extract and output the first feature maps of five different scales, whose scales are 128, 64, 32, 16, and 8 respectively.

[0039] When the first feature map of scale 8 passes through the channel splicing unit, it only passes through the convolution layer, outputting a second feature map of scale 8. After the second feature map of scale 8 is upsampled to twice its original size, it is added to the eigenvalues ​​of the secondary first feature map to generate a second feature map of scale 16. The second feature map of scale 16 is further upsampled to twice its original size and added to the eigenvalues ​​of the secondary first feature map to generate a second feature map of scale 32. The second feature map of scale 32 is further upsampled to twice its original size and added to the eigenvalues ​​of the secondary first feature map to generate a second feature map of scale 64. In this way, four second feature maps of different scales are obtained.

[0040] Furthermore, the second feature maps with scales of 8, 16, 32, and 64 are upsampled by 8 times, 4 times, 2 times, and 1 times, respectively. Then, the second feature maps of the same size are concat-dimensionally added in the splicing module to obtain a third feature map with a feature map size of 64×64.

[0041] Furthermore, the third feature map is upsampled by a factor of 2 and dimensionally added to the first feature map of scale 128 to obtain a fourth feature map of size 128×128; finally, after another upsampling by a factor of 2, the feature values ​​are added to the original image, and the generated image of size 256×256 is output.

[0042] The 7×7 convolutional layer in the feature extraction module is used to preserve the feature information of the blurred image as much as possible. Large convolution kernels can extract the most image information possible. Placing it after the input layer not only extracts the most feature information but also does not significantly increase the actual weights.

[0043] Further optionally, the dense convolution connection module includes a dense connection unit and a deformable convolution unit.

[0044] The densely connected unit includes a BN layer, a ReLU layer, and a Transition layer connected in sequence; the Transition layer includes a BN layer, a ReLU layer, a 1×1 convolution kernel, and an average pooling layer connected in sequence;

[0045] The deformable convolution unit is used to calculate the offset of each pixel of the input feature map, and adjust the sampling position of the convolution kernel according to the offset before performing the convolution operation.

[0046] Compared to residual blocks, densely connected units are placed before convolutional layers, which produces better results. The purpose of batch normalization is to convert the pixel values ​​in the image into values ​​within a normal distribution between 0 and 1. This is done to place all values ​​in an area that is sensitive to the activation function, thereby amplifying the gradient and accelerating model convergence.

[0047] The 1×1 convolution kernel in the Transition layer reduces the number of channels, which helps improve model training efficiency. If the number of channels is not limited, the resulting data volume will be extremely large, easily causing memory overflow and hindering training. The average pooling layer reduces the feature map by a factor of two to produce feature maps of different scales.

[0048] This embodiment uses a deformable convolution unit, which has the ability to learn and adapt to the deformation of irregular objects compared to ordinary convolution. For deformable convolution, no matter how complex the object is, its deformation can be learned. The receptive field after deformable convolution is not rectangular, but irregular in shape, which can completely cover the surface of the object. The receptive field of an image after standard convolution cannot completely cover the surface of the object, while the receptive field of an image after deformable convolution can completely cover the surface of the object.

[0049] Further optionally, the splicing module includes 4 concat layers, which are used to perform dimensional addition operations on the second feature maps that have undergone 8x, 4x, 2x and 1x upsampling operations respectively; the last concat layer is connected to a channel splicing unit, and the output end of each add layer is connected to a channel splicing unit; wherein the channel splicing unit includes 2 convolution layers with the same padding value.

[0050] Further optionally, the output ends of the 7×7 convolution layer and the dense convolution connection module are respectively connected to 1×1 convolution kernels, which are used to unify the dimensions of the output feature maps so that the number of channels generated by the first feature maps of each scale are unified.

[0051] In an optional embodiment, the discriminator includes 6 convolutional layers; wherein, in the sequentially connected 2nd to 4th level convolutional layers, each convolutional layer is connected to an instance normalization layer and an activation function layer; the receptive field of the discriminator is 142.

[0052] The discriminator in this embodiment improves the PatchGAN network by expanding its receptive field to evaluate the evaluation value of each local area. The receptive field can cover the surface of the object as much as possible, so that all the feature information of the object can be obtained, and the evaluation value of the area can be better evaluated.

