A remote sensing image extreme value inspection method, device and medium based on deep learning
By using a deep learning-based convolutional kernel model and GPU-accelerated computation, the problems of low efficiency and insufficient accuracy in traditional remote sensing image extremum detection methods are solved, achieving efficient and accurate identification and detection of extremum points in remote sensing images.
Patent Information
- Application Number
- CN202411171037.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-26
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-08-26
AI Technical Summary
Traditional remote sensing image extreme value detection methods rely on manually set thresholds or rules, which are difficult to effectively handle complex image conditions, resulting in low detection efficiency and insufficient accuracy.
We employ a deep learning-based inspection convolution kernel model, combined with GPU-accelerated computation, to identify extreme points in remote sensing images by training and optimizing the model. We use 1x1 convolution kernels for convolution operations and combine GPU parallel computing to improve detection efficiency and accuracy.
It enables efficient and accurate identification of extreme points in remote sensing images, improving detection speed and result quality, and is applicable to remote sensing images of various types and resolutions.
Smart Images

Figure CN118675064B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of remote sensing image processing, and in particular to a remote sensing image extreme value inspection method, device and medium based on deep learning. Background Art
[0002] The valid data coverage area and invalid data coverage area of remote sensing images are represented by different pixel values. For 8-bit remote sensing images, the pixel values of each band in the valid data coverage area are required to be between 1-254, and the pixel values of each band in the invalid data coverage area are required to be 0. However, during the data acquisition and data processing of remote sensing images, the pixel extreme values 0 or 255 will appear in the valid data coverage area. It is necessary to find the location of the pixel value 0 or 255 in the valid data coverage area and modify it to other specified pixel values.
[0003] In the field of remote sensing image processing, detecting and processing pixel extreme values (0 or 255) is a key step to ensure the quality of image data and the accuracy of subsequent analysis. However, traditional extreme value detection methods usually rely on manually set thresholds or rules, which may have certain limitations for complex image situations. Summary of the invention
[0004] The purpose of this application is to provide a remote sensing image extreme value inspection method, equipment and medium based on deep learning, which can improve the detection efficiency and quality.
[0005] To achieve the above objectives, this application provides the following solutions:
[0006] In a first aspect, the present application provides a remote sensing image extreme value inspection method based on deep learning, comprising:
[0007] Acquire remote sensing image data of the target area.
[0008] The remote sensing image data is input into the trained inspection convolution kernel model, and the GPU accelerated calculation method is used to obtain the position and number of each extreme point in the remote sensing image data.
[0009] Among them, the training process of checking the convolution kernel model is:
[0010] Get sample remote sensing image data.
[0011] The sample remote sensing image data is subjected to target marking to obtain marked sample remote sensing image data; the target includes the position and quantity of each extreme value point in the sample remote sensing image data.
[0012] A remote sensing image data set is constructed according to each of the labeled sample remote sensing image data; the remote sensing image data set includes a training set, a test set and a validation set divided according to a set ratio.
[0013] Based on the remote sensing image data set, the inspection convolution kernel model is trained to obtain a trained inspection convolution kernel model; wherein the inspection convolution kernel model is a convolution kernel model that uses a 1x1 convolution kernel for superposition calculation.
[0014] Optionally, the remote sensing image data is input into a trained inspection convolution kernel model, and a GPU accelerated calculation method is used to obtain the position and number of each extreme point in the remote sensing image data, specifically including:
[0015] According to the coordinate data in the remote sensing image data, based on the number of feature map channels, feature map channel index and convolution kernel type in the inspection convolution kernel model, the remote sensing image data is convolved to obtain feature maps on each channel of the remote sensing image data.
[0016] Based on the feature map, the extreme point positioning algorithm is executed on the feature map of each channel using the GPU parallel computing method to obtain the extreme value image in the remote sensing image data; the extreme value image is a local maximum value or a local minimum value.
[0017] The data format of the extreme value image is converted to obtain the position and number of each extreme value point in the remote sensing image data.
[0018] Optionally, according to the coordinate data in the remote sensing image data, based on the number of feature map channels, feature map channel index and convolution kernel type in the inspection convolution kernel model, the remote sensing image data is convolved to obtain a feature map on each channel of the remote sensing image data, specifically including:
[0019] According to the formula , and obtain the feature mapping diagram on each channel of the remote sensing image data.
