Remote sensing image classification method and device based on supply chain management, equipment and medium

Abstract

The invention discloses a remote sensing image classification method and device based on supply chain management, equipment and a medium, and relates to the technical field of image processing. Through combination of an image fusion technology and a generative adversarial network, a plurality of remote sensing images with different angles and different resolutions are fused, and the fused images are adopted to train the generative adversarial network, so that a more real category label can be obtained, and the accuracy, reliability and robustness of image classification are improved. The method comprises the following steps: preprocessing a plurality of initial remote sensing images to obtain a plurality of target remote sensing images; splicing the plurality of target remote sensing images by using an image fusion technology, and performing curve fitting on the spliced image boundary by using a nonlinear interpolation technology to obtain a fused image; training the initial generative adversarial network by adopting the fused image to obtain a target generative adversarial network; and obtaining a plurality of to-be-classified remote sensing images, and inputting the plurality of to-be-classified remote sensing images into the target generative adversarial network to obtain a plurality of category labels.

Description

Technical Field

[0001] The present application relates to the field of image processing technologies, and particularly to a remote sensing image classification method, device, equipment and medium based on supply chain management. Background Art

[0002] With the rapid development of the global supply chain, the requirements for the accuracy and efficiency of supply chain management are getting higher and higher. In order to better understand and manage the supply chain, we need to obtain and process remote sensing images of it with high precision.

[0003] In the related art, a convolutional neural network is used to train a sample set to realize the classification of remote sensing images. However, the applicant realizes that there are many problems in the traditional remote sensing image classification method, such as large image noise and serious information loss, etc., which make the traditional remote sensing image classification method have certain uncertainty and limitations, resulting in low accuracy of remote sensing image classification. Summary of the Invention

[0004] In view of this, the present application provides a remote sensing image classification method, device, equipment and medium based on supply chain management, and the main purpose is to solve the problems such as large image noise and serious information loss existing in the traditional remote sensing image classification method.

[0005] According to the first aspect of the present application, a remote sensing image classification method based on supply chain management is provided, and the method includes:

[0006] Obtain a plurality of initial remote sensing images, and perform preprocessing on the plurality of initial remote sensing images to obtain a plurality of target remote sensing images;

[0007] Obtain an image fusion technology and a non-linear interpolation technology, splice the plurality of target remote sensing images by using the image fusion technology, and perform curve fitting on the boundary of the spliced image by using the non-linear interpolation technology to obtain a fused image;

[0008] Obtain an initial generative adversarial network, and train the initial generative adversarial network by using the fused image to obtain a target generative adversarial network;

[0009] Obtain a plurality of remote sensing images to be classified, and input the plurality of remote sensing images to be classified into the target generative adversarial network to obtain a plurality of class labels.

[0010] Optionally, the performing preprocessing on the plurality of initial remote sensing images to obtain a plurality of target remote sensing images includes:

[0011] Based on an image processing software, obtain the coordinate information of each of the initial remote sensing images to obtain a plurality of initial coordinate information;

[0012] Obtain a feature detection algorithm and a feature matching algorithm. Use the feature detection algorithm to detect the multiple initial coordinate information to obtain multiple initial image feature points. Use the feature matching algorithm to perform feature point matching on the multiple initial image feature points to obtain an initial feature point matching result. Use the initial feature point matching result to determine multiple translation amounts;

[0013] Based on the image processing software, perform translation processing on the multiple initial remote sensing images according to the multiple translation amounts to obtain multiple first remote sensing images;

[0014] Use the initial feature point matching result to calculate the rotation angle of each of the initial remote sensing images to obtain multiple rotation angles;

[0015] Based on the image processing software, perform rotation processing on the multiple first remote sensing images according to the multiple rotation angles to obtain multiple second remote sensing images;

[0016] Obtain a target resolution, determine the resolution of each of the second remote sensing images, and calculate the scaling ratio of each of the second remote sensing images using the target resolution and the resolution of each of the second remote sensing images;

[0017] Obtain an interpolation algorithm. Based on the image processing software, use the interpolation algorithm to adjust each of the second remote sensing images according to the scaling ratio of each of the second remote sensing images to obtain multiple third remote sensing images;

[0018] Obtain a grayscale function. Based on the image processing software, use the grayscale function to perform grayscale processing on the multiple third remote sensing images to obtain multiple fourth remote sensing images;

[0019] Based on the image processing software, perform denoising processing on the multiple fourth remote sensing images to obtain multiple fifth remote sensing images. The denoising processing is any one of mean filtering processing, median filtering processing, and Gaussian filtering processing;

[0020] Obtain an image correction model. Based on the image correction model, calculate the correction parameters of each of the fifth remote sensing images. Based on the image correction model, perform geometric correction processing on each of the fifth remote sensing images according to the correction parameters of each of the fifth remote sensing images to obtain multiple sixth remote sensing images;

[0021] Obtain a convolutional neural network model. Based on the convolutional neural network model, perform image enhancement on the multiple sixth remote sensing images to obtain the multiple target remote sensing images.

[0022] Optionally, acquiring the image fusion technology and the non-linear interpolation technology, and using the image fusion technology to splice the multiple target remote sensing images, and using the non-linear interpolation technology to perform curve fitting on the boundaries of the spliced image to obtain a fused image, includes:

[0023] Obtaining the coordinate information of each of the target remote sensing images based on an image processing software to obtain a plurality of target coordinate information;

[0024] Acquiring the feature detection algorithm and the feature matching algorithm included in the image fusion technology, using the feature detection algorithm to detect the plurality of target coordinate information to obtain a plurality of target image feature points, and using the feature matching algorithm to perform feature point matching on the plurality of target image feature points to obtain a target feature point matching result;

[0025] Splicing the multiple target remote sensing images according to the target feature point matching result to obtain a spliced image, where the spliced image includes a plurality of image splicing boundaries;

[0026] Using the non-linear interpolation technology to perform curve fitting on the plurality of image splicing boundaries to obtain the fused image.

[0027] Optionally, the method further includes:

[0028] Acquiring a spatial pyramid pooling network, and inputting the multiple target remote sensing images into the spatial pyramid pooling network;

[0029] Dividing the multiple target remote sensing images into a plurality of time windows based on the spatial pyramid pooling network, where the plurality of time windows do not overlap with each other;

[0030] Performing local feature extraction on the plurality of time windows based on the encoder of the spatial pyramid pooling network to obtain a plurality of feature maps, where the encoder is any one of a convolutional neural network, a recurrent neural network, and a long short-term memory network;

[0031] Performing splicing processing on the plurality of feature maps based on the spatial pyramid pooling network to obtain a spliced feature map;

[0032] Performing a decoding operation on the spliced feature map based on the decoder of the spatial pyramid pooling network to obtain a reconstructed image, and performing normalization processing on the reconstructed image based on the spatial pyramid pooling network to obtain the fused image.