[0053] For example, Figure 3 , which is a diagram showing the architecture of the discriminator of this embodiment.

[0054] The original PatchGAN network consists of 5 convolutional layers with a receptive field of 70. This embodiment expands its receptive field by adding a convolutional layer. The receptive field of the improved discriminator network is 142, which improves the discriminant's discrimination ability without increasing the number of parameters. Finally, a 16 16 The matrix is ​​1-sized and the matrix values ​​are averaged to determine whether the single image is real. If the image is fake, the data is fed back to the generator to update the data.

[0055] As an example, the calculation formula of the receptive field is as follows:

[0056]

[0057] in, Indicates the output size of the image; is the step length; is the convolution kernel size. It can be seen that the receptive field of this embodiment is doubled compared to the original PatchGAN network.

[0058] In an optional embodiment, the method further comprises the following steps:

[0059] Collect training data and train the generative adversarial network; wherein, the VGG16 network is used to extract the feature map of the generated image by combining the L1 mean absolute error loss and the perceptual loss, and the generator loss value is calculated by the mean square error ; Its expression is:

[0060]

[0061]

[0062]

[0063] in, Represents the L1 error loss function. The principle of the L1 loss function is to convert the true value With generated values Difference between them, and finally find the average; Represents the perceptual loss function. The perceptual loss function compares the feature vector output after convolution of the real sample with the feature vector obtained by the constructed sample. Generally, the mean square error function is used to compare the feature maps so that their feature information content is similar. is the weight coefficient; is the weight coefficient; is the true value, To generate the image i pixel values, n is the number of pixels in the generated image; Indicates the first p The convolution q The feature map obtained in the layer, and Represent the width and height of the feature map respectively; and represent real images and generated images respectively;

[0064] Use the WGAN-GP network to optimize the objective function of the generated adversarial network and calculate the discriminator loss value ; Its expression is:

[0065]

[0066] in, represents the expected value of the distribution function; represents the real data distribution, represents the noise distribution, represents the interpolated data distribution; represents the output of the discriminator; represents a sample drawn from the true data distribution; represents samples sampled from the noise distribution; represents the gradient operator.

[0067] This example uses the VGG network to extract image feature maps. In the field of image reconstruction, the mean squared error loss function is commonly used as the image pixel space loss function. However, this loss function is not capable of generating good image samples and has been gradually replaced by perceptual loss functions. Perceptual loss squares the differences of multidimensional feature maps, so using perceptual loss can construct higher-quality images.

[0068] The real image and the generated image are put into the VGG16 network to obtain the feature map. The feature map obtained by the third convolution layer in the third convolution of the VGG16 network is set to be compared, and the loss value is calculated by using the mean square error.

[0069] In an optional embodiment, the method further comprises the following steps:

[0070] The generated image output by the edge computing device is transmitted back to the cloud server via a network protocol; at the same time, the edge computing device forms a self-organizing network and transmits the generated image to a preset mobile terminal;

[0071] The cloud server analyzes the received generated images, generates real-time and / or offline reports, visual charts and / or alarm signals, and transmits them to a preset mobile terminal.

[0072] In this embodiment, the edge computing devices and the cloud form a tightly coordinated architecture, making image data acquisition, real-time processing, cloud analysis, and mobile presentation more efficient, reliable, and effective. This provides users with faster, more flexible, and more intelligent image analysis and data processing services.

[0073] For example, in the specific implementation process, the user remotely controls the movement of the pod-mounted equipment through a mobile device and sends image acquisition instructions to the cloud intelligent control computing center. The cloud dispatches collaborative work tasks to the edge computing device according to the user's instructions, and the edge computing device collects image data according to the task content.

[0074] Furthermore, edge computing devices are used for image processing tasks. The GPU of the edge computing device is used for parallel computing to accelerate the image deblurring algorithm. The optimized image processing algorithm and model are then run to improve processing speed and accuracy.

[0075] The edge computing device processes, calculates, and stores the image data collected by the camera in real time, and then returns the processing results to the cloud. The cloud analyzes the processing results of multiple machines and presents the analysis results on the mobile device according to user needs. In addition, there is also a self-organizing network between edge computing devices, which presents the image results processed by the edge device on the mobile device to complete the image motion deblurring processing.