[0020] in, k is the channel index of the output feature map, ( i , j , k ) represents the coordinate point in the remote sensing image data ( i , j ) is the pixel value of the kth channel after convolution, C is the number of channels of the input feature map, c is the channel index of the input feature map.
[0021] Optionally, the extreme point positioning algorithm specifically includes:
[0022] Select the convolution kernel.
[0023] The convolution kernel is slid on the remote sensing image, the elements of each channel are weighted and summed, and the result of the weighted summation is output to a new image.
[0024] Find the local maximum or minimum value as the extreme point in the new image; the extreme point includes the local maximum value and the local minimum value; the local maximum value is that the pixel values of the three bands are all maximum values and the convolution result is equal to a specific value; the local minimum value is that the pixel values of the three bands are all minimum values and the convolution result is 0.
[0025] The new image is binarized, and non-extreme points are set as background values to obtain a mask.
[0026] Use masks to filter remote sensing images and determine the locations of extreme points in remote sensing image data.
[0027] Optionally, after determining the location of the extreme points in the remote sensing image data, the following steps are further included:
[0028] Correct the pixel values of the convolution results, including:
[0029] For positions where the pixel value is equal to 0, it is corrected to a pixel value of 1.
[0030] For positions where the pixel value is equal to 255, it is corrected to the pixel value 254.
[0031] Optionally, performing data format conversion on the extreme value image to obtain the position and number of each extreme value point in the remote sensing image data specifically includes:
[0032] The extreme value image is converted from raster to vector data format, and the CGCS2000 geographic coordinate system is selected to perform projection coordinate transformation to obtain the coordinates of each extreme point of the extreme value image.
[0033] According to the coordinates of each extreme point, the SHP data format is used to display the position and number of each extreme point in the remote sensing image data.
[0034] Optionally, after obtaining the remote sensing image data of the target area, the following is also included:
[0035] The acquired remote sensing image data is preprocessed; the data preprocessing includes: image correction, radiation correction, geometric correction, atmospheric correction, denoising and smoothing.
[0036] Optionally, the remote sensing image data includes satellite images and drone images.
[0037] In a second aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any one of the above-mentioned remote sensing image extreme value inspection methods based on deep learning.
[0038] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any one of the above-mentioned methods for detecting extreme values of remote sensing images based on deep learning.
[0039] According to the specific embodiments provided in this application, this application discloses the following technical effects:
[0040] The present application provides a remote sensing image extreme value inspection method, device and medium based on deep learning, the method includes inputting remote sensing image data into a pre-trained inspection convolution kernel model, which uses the computing advantage of GPU acceleration to efficiently identify the precise position and number of each extreme value point in the image. This process improves the speed of remote sensing image processing and enhances the accuracy of the results. In the training stage of the inspection convolution kernel model, the specific position and number of extreme value points in each sample data are determined through refined target marking work based on the sample remote sensing image data, thereby constructing a high-quality remote sensing image data set. Based on the data set, the inspection convolution kernel model is trained and optimized using deep learning technology. After the training is completed, the model can accurately identify the extreme value points in the remote sensing image and quickly give information about its position and quantity. The inspection convolution kernel model in the present application is specially designed to drive the GPU for raster calculation. The operation efficiency of the model is improved, so that the extreme value inspection of the remote sensing image can be completed in a shorter time and the accuracy of the results is guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0042] Figure 1 This is an application environment diagram of a remote sensing image extreme value inspection method based on deep learning in one embodiment of the present application.
[0043] Figure 2 A flowchart of a remote sensing image extreme value inspection method based on deep learning is provided in one embodiment of the present application.
[0044] Figure 3 A schematic diagram of 1x1 convolution kernel superposition calculation provided in one embodiment of the present application.
[0045] Figure 4 A schematic diagram of an extreme point checking algorithm provided in an embodiment of the present application.
[0046] Figure 5 This is a Gaofen-1 multispectral satellite image provided in one embodiment of the present application.
[0047] Figure 6 This is a diagram of an extreme value check graphical interface provided in an embodiment of the present application.
[0048] Figure 7 This is an image extreme value distribution effect diagram provided by an embodiment of the present application.