[0033] Optionally, the method further includes:

[0034] Obtaining the weight coefficient corresponding to each of the target remote sensing images to obtain a plurality of weight coefficients, and extracting a plurality of pixel points included in each of the target remote sensing images;

[0035] Arbitrarily select one of the multiple target remote sensing images as the reference image. For each pixel point of the reference image, extract the corresponding pixel points in the multiple target remote sensing images other than the reference image to obtain multiple target pixel points, and calculate the weighted sum using the multiple weight coefficients, the pixel point, and the multiple target pixel points to obtain the fused pixel point;

[0036] Use the multiple target remote sensing images other than the reference image to perform weighted average calculation on each pixel point of the reference image to obtain multiple fused pixel points;

[0037] Generate the fused image using the multiple fused pixel points.

[0038] Optionally, after splicing the multiple target remote sensing images and using the non - linear interpolation technique to perform curve fitting on the boundary of the spliced image to obtain the fused image, the method further includes:

[0039] Perform post - processing on the fused image, and the post - processing includes noise removal and color correction.

[0040] Optionally, the step of training the initial generative adversarial network using the fused image to obtain the target generative adversarial network includes:

[0041] Generate a training set, a validation set, and a test set based on the fused image;

[0042] Use the training set to train the initial generative adversarial network, and use the test set and the validation set to test and validate the trained initial generative adversarial network respectively to obtain the target generative adversarial network.

[0043] According to the second aspect of the present application, a remote sensing image classification device based on supply chain management is provided, and the device includes:

[0044] A pre - processing module, configured to obtain multiple initial remote sensing images and perform pre - processing on the multiple initial remote sensing images to obtain multiple target remote sensing images;

[0045] A fusion module, configured to obtain an image fusion technique and a non - linear interpolation technique, splice the multiple target remote sensing images using the image fusion technique, and perform curve fitting on the boundary of the spliced image using the non - linear interpolation technique to obtain a fused image;

[0046] A training module, configured to obtain an initial generative adversarial network and train the initial generative adversarial network using the fused image to obtain a target generative adversarial network;

[0047] A classification module, configured to obtain a plurality of remotely sensed images to be classified, input the plurality of remotely sensed images to be classified into the target generative adversarial network, and obtain a plurality of class labels.

[0048] Optionally, the preprocessing module is configured to obtain coordinate information of each of the initial remotely sensed images based on an image processing software, so as to obtain a plurality of initial coordinate information; obtain a feature detection algorithm and a feature matching algorithm, use the feature detection algorithm to detect the plurality of initial coordinate information, so as to obtain a plurality of initial image feature points, use the feature matching algorithm to perform feature point matching on the plurality of initial image feature points, so as to obtain an initial feature point matching result, and use the initial feature point matching result to determine a plurality of translation amounts; based on the image processing software, perform translation processing on the plurality of initial remotely sensed images according to the plurality of translation amounts, so as to obtain a plurality of first remotely sensed images; calculate the rotation angle of each of the initial remotely sensed images by using the initial feature point matching result, so as to obtain a plurality of rotation angles; based on the image processing software, perform rotation processing on the plurality of first remotely sensed images according to the plurality of rotation angles, so as to obtain a plurality of second remotely sensed images; obtain a target resolution, determine the resolution of each of the second remotely sensed images, and calculate the scaling ratio of each of the second remotely sensed images by using the target resolution and the resolution of each of the second remotely sensed images; obtain an interpolation algorithm, and based on the image processing software, use the interpolation algorithm to adjust each of the second remotely sensed images according to the scaling ratio of each of the second remotely sensed images, so as to obtain a plurality of third remotely sensed images; obtain a grayscale function, and based on the image processing software, use the grayscale function to perform grayscale processing on the plurality of third remotely sensed images, so as to obtain a plurality of fourth remotely sensed images; based on the image processing software, perform denoising processing on the plurality of fourth remotely sensed images, so as to obtain a plurality of fifth remotely sensed images, where the denoising processing is any one of mean filtering processing, median filtering processing, and Gaussian filtering processing; obtain an image correction model, calculate the correction parameter of each of the fifth remotely sensed images based on the image correction model, and based on the image correction model, perform geometric correction processing on each of the fifth remotely sensed images according to the correction parameter of each of the fifth remotely sensed images, so as to obtain a plurality of sixth remotely sensed images; obtain a convolutional neural network model, and perform image enhancement on the plurality of sixth remotely sensed images based on the convolutional neural network model, so as to obtain the plurality of target remotely sensed images.

[0049] Optionally, the fusion module is configured to obtain the coordinate information of each of the target remote sensing images based on image processing software, obtaining a plurality of target coordinate information; obtain the feature detection algorithm and the feature matching algorithm included in the image fusion technology, use the feature detection algorithm to detect the plurality of target coordinate information, obtaining a plurality of target image feature points, use the feature matching algorithm to perform feature point matching on the plurality of target image feature points, obtaining a target feature point matching result; stitch the plurality of target remote sensing images according to the target feature point matching result, obtaining a stitched image, the stitched image including a plurality of image stitching boundaries; use the non-linear interpolation technique to perform curve fitting on the plurality of image stitching boundaries, obtaining the fused image.

[0050] Optionally, the fusion module is configured to obtain a spatial pyramid pooling network, input the plurality of target remote sensing images into the spatial pyramid pooling network; divide the plurality of target remote sensing images into a plurality of time windows based on the spatial pyramid pooling network, wherein the plurality of time windows do not overlap; perform local feature extraction on the plurality of time windows based on the encoder of the spatial pyramid pooling network, obtaining a plurality of feature maps, the encoder being any one of a convolutional neural network, a recurrent neural network, and a long short-term memory network; perform stitching processing on the plurality of feature maps based on the spatial pyramid pooling network, obtaining a stitched feature map; perform decoding operations on the stitched feature map based on the decoder of the spatial pyramid pooling network, obtaining a reconstructed image, and perform normalization processing on the reconstructed image based on the spatial pyramid pooling network, obtaining the fused image.