[0076] The optoelectronic pod used in this embodiment is a camera device mounted on a mechanical arm or a crane arm, and is used to monitor and control the operation of the pod. In the optoelectronic pod, image acquisition is performed by a camera.

[0077] For example, the image acquisition process mainly includes the following steps:

[0078] 1) Use high-resolution cameras: The cameras in the optoelectronic sensors of the optoelectronic pod consist of single-photon cameras, infrared cameras, and high-resolution cameras.

[0079] 2) Camera Installation and Positioning: Mount the camera on the pod and determine its viewing range and field of view by adjusting its angle and position. The camera's position and angle should be adjusted based on the specific application to fully capture image information of the pod's operating area.

[0080] 3) Connect and configure the camera: Install the camera into the optoelectronic pod and configure it as needed. This includes setting parameters such as image resolution, frame rate, exposure time, and sensor gain.

[0081] 4) Image Capture: Once the camera is connected and configured, users can remotely control the pod's onboard equipment and send commands using their mobile device. The camera periodically generates image frames and transmits them via digital signals. The optoelectronic pod's edge computing device receives and processes these frames, enabling real-time monitoring of the pod's operating area.

[0082] Through the above steps, the optoelectronic pod can realize image acquisition of the operating area.

[0083] Furthermore, a single blurry image captured by the optoelectronic pod is processed through a feature extraction module based on a feature pyramid network structure, which outputs feature maps at different scales. Next, a single 256×256 three-channel image is input, and feature maps are extracted through a dense convolutional connection module. Feature maps at each scale are upsampled and summed to generate a sharp image. Finally, after dimensional summation, the feature maps are output, and further upsampling and dimensional summation are performed to obtain the final sharp generated image.

[0084] The generated image output by the generator is fed into the discriminator for evaluation. The discriminator uses the PatchGAN network to add convolutional layers to expand the receptive field and improve its discriminative ability. The discriminator ultimately generates a 16×16×1 matrix, which is averaged to assess the authenticity of the image.

[0085] Furthermore, the processed image data can be stored temporarily or permanently within the edge computing device. At the same time, the edge computing device transmits the processing results back to the cloud server via a network protocol. After receiving the image data results from the edge computing device, the cloud server uses more powerful computing power, storage, and algorithms for further analysis and processing.

[0086] The results of cloud processing can be presented on mobile devices through real-time or offline reports, visual charts, alerts, or other forms of information.

[0087] The edge computing devices within the pod can also form a self-organizing network, sharing the processed image results and presenting them on the mobile device. This collaboration can improve real-time performance, accuracy, and reliability.

[0088] When performing optoelectronic pod cloud-edge collaborative processing, this embodiment will combine the entire system through the pod's real-time data processing, enabling it to comprehensively respond to various processing services. Example

[0089] This embodiment proposes a device including a memory and a processor, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes all or part of the steps of the image motion deblurring method proposed in Example 1. Example

[0090] This embodiment provides a storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement all or part of the steps of the image motion deblurring method provided in Embodiment 1.

[0091] Exemplarily, the storage medium includes, but is not limited to, a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and other media that can store program codes.

[0092] Exemplarily, the instructions, programs, code sets or instruction sets may be implemented using conventional programming languages.

[0093] Exemplarily, the processor includes but is not limited to a smart phone, a personal computer, a server, a network device, etc., and is configured to execute all or part of the steps of the image motion deblurring method described in Example 1.

[0094] The terms in the drawings are for illustrative purposes only and are not to be construed as limiting the present invention;