[0051] Optionally, the fusion module is configured to obtain the weight coefficient corresponding to each of the target remote sensing images, obtaining a plurality of weight coefficients, and extract a plurality of pixel points included in each of the target remote sensing images; arbitrarily select one of the target remote sensing images as a reference image from the plurality of target remote sensing images, for each pixel point of the reference image, extract the pixel point corresponding to the pixel point from the plurality of target remote sensing images other than the reference image, obtaining a plurality of target pixel points, calculate the weighted sum using the plurality of weight coefficients, the pixel point, and the plurality of target pixel points, obtaining a fused pixel point; perform weighted average calculation on each pixel point of the reference image using the plurality of target remote sensing images other than the reference image, obtaining a plurality of fused pixel points; generate the fused image using the plurality of fused pixel points.

[0052] Optionally, the apparatus further includes:

[0053] A post-processing module, configured to perform post-processing on the fused image, the post-processing including noise removal and color correction.

[0054] Optionally, the training module is configured to generate a training set, a validation set, and a test set based on the fused image; train the initial generative adversarial network using the training set, and test and validate the trained initial generative adversarial network using the test set and the validation set respectively to obtain the target generative adversarial network.

[0055] According to a third aspect of the present application, there is provided a device including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the method according to any one of the first aspects above are implemented.

[0056] According to a fourth aspect of the present application, there is provided a medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method according to any one of the first aspects above are implemented.

[0057] By means of the above technical solutions, a remote sensing image classification method, device, equipment, and medium based on supply chain management provided by the present application. The present application obtains multiple initial remote sensing images, preprocesses the multiple initial remote sensing images to obtain multiple target remote sensing images, obtains image fusion technology and non-linear interpolation technology, stitches the multiple target remote sensing images using the image fusion technology, and performs curve fitting on the boundaries of the stitched images using the non-linear interpolation technology to obtain a fused image, obtains an initial generative adversarial network, trains the initial generative adversarial network using the fused image to obtain a target generative adversarial network, obtains multiple remote sensing images to be classified, and inputs the multiple remote sensing images to be classified into the target generative adversarial network to obtain multiple class labels. By combining the image fusion technology and the generative adversarial network, multiple remote sensing images with different angles and different resolutions are fused, and the fused image is used to train the generative adversarial network, which can obtain more real class labels, improve the accuracy, reliability, and robustness of image classification, and thus be widely applied to fields such as supply chain management and environmental monitoring, providing a more accurate and reliable remote sensing image processing solution for related industries.

[0058] The above description is only an overview of the technical solutions of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of the present application more obvious and understandable, the specific embodiments of the present application are specifically described below. Description of the Drawings

[0059] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present application. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0060] Figure 1 It shows a schematic flow chart of a method for remote sensing image classification based on supply chain management provided by an embodiment of the present application;

[0061] Figure 2 It shows another schematic flow chart of a method for remote sensing image classification based on supply chain management provided by an embodiment of the present application;

[0062] Figure 3A It shows a schematic structural diagram of a remote sensing image classification based on supply chain management provided by an embodiment of the present application;

[0063] Figure 3B It shows another schematic structural diagram of a remote sensing image classification based on supply chain management provided by an embodiment of the present application;

[0064] Figure 4 It shows a schematic structural diagram of a device provided by an embodiment of the present application. Detailed implementation manners

[0065] Hereinafter, exemplary embodiments of the present application will be described in more detail with reference to the drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present application can be more thoroughly understood and the scope of the present application can be fully conveyed to those skilled in the art.

[0066] An embodiment of the present application provides a method for remote sensing image classification based on supply chain management, as Figure 1 shown, the method includes:

[0067] 101. Obtain a plurality of initial remote sensing images, and perform preprocessing on the plurality of initial remote sensing images to obtain a plurality of target remote sensing images.

[0068] Due to the increasing demand for high-precision classification and enhancement of remote sensing images. However, there are many problems with traditional remote sensing image processing methods, such as large image noise and serious information loss, which often lead to certain uncertainties and limitations in traditional remote sensing image processing methods. To solve this problem, this application proposes a remote sensing image classification method based on supply chain management, which uses preprocessing means such as denoising, image registration, geometric correction, and image enhancement to improve the quality and consistency of data. Then, an image fusion technology is used to fuse the remote sensing images to provide high-quality input for the generative adversarial network. Then, the fused images are used to train the generative adversarial network, which can achieve high-precision classification, avoid the time and labor costs brought by manual annotation, and ensure the consistency and accuracy of labels. The execution entity of this application can be a remote sensing image processing system, and the remote sensing image processing system relies on the computing power of the server to provide services for users. The server can be an independent server or can provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms, so that the remote sensing image processing system can generate more realistic class labels to ensure the accuracy and reliability of classification.

[0069] In the embodiment of this application, the remote sensing image processing system acquires multiple initial remote sensing images. Among them, the remote sensing images can be images taken at the same shooting location on the supply chain at different times or under different weather conditions, or images taken at different geographical locations on the supply chain. It should be noted that in order to ensure the efficiency and stability of the image acquisition process, the most suitable sensor type and appropriate shooting time are selected to collect remote sensing images related to the supply chain to ensure the high accuracy and integrity of the collected data. Then, the remote sensing image processing system preprocesses the multiple initial remote sensing images to obtain multiple target remote sensing images. In practical applications, due to the different sources and shooting methods of remote sensing images, there will be some unnecessary information or noise, so it is necessary to preprocess the original remote sensing images to improve the fusion quality.

[0070] 102. Acquire the image fusion technology and the non-linear interpolation technology, splice the multiple target remote sensing images by using the image fusion technology, and perform curve fitting on the boundaries of the spliced images by using the non-linear interpolation technology to obtain the fused image.

[0071] In the embodiment of the present application, a remote sensing image processing system acquires an image fusion technology and a non-linear interpolation technology, stitches multiple target remote sensing images using the image fusion technology, and performs curve fitting on the boundaries of the stitched images using the non-linear interpolation technology to obtain a fused image. It should be noted that the stitching method can be implemented in various ways, such as pixel-level stitching based on coordinate matching, descriptor-level stitching based on feature matching, curve fitting based on non-linear interpolation, etc. These stitching methods can be selected according to specific requirements and data characteristics to achieve the best fusion effect, and the embodiment of the present application does not specifically limit them here.

[0072] 103. Acquire an initial generative adversarial network, and train the initial generative adversarial network using the fused image to obtain a target generative adversarial network.

[0073] In the embodiment of the present application, a remote sensing image processing system acquires an initial generative adversarial network, and trains the initial generative adversarial network using the fused image to obtain a target generative adversarial network. Among them, a generative adversarial network (GAN) is a deep learning model that can be used to generate new, seemingly real images, videos, and other types of data. This network structure utilizes two neural networks, a generator network and a discriminator network, to compete with each other. The task of the generator network is to generate target data as realistically as possible based on conditions such as random noise or text descriptions, while the goal of the discriminator network is to distinguish between the generated data and the real data. The two continuously conduct adversarial training, and ultimately the generator network will learn how to generate as realistic data as possible. Training using a generative adversarial network can improve the accuracy and reliability of classification.