[0095] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A method for image motion deblurring of intelligent reconnaissance equipment, characterized in that: The following steps are involved: Capturing target images through intelligent reconnaissance equipment, sending the target images to edge computing devices for motion deblurring, and outputting generated images; The edge computing device is equipped with a trained semi-supervised generative adversarial network consisting of a generator and a discriminator; The generator includes a feature pyramid network for performing motion deblurring on an input image and outputting a generated image; The discriminator includes a discriminator network based on the PatchGAN network, which is used to judge whether the generated clear image is true or false; when the discriminator outputs true, the generated image output by the generator is output as the image motion deblurring result; when the discriminator outputs false, the parameters of the generator are updated according to a preset loss function and the generated image is re-output; The generator includes a feature extraction module and a splicing module based on a feature pyramid network structure; wherein: The feature extraction module includes a 7×7 convolutional layer and four dense convolution connection modules connected in sequence, which are used to extract and output first feature maps of five different scales; The feature extraction module also includes a channel splicing unit for upsampling the first feature maps of each level to the same size and then adding them through an add layer to generate four second feature maps of different scales; The second feature map output by the feature extraction module is input into the splicing module after being upsampled by 8 times, 4 times, 2 times and 1 times respectively, for upsampling the second feature map to the same size and then performing a dimensional addition operation to obtain a third feature map; Inputting the first feature map of the first scale output by the 7×7 convolutional layer into a splicing module for upsampling the third feature map to the same size and then performing a dimension addition operation on the third feature map and the first feature map to obtain a fourth feature map; Inputting the input image into a stitching module, for upsampling the fourth feature map to the same size and then performing a feature value addition operation with the input image to obtain a generated image; The dense convolution connection module includes a dense connection unit and a deformable convolution unit.

2. The image motion deblurring method according to claim 1, wherein: in, The densely connected unit includes a BN layer, a ReLU layer, and a Transition layer connected in sequence; the Transition layer includes a BN layer, a ReLU layer, a 1×1 convolution kernel, and an average pooling layer connected in sequence; The deformable convolution unit is used to calculate the offset of each pixel of the input feature map, and adjust the sampling position of the convolution kernel according to the offset before performing the convolution operation.

3. The image motion deblurring method according to claim 1, wherein: The splicing module includes four concat layers, which are used to perform dimensional addition operations on the second feature maps that have been upsampled by 8 times, 4 times, 2 times, and 1 times respectively; The last concat layer is connected to a channel splicing unit, and the output end of each add layer is connected to a channel splicing unit; wherein the channel splicing unit includes two convolutional layers with the same padding value.

4. The image motion deblurring method according to claim 1, wherein: In the generator, the output ends of the 7×7 convolution layer and the dense convolution connection module are respectively connected to 1×1 convolution kernels.

5. The image motion deblurring method according to claim 1, wherein: The discriminator includes 6 convolutional layers; among the sequentially connected second-level convolutional layers to the fourth-level convolutional layers, each convolutional layer is connected to an instance normalization layer and an activation function layer; the receptive field of the discriminator is 142.

6. The image motion deblurring method according to any one of claims 1 to 5, characterized in that: The method further comprises the following steps: Collect training data and train the generative adversarial network; wherein, the VGG16 network is used to extract the feature map of the generated image by combining the L1 mean absolute error loss and the perceptual loss, and the generator loss value is calculated by the mean square error ; Its expression is: in, represents the L1 mean absolute error loss, Indicates perceived loss; is the weight coefficient; is the weight coefficient; is the true value, To generate the image i pixel values, n is the number of pixels in the generated image; Indicates the first p The convolution q The feature map obtained in the layer, and Represent the width and height of the feature map respectively; and represent real images and generated images respectively; Use the WGAN-GP network to optimize the objective function of the generated adversarial network and calculate the discriminator loss value ; Its expression is: in, represents the expected value of the distribution function; represents the real data distribution, represents the noise distribution, represents the interpolated data distribution; represents the output of the discriminator; represents a sample drawn from the true data distribution; represents samples sampled from the noise distribution; represents the gradient operator.

7. The image motion deblurring method according to any one of claims 1 to 5, characterized in that: The method further comprises the following steps: The generated image output by the edge computing device is transmitted back to the cloud server via a network protocol; at the same time, the edge computing device forms a self-organizing network and transmits the generated image to a preset mobile terminal; The cloud server analyzes the received generated images, generates real-time and / or offline reports, visual charts and / or alarm signals, and transmits them to a preset mobile terminal.

8. A device comprising a memory and a processor, wherein the memory stores computer-readable instructions, wherein: When the computer-readable instructions are executed by the processor, the processor is caused to perform all or part of the steps of the image motion deblurring method according to any one of claims 1 to 7.

9. A storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, all or part of the steps of the image motion deblurring method according to any one of claims 1 to 7 are implemented.

Images