[0074] 104. Acquire multiple remote sensing images to be classified, and input the multiple remote sensing images to be classified into the target generative adversarial network to obtain multiple class labels.

[0075] In the embodiment of the present application, a remote sensing image processing system acquires multiple remote sensing images to be classified, and inputs the multiple remote sensing images to be classified into the target generative adversarial network to obtain multiple class labels. This can not only achieve high-precision classification, but also obtain more realistic class labels, avoid the time and labor costs brought by manual annotation, and ensure the consistency and accuracy of the labels.

[0076] The method provided by the embodiments of the present application obtains multiple initial remote sensing images, preprocesses the multiple initial remote sensing images to obtain multiple target remote sensing images, obtains image fusion technology and non-linear interpolation technology, stitches the multiple target remote sensing images using the image fusion technology, and performs curve fitting on the boundaries of the stitched images using the non-linear interpolation technology to obtain a fused image. An initial generative adversarial network is obtained, and the initial generative adversarial network is trained using the fused image to obtain a target generative adversarial network. Multiple remote sensing images to be classified are obtained, and the multiple remote sensing images to be classified are input into the target generative adversarial network to obtain multiple class labels. By combining the image fusion technology and the generative adversarial network, multiple remote sensing images with different angles and resolutions are fused, and the fused image is used to train the generative adversarial network, which can obtain more realistic class labels, improve the accuracy, reliability, and robustness of image classification, and thus be widely applied in fields such as supply chain management and environmental monitoring, providing a more accurate and reliable remote sensing image processing solution for related industries.

[0077] Further, as a refinement and extension of the specific implementation manner of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, the embodiments of the present application provide another remote sensing image classification method based on supply chain management, as Figure 2 shown, the method includes:

[0078] 201. Obtain multiple initial remote sensing images, and perform translation processing on the multiple initial remote sensing images based on image processing software to obtain multiple first remote sensing images.

[0079] In the embodiments of the present application, during the preprocessing process, the remote sensing image processing system performs various processes on the collected initial remote sensing images, including translation, rotation, scaling, grayscale conversion, denoising, image correction, and image enhancement. The purpose of these processing steps is to eliminate noise and ensure the consistency of the geometric relationships between images, thereby improving the quality and consistency of the data. This guarantee of consistency is crucial for subsequent feature extraction and model training because model training relies on a stable data set.

[0080] First, the remote sensing image processing system performs translation processing to adjust the positions of the initial remote sensing images to align them with other initial remote sensing images.

[0081] Specifically, the remote sensing image processing system obtains the coordinate information of each initial remote sensing image based on the image processing software, and obtains a plurality of initial coordinate information. Then, the remote sensing image processing system obtains the feature detection algorithm and the feature matching algorithm, and uses the feature detection algorithm to detect the plurality of initial coordinate information to obtain a plurality of initial image feature points. For example, the feature detection algorithm can be the Scale Invariant Feature Transform (SIFT), the Speeded Up Robust Features (SURF), the algorithm for feature point extraction and description in computer vision (Oriented FAST and Rotated BRIEF, ORB), etc. Subsequently, the remote sensing image processing system uses the feature matching algorithm to perform feature point matching on the plurality of initial image feature points to obtain an initial feature point matching result, and uses the initial feature point matching result to determine a plurality of translation amounts. For example, the feature matching algorithm can be the Fast Library for Approximate Nearest Neighbors (FLANN), etc. Then, the remote sensing image processing system, based on the image processing software, performs translation processing on the plurality of initial remote sensing images according to the plurality of translation amounts to obtain a plurality of first remote sensing images. Optionally, the remote sensing image processing system can use a translation function in a programming language (such as the OpenCV library of Python) to move the remote sensing image according to the calculated translation amount. It should be noted that if blank areas appear after translation, they are filled with black or adjacent pixel values.

[0082] 202. Rotate the plurality of first remote sensing images based on the image processing software to obtain a plurality of second remote sensing images.

[0083] Secondly, the remote sensing image processing system performs rotation processing to adjust the posture of the remote sensing image and eliminate the perspective difference.

[0084] In the embodiment of the present application, the remote sensing image processing system calculates the rotation angle of each initial remote sensing image using the initial feature point matching result to obtain a plurality of rotation angles. Then, the remote sensing image processing system, based on the image processing software, performs rotation processing on the plurality of first remote sensing images according to the plurality of rotation angles to obtain a plurality of second remote sensing images. Optionally, the remote sensing image processing system can use a rotation function in a programming language (such as the OpenCV library of Python) to rotate the remote sensing image according to the calculated angle. It should be noted that if blank areas appear after rotation, they are filled with black or adjacent pixel values.

[0085] 203. Obtain an interpolation algorithm. Based on image processing software, use the interpolation algorithm to adjust each second remote sensing image to obtain multiple third remote sensing images.

[0086] Then, the remote sensing image processing system performs a scaling process to adjust the size of the remote sensing image to be consistent with the resolution of other remote sensing images.

[0087] In the embodiment of the present application, the remote sensing image processing system obtains the target resolution, determines the resolution of each second remote sensing image, and calculates the scaling ratio of each second remote sensing image using the target resolution and the resolution of each second remote sensing image. Among them, the target resolution can be set according to the size of the remote sensing image, and the embodiment of the present application does not make specific limitations. Then, the remote sensing image processing system obtains an interpolation algorithm. Based on image processing software, use the interpolation algorithm to adjust each second remote sensing image according to the scaling ratio of each second remote sensing image to obtain multiple third remote sensing images. Among them, the interpolation algorithm can be, for example, bilinear interpolation, nearest neighbor interpolation, etc., which are used to generate new pixel values. Optionally, the remote sensing image processing system can use a scaling function in a programming language (such as the OpenCV library of Python) to enlarge or reduce the remote sensing image according to the calculated ratio.

[0088] Based on the above process, the remote sensing image processing system completes the image registration of the remote sensing image. Through translation, rotation, and scaling processing, it ensures that the geometric relationships between multiple remote sensing images are consistent.

[0089] 204. Obtain a grayscale function. Based on image processing software, use the grayscale function to perform grayscale processing on multiple third remote sensing images to obtain multiple fourth remote sensing images.

[0090] Next, the remote sensing image processing system performs grayscale processing to convert the color remote sensing image into a grayscale remote sensing image, reducing the impact of color differences on the fusion effect.

[0091] In the embodiment of the present application, the remote sensing image processing system obtains a grayscale function. Based on image processing software, use the grayscale function to perform grayscale processing on multiple third remote sensing images to obtain multiple fourth remote sensing images. For example, the grayscale function can be the simple average grayscale method: Gray = (R + G + B) / 3, the weighted average grayscale method: Gray = 0.299*R + 0.587*G + 0.114*B, etc. Optionally, the remote sensing image processing system can use a grayscale function in a programming language (such as the OpenCV library of Python) to convert the color remote sensing image into a grayscale remote sensing image.

[0092] 205. Based on image processing software, perform denoising processing on multiple fourth remote sensing images to obtain multiple fifth remote sensing images.

[0093] Subsequently, the remote sensing image processing system performs denoising processing to reduce the noise in the remote sensing image and improve the quality of the remote sensing image.

[0094] In the embodiment of the present application, the remote sensing image processing system performs denoising processing on multiple fourth remote sensing images based on image processing software to obtain multiple fifth remote sensing images. Among them, the denoising processing is any one of mean filtering processing, median filtering processing, and Gaussian filtering processing. Optionally, the remote sensing image processing system can use a denoising function in a programming language (such as the OpenCV library of Python) to perform denoising processing on the remote sensing image.

[0095] 206. Obtain an image correction model, and perform geometric correction processing on each fifth remote sensing image based on the image correction model to obtain multiple sixth remote sensing images.

[0096] Then, the remote sensing image processing system performs geometric correction processing to correct the geometric distortion of the remote sensing image and ensure the geometric accuracy of the remote sensing image.

[0097] In the embodiment of the present application, the remote sensing image processing system obtains an image correction model and calculates the correction parameters of each fifth remote sensing image based on the image correction model. It should be noted that the image correction model can be a polynomial model, a perspective transformation model, etc., and the embodiment of the present application does not make specific limitations. Then, the remote sensing image processing system performs geometric correction processing on each fifth remote sensing image according to the correction parameters of each fifth remote sensing image based on the image correction model to obtain multiple sixth remote sensing images.

[0098] 207. Obtain a convolutional neural network model, and perform image enhancement on multiple sixth remote sensing images based on the convolutional neural network model to obtain multiple target remote sensing images.

[0099] Finally, the remote sensing image processing system performs image enhancement processing to improve the quality of the remote sensing image.

[0100] In the embodiment of the present application, the remote sensing image processing system obtains a convolutional neural network model and performs image enhancement on multiple sixth remote sensing images based on the convolutional neural network model to obtain multiple target remote sensing images. Through the convolutional neural network model, the defects in the remote sensing image are deeply mined, and the remote sensing image is enhanced, which can enhance the robustness of the image, so that even in the subsequent training process, facing different types of interference or attacks, it can maintain high classification accuracy and robustness, thereby improving the accuracy and robustness of image classification.

[0101] In summary, the preprocessing methods include, but are not limited to, geometric transformations such as translation and rotation, as well as color conversions such as grayscale conversion, with the aim of improving the image quality. Among them, geometric transformations such as translation, rotation, and scaling can eliminate perspective differences by changing the position, attitude, and size of the remote sensing images, ensuring the correct relative position between the remote sensing images, thereby improving the fusion effect of the remote sensing images; color conversions such as grayscale conversion can convert remote sensing images in different bands into the same color space, thereby reducing the impact of color differences on the fusion effect. Through these preprocessing operations, the quality of the remote sensing images can be improved, thus providing high-quality inputs for the generative adversarial network.

[0102] 208. Obtain the image fusion technology and the non-linear interpolation technology, splice multiple target remote sensing images using the image fusion technology, and perform curve fitting on the boundaries of the spliced images using the non-linear interpolation technology to obtain the fused image.

[0103] In the embodiment of the present application, the remote sensing image processing system directly performs pixel-level matching and splicing based on the remote sensing image coordinate information. Specifically, the remote sensing image processing system obtains the coordinate information of each target remote sensing image based on the image processing software to obtain multiple target coordinate information. Then, the remote sensing image processing system obtains the feature detection algorithm and the feature matching algorithm included in the image fusion technology, uses the feature detection algorithm to detect the multiple target coordinate information to obtain multiple target image feature points, and uses the feature matching algorithm to perform feature point matching on the multiple target image feature points to obtain the target feature point matching result. Subsequently, the remote sensing image processing system splices the multiple target remote sensing images according to the target feature point matching result to obtain a spliced image, where the spliced image includes multiple image splicing boundaries. Then, the remote sensing image processing system uses the non-linear interpolation technology to perform curve fitting on the multiple image splicing boundaries to achieve a more natural transition and obtain the fused image. It should be noted that the splicing process can adopt various methods, such as pixel-level splicing based on coordinate matching, descriptor-level splicing based on feature matching, curve fitting based on non-linear interpolation, etc., and the most suitable splicing method can be selected according to specific requirements and data characteristics to achieve the best visual effect and application requirements.

[0104] In an alternative embodiment, the remote sensing image processing system may adopt other image fusion techniques, such as Spatial Pyramid Pooling (SPP), to fuse multiple remote sensing images, forming a more comprehensive and accurate feature space, enabling the generative adversarial network to capture the complex relationships between images, thereby further improving the classification accuracy. Specifically, the remote sensing image processing system obtains a spatial pyramid pooling network and inputs multiple target remote sensing images into the spatial pyramid pooling network. Then, the remote sensing image processing system divides the multiple target remote sensing images into multiple time windows based on the spatial pyramid pooling network, where the multiple time windows do not overlap, and the data within each time window usually has similar attributes or features. Subsequently, the remote sensing image processing system performs local feature extraction on the multiple time windows based on the encoder of the spatial pyramid pooling network to obtain multiple feature maps, where the encoder is any one of a convolutional neural network, a recurrent neural network, and a long short-term memory network. Then, the remote sensing image processing system splices the multiple feature maps based on the spatial pyramid pooling network to form a higher-dimensional feature representation, i.e., a spliced feature map, where this process can be achieved by simple splicing, average pooling, max pooling, etc. Then, the remote sensing image processing system decodes the spliced feature map based on the decoder of the spatial pyramid pooling network to obtain a reconstructed image. It should be noted that the decoder adopts a design similar to that of the encoder, and the decoder introduces an attention mechanism to focus on different parts of the input data. Finally, the remote sensing image processing system normalizes the reconstructed image based on the spatial pyramid pooling network to obtain a fused image.

[0105] In an alternative embodiment, the remote sensing image processing system may also adopt the weighted average method for image fusion. Specifically, the remote sensing image processing system obtains the weight coefficients corresponding to each target remote sensing image, obtaining multiple weight coefficients, and extracts multiple pixel points included in each target remote sensing image. Then, the remote sensing image processing system arbitrarily selects one target remote sensing image as the reference image among the multiple target remote sensing images. For each pixel point of the reference image, the remote sensing image processing system extracts the corresponding pixel points from the multiple target remote sensing images other than the reference image to obtain multiple target pixel points, and calculates the weighted sum using the multiple weight coefficients, the pixel point, and the multiple target pixel points to obtain a fused pixel point. In this way, the remote sensing image processing system performs weighted average calculation on each pixel point of the reference image using the multiple target remote sensing images other than the reference image to obtain multiple fused pixel points. Then, the remote sensing image processing system generates a fused image using the multiple fused pixel points.

[0106] 209. Perform post-processing on the fused image.

[0107] In the embodiments of the present application, the remote sensing image processing system performs post-processing on the fused image to optimize the quality of the fused image and make it more suitable for specific application scenarios. Among them, the post-processing includes noise removal and color correction. It should be noted that the post-processing method can be selected according to the actual application needs. For example, a filtering algorithm can be used to reduce the random noise in the fused image; the brightness and contrast of the fused image can be adjusted to make the details of the fused image more obvious; color correction can be performed to improve the visual effect and practicality of the fused image, etc. These post-processing methods can make the fused image clearer, easier to understand, and better adapt to the needs of actual applications.

[0108] 210. Obtain an initial generative adversarial network, and train the initial generative adversarial network with the fused image to obtain a target generative adversarial network.

[0109] In the embodiments of the present application, the remote sensing image processing system generates a training set, a validation set, and a test set based on the fused image. Then, the remote sensing image processing system uses the training set to train the initial generative adversarial network, and uses the test set and the validation set to test and validate the trained initial generative adversarial network respectively to obtain a target generative adversarial network. It should be noted that the embodiments of the present application use advanced algorithms such as an adaptive loss function in the training process to train the generative adversarial network in the training set, so as to better learn and map image features, thereby achieving high-precision classification and enhancement. The obtained target generative adversarial network can automatically generate real class labels, realizing the accuracy and reliability of image classification, avoiding the time and labor costs brought by manual annotation, and at the same time ensuring the consistency and accuracy of class labels.

[0110] 211. Obtain multiple remote sensing images to be classified, and input the multiple remote sensing images to be classified into the target generative adversarial network to obtain multiple class labels.

[0111] In the embodiments of the present application, the remote sensing image processing system obtains multiple remote sensing images to be classified, and inputs the multiple remote sensing images to be classified into the target generative adversarial network to obtain multiple class labels. The remote sensing image processing system combines image fusion technology and a deep learning model, which can improve the accuracy and robustness of classification, and the obtained target generative adversarial network can adapt to different scenarios and conditions, with good generalization ability and robustness. In addition, the remote sensing image processing system can achieve real-time processing and high-concurrency processing based on the target generative adversarial network, thereby improving the efficiency and accuracy of data processing, providing strong support for supply chain management, and can be widely applied in fields such as supply chain management and environmental monitoring, providing a more accurate and reliable remote sensing image processing solution for related industries.

[0112] The method provided by the embodiments of the present application obtains multiple initial remote sensing images, preprocesses the multiple initial remote sensing images to obtain multiple target remote sensing images, obtains image fusion technology and non-linear interpolation technology, stitches the multiple target remote sensing images using the image fusion technology, and performs curve fitting on the boundaries of the stitched images using the non-linear interpolation technology to obtain a fused image. An initial generative adversarial network is obtained, and the initial generative adversarial network is trained using the fused image to obtain a target generative adversarial network. Multiple remote sensing images to be classified are obtained, and the multiple remote sensing images to be classified are input into the target generative adversarial network to obtain multiple class labels. By combining the image fusion technology and the generative adversarial network, multiple remote sensing images with different angles and resolutions are fused, and the generative adversarial network is trained using the fused image, which can obtain more realistic class labels and improve the accuracy, reliability, and robustness of image classification. Therefore, it can be widely applied in fields such as supply chain management and environmental monitoring, providing a more accurate and reliable remote sensing image processing solution for related industries.

[0113] Further, as Figure 1 a specific implementation of the method, the embodiments of the present application provide a remote sensing image classification device based on supply chain management, as Figure 3A shown. The device includes: a preprocessing module 301, a fusion module 302, a training module 303, and a classification module 304.

[0114] The preprocessing module 301 is configured to obtain multiple initial remote sensing images and preprocess the multiple initial remote sensing images to obtain multiple target remote sensing images;

[0115] The fusion module 302 is configured to obtain image fusion technology and non-linear interpolation technology, stitch the multiple target remote sensing images using the image fusion technology, and perform curve fitting on the boundaries of the stitched images using the non-linear interpolation technology to obtain a fused image;

[0116] The training module 303 is configured to obtain an initial generative adversarial network and train the initial generative adversarial network using the fused image to obtain a target generative adversarial network;

[0117] The classification module 304 is configured to obtain multiple remote sensing images to be classified and input the multiple remote sensing images to be classified into the target generative adversarial network to obtain multiple class labels.

[0118] In a specific application scenario, the preprocessing module 301 is configured to obtain the coordinate information of each of the initial remote sensing images based on image processing software, resulting in a plurality of initial coordinate information; obtain a feature detection algorithm and a feature matching algorithm, use the feature detection algorithm to detect the plurality of initial coordinate information to obtain a plurality of initial image feature points, use the feature matching algorithm to perform feature point matching on the plurality of initial image feature points to obtain an initial feature point matching result, and use the initial feature point matching result to determine a plurality of translation amounts; based on the image processing software, translate the plurality of initial remote sensing images according to the plurality of translation amounts to obtain a plurality of first remote sensing images; calculate the rotation angle of each of the initial remote sensing images using the initial feature point matching result to obtain a plurality of rotation angles; based on the image processing software, rotate the plurality of first remote sensing images according to the plurality of rotation angles to obtain a plurality of second remote sensing images; obtain a target resolution, determine the resolution of each of the second remote sensing images, and calculate the scaling ratio of each of the second remote sensing images using the target resolution and the resolution of each of the second remote sensing images; obtain an interpolation algorithm, and based on the image processing software, use the interpolation algorithm to adjust each of the second remote sensing images according to the scaling ratio of each of the second remote sensing images to obtain a plurality of third remote sensing images; obtain a grayscale function, and based on the image processing software, use the grayscale function to perform grayscale processing on the plurality of third remote sensing images to obtain a plurality of fourth remote sensing images; based on the image processing software, perform denoising processing on the plurality of fourth remote sensing images to obtain a plurality of fifth remote sensing images, where the denoising processing is any one of mean filtering processing, median filtering processing, and Gaussian filtering processing; obtain an image correction model, calculate the correction parameters of each of the fifth remote sensing images based on the image correction model, and based on the image correction model, perform geometric correction processing on each of the fifth remote sensing images according to the correction parameters of each of the fifth remote sensing images to obtain a plurality of sixth remote sensing images; obtain a convolutional neural network model, and perform image enhancement on the plurality of sixth remote sensing images based on the convolutional neural network model to obtain the plurality of target remote sensing images.

[0119] In a specific application scenario, the fusion module 302 is configured to obtain the coordinate information of each of the target remote sensing images based on image processing software, so as to obtain a plurality of target coordinate information; obtain the feature detection algorithm and feature matching algorithm included in the image fusion technology, use the feature detection algorithm to detect the plurality of target coordinate information, so as to obtain a plurality of target image feature points, use the feature matching algorithm to perform feature point matching on the plurality of target image feature points, so as to obtain a target feature point matching result; splice the plurality of target remote sensing images according to the target feature point matching result, so as to obtain a spliced image, where the spliced image includes a plurality of image splicing boundaries; use the non-linear interpolation technology to perform curve fitting on the plurality of image splicing boundaries, so as to obtain the fused image.

[0120] In a specific application scenario, the fusion module 302 is configured to obtain a spatial pyramid pooling network, and input the plurality of target remote sensing images into the spatial pyramid pooling network; divide the plurality of target remote sensing images into a plurality of time windows based on the spatial pyramid pooling network, where the plurality of time windows do not overlap with each other; perform local feature extraction on the plurality of time windows based on the encoder of the spatial pyramid pooling network, so as to obtain a plurality of feature maps, where the encoder is any one of a convolutional neural network, a recurrent neural network, and a long short-term memory network; perform splicing processing on the plurality of feature maps based on the spatial pyramid pooling network, so as to obtain a spliced feature map; perform a decoding operation on the spliced feature map based on the decoder of the spatial pyramid pooling network, so as to obtain a reconstructed image, and perform normalization processing on the reconstructed image based on the spatial pyramid pooling network, so as to obtain the fused image.

[0121] In a specific application scenario, the fusion module 302 is configured to obtain the weight coefficient corresponding to each of the target remote sensing images, so as to obtain a plurality of weight coefficients, and extract a plurality of pixel points included in each of the target remote sensing images; arbitrarily select one of the target remote sensing images as a reference image from the plurality of target remote sensing images, for each pixel point of the reference image, extract the pixel point corresponding to the pixel point from the plurality of target remote sensing images other than the reference image, so as to obtain a plurality of target pixel points, calculate the weighted sum using the plurality of weight coefficients, the pixel point, and the plurality of target pixel points, so as to obtain a fused pixel point; perform weighted average calculation on each pixel point of the reference image using the plurality of target remote sensing images other than the reference image, so as to obtain a plurality of fused pixel points; generate the fused image using the plurality of fused pixel points.

[0122] In a specific application scenario, as Figure 3B shown, the device further includes: a post-processing module 305.

[0123] A post - processing module 305 is used to perform post - processing on the fused image, and the post - processing includes noise removal and color correction.

[0124] In a specific application scenario, the training module 303 is used to generate a training set, a validation set, and a test set based on the fused image; use the training set to train the initial generative adversarial network, and use the test set and the validation set to test and validate the trained initial generative adversarial network respectively to obtain the target generative adversarial network.

[0125] The device provided in the embodiments of the present application acquires multiple initial remote - sensing images, pre - processes the multiple initial remote - sensing images to obtain multiple target remote - sensing images, acquires an image fusion technology and a non - linear interpolation technology, stitches the multiple target remote - sensing images using the image fusion technology, and performs curve fitting on the stitched image boundary using the non - linear interpolation technology to obtain a fused image, acquires an initial generative adversarial network, trains the initial generative adversarial network using the fused image to obtain a target generative adversarial network, acquires multiple remote - sensing images to be classified, and inputs the multiple remote - sensing images to be classified into the target generative adversarial network to obtain multiple class labels. By combining the image fusion technology and the generative adversarial network, multiple remote - sensing images with different angles and different resolutions are fused, and the fused image is used to train the generative adversarial network, which can obtain more real class labels, improve the accuracy, reliability, and robustness of image classification, and thus be widely applied to fields such as supply - chain management and environmental monitoring, providing a more accurate and reliable remote - sensing image processing solution for related industries.

[0126] It should be noted that for other corresponding descriptions of each functional unit involved in the remote - sensing image classification device based on supply - chain management provided in the embodiments of the present application, reference can be made to Figure 1 and Figure 2 the corresponding descriptions therein, which will not be elaborated here.

[0127] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties.

[0128] The technical features of the above - mentioned embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above - mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0129] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

[0130] In an exemplary embodiment, referring to Figure 4 , a device is further provided. The device includes a bus, a processor, a memory, and a communication interface, and may further include an input / output interface and a display device. Among them, each functional unit can complete mutual communication through the bus. The memory stores a computer program, and the processor is configured to execute the program stored on the memory to execute the remote sensing image classification method based on supply chain management in the above embodiments.

[0131] A medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the remote sensing image classification method based on supply chain management are implemented.

[0132] Through the description of the above implementation manners, those skilled in the art can clearly understand that the present application can be implemented by hardware or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various implementation scenarios of the present application.

[0133] Those skilled in the art can understand that the drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the drawings are not necessarily essential for implementing the present application.

[0134] Those skilled in the art can understand that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the description of the implementation scenario, or can be correspondingly changed to be located in one or more devices different from the present implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple sub-modules.

[0135] The above serial numbers of the present application are only for description and do not represent the advantages or disadvantages of the implementation scenarios.

[0136] The above-disclosed are only several specific implementation scenarios of the present application. However, the present application is not limited thereto, and any changes that can be thought of by those skilled in the art should fall within the protection scope of the present application.

Claims

1. A remote sensing image classification method based on supply chain management, characterized in that: include: Acquire a plurality of initial remote sensing images, and preprocess the plurality of initial remote sensing images to obtain a plurality of target remote sensing images; Acquire image fusion technology and nonlinear interpolation technology, use the image fusion technology to splice the multiple target remote sensing images, and use the nonlinear interpolation technology to perform curve fitting on the boundaries of the spliced ​​images to obtain a fused image; Acquire an initial generative adversarial network, and use the fused image to train the initial generative adversarial network to obtain a target generative adversarial network; A plurality of remote sensing images to be classified are obtained, and the plurality of remote sensing images to be classified are input into the target generative adversarial network to obtain a plurality of category labels.

2. The method according to claim 1, characterized in that The preprocessing of the multiple initial remote sensing images to obtain multiple target remote sensing images includes: Acquire the coordinate information of each of the initial remote sensing images based on image processing software to obtain a plurality of initial coordinate information; Acquire a feature detection algorithm and a feature matching algorithm, use the feature detection algorithm to detect the multiple initial coordinate information to obtain multiple initial image feature points, use the feature matching algorithm to perform feature point matching on the multiple initial image feature points to obtain initial feature point matching results, and use the initial feature point matching results to determine multiple translation amounts; Based on the image processing software, performing translation processing on the multiple initial remote sensing images according to the multiple translation amounts to obtain multiple first remote sensing images; Calculating the rotation angle of each of the initial remote sensing images using the initial feature point matching result to obtain multiple rotation angles; Based on the image processing software, the plurality of first remote sensing images are rotated according to the plurality of rotation angles to obtain a plurality of second remote sensing images; Acquire a target resolution, determine a resolution of each of the second remote sensing images, and calculate a scaling ratio of each of the second remote sensing images using the target resolution and the resolution of each of the second remote sensing images; Acquire an interpolation algorithm, and based on the image processing software, use the interpolation algorithm to adjust each of the second remote sensing images according to the scaling ratio of each of the second remote sensing images to obtain a plurality of third remote sensing images; Obtaining a grayscale function, and based on the image processing software, using the grayscale function to perform grayscale processing on the plurality of third remote sensing images to obtain a plurality of fourth remote sensing images; Based on the image processing software, denoising is performed on the plurality of fourth remote sensing images to obtain a plurality of fifth remote sensing images, wherein the denoising is any one of mean filtering, median filtering, and Gaussian filtering; Acquire an image correction model, calculate a correction parameter of each of the fifth remote sensing images based on the image correction model, and perform geometric correction processing on each of the fifth remote sensing images according to the correction parameter of each of the fifth remote sensing images based on the image correction model to obtain a plurality of sixth remote sensing images; A convolutional neural network model is obtained, and image enhancement is performed on the multiple sixth remote sensing images based on the convolutional neural network model to obtain the multiple target remote sensing images.

3. The method according to claim 1, characterized in that The image fusion technology and the nonlinear interpolation technology are acquired, the image fusion technology is used to splice the multiple target remote sensing images, and the nonlinear interpolation technology is used to perform curve fitting on the spliced ​​image boundaries to obtain a fused image, including: Acquire the coordinate information of each of the target remote sensing images based on image processing software to obtain multiple target coordinate information; Acquire a feature detection algorithm and a feature matching algorithm included in the image fusion technology, use the feature detection algorithm to detect the multiple target coordinate information to obtain multiple target image feature points, use the feature matching algorithm to perform feature point matching on the multiple target image feature points to obtain target feature point matching results; splicing the multiple target remote sensing images according to the target feature point matching result to obtain a spliced ​​image, wherein the spliced ​​image includes multiple image splicing boundaries; The nonlinear interpolation technology is used to perform curve fitting on the stitching boundaries of the multiple images to obtain the fused image.

4. The method according to claim 3, characterized in that The method further comprises: Acquire a spatial pyramid pooling network, and input the multiple target remote sensing images into the spatial pyramid pooling network; Dividing the multiple target remote sensing images into multiple time windows based on the spatial pyramid pooling network, wherein the multiple time windows do not overlap each other; An encoder based on the spatial pyramid pooling network performs local feature extraction on the multiple time windows to obtain multiple feature maps, wherein the encoder is any one of a convolutional neural network, a recurrent neural network, and a long short-term memory network; Based on the spatial pyramid pooling network, the multiple feature maps are spliced ​​to obtain a spliced ​​feature map; A decoder based on the spatial pyramid pooling network performs a decoding operation on the spliced ​​feature map to obtain a reconstructed image, and a normalization process is performed on the reconstructed image based on the spatial pyramid pooling network to obtain the fused image.

5. The method according to claim 3, characterized in that: The method further comprises: Obtaining a weight coefficient corresponding to each of the target remote sensing images, obtaining a plurality of weight coefficients, and extracting a plurality of pixel points contained in each of the target remote sensing images; arbitrarily selecting a target remote sensing image from the multiple target remote sensing images as a reference image, for each pixel point of the reference image, extracting a pixel point corresponding to the pixel point from the multiple target remote sensing images other than the reference image to obtain multiple target pixel points, and calculating a weighted sum using the multiple weight coefficients, the pixel point, and the multiple target pixel points to obtain a fused pixel point; Using the multiple target remote sensing images except the reference image to perform weighted average calculation on each pixel point of the reference image, to obtain multiple fused pixel points; The fused image is generated using the multiple fused pixel points.

6. The method according to claim 1, characterized in that After the plurality of target remote sensing images are stitched together and the nonlinear interpolation technique is used to curve fit the stitched image boundaries to obtain a fused image, the method further includes: The fused image is post-processed, and the post-processing includes noise removal and color correction.

7. The method according to claim 1, characterized in that The adopting the fused image to train the initial generative adversarial network to obtain a target generative adversarial network includes: Generate a training set, a validation set and a test set based on the fused image; The initial generative adversarial network is trained using the training set, and the trained initial generative adversarial network is tested and verified using the test set and the verification set, respectively, to obtain the target generative adversarial network.

8. A remote sensing image classification device based on supply chain management, characterized in that: include: A preprocessing module is used to obtain a plurality of initial remote sensing images, and preprocess the plurality of initial remote sensing images to obtain a plurality of target remote sensing images; A fusion module is used to obtain image fusion technology and nonlinear interpolation technology, use the image fusion technology to splice the multiple target remote sensing images, and use the nonlinear interpolation technology to perform curve fitting on the spliced ​​image boundaries to obtain a fused image; A training module, used to obtain an initial generative adversarial network, and train the initial generative adversarial network using the fused image to obtain a target generative adversarial network; The classification module is used to obtain a plurality of remote sensing images to be classified, input the plurality of remote sensing images to be classified into the target generative adversarial network, and obtain a plurality of category labels.

9. A device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images