[0049] Figure 8 This is a diagram showing the inspection result and image extreme value range matching effect provided in one embodiment of the present application.
[0050] Fig. 9 This is a SHP result diagram of extreme value check provided in an embodiment of the present application.
[0051] Fig.10 This is a diagram showing the inspection result and image extreme value range matching effect provided in one embodiment of the present application.
[0052] Fig.11 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0053] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0054] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0055] The remote sensing image extreme value inspection method based on deep learning provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, or it can be integrated on the server 104, or it can be placed on the cloud or other servers. The terminal 102 can send the remote sensing image data of the target area to be processed to the server 104. After the server 104 receives the remote sensing image data of the target area to be processed, for the remote sensing image data of the target area to be processed, the server 104 obtains the remote sensing image data of the target area; the remote sensing image data is input into the trained inspection convolution kernel model, and the GPU accelerated calculation method is used to obtain the position and number of each extreme point in the remote sensing image data. The server 104 can feedback the position and number of each extreme point in the obtained remote sensing image data to the terminal 102. In addition, in some embodiments, a remote sensing image extreme value inspection method based on deep learning can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly perform data processing on the remote sensing image data of the target area to be processed, or the server 104 can obtain the remote sensing image data of the target area to be processed from the data storage system and perform data processing on the remote sensing image data of the target area to be processed.
[0056] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.
[0057] In an exemplary embodiment, a remote sensing image extreme value inspection method based on deep learning is provided. The method is executed by a computer device, which can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following steps 101 to 102. Among them:
[0058] Step 101: Acquire remote sensing image data of a target area.
[0059] Step 102: input the remote sensing image data into the trained inspection convolution kernel model, and use a GPU accelerated calculation method to obtain the position and number of each extreme point in the remote sensing image data.
[0060] When executing step 101, the specific steps may be as follows:
[0061] Obtain satellite images and / or unmanned aerial vehicle (UAV) images of the target area, which may be a city, farmland, or ocean.
[0062] After remote sensing image data, it also includes:
[0063] A series of processing and optimization operations are performed on the acquired remote sensing image data to improve data quality and usability. The preprocessing process includes image correction, radiation correction, geometric correction, atmospheric correction, denoising and smoothing, etc., which is to eliminate and reduce noise, artifacts and other unnecessary interference in the image data, while enhancing the characteristics and details of the image.
[0064] Before executing step 102, a convolution kernel model is also trained and checked, such as Figure 2 As shown, the training process of checking the convolution kernel model is:
[0065] Step 201, obtaining sample remote sensing image data.
[0066] Step 202, target marking is performed on the sample remote sensing image data to obtain marked sample remote sensing image data; the target includes the position and number of each extreme point in the sample remote sensing image data.
[0067] Step 203, constructing a remote sensing image data set according to each of the labeled sample remote sensing image data; the remote sensing image data set includes a training set, a test set and a validation set divided according to a set ratio.
[0068] Step 204: Based on the remote sensing image data set, the inspection convolution kernel model is trained to obtain a trained inspection convolution kernel model; wherein the inspection convolution kernel model is a convolution kernel model that uses a 1x1 convolution kernel for superposition calculation.
[0069] When executing step 201, the specific steps may be as follows:
[0070] Collect remote sensing image data from different sensors, resolutions, and spectral ranges, including satellite images and unmanned aerial vehicle (UAV) images. In order to improve the generalization ability of the model, the collected data is divided into different categories, such as cities, farmlands, oceans, etc.
[0071] After executing step 201, the method further includes:
[0072] A series of processing and optimization operations are performed on the acquired remote sensing image data to improve data quality and usability. The preprocessing process includes image correction, radiation correction, geometric correction, atmospheric correction, denoising and smoothing, etc., which is to eliminate and reduce noise, artifacts and other unnecessary interference in the image data, while enhancing the characteristics and details of the image.
[0073] Before executing step 204, the method further includes: designing and checking a convolution kernel.
[0074] Specifically, determine the design concept of the inspection convolution kernel: Inspection convolution kernel design is a key part in model design. The selection of the size, shape and number of convolution kernels directly affects the performance and feature extraction capabilities of the model. Common convolution kernel sizes include 3x3, 5x5 and 1x1. The size of the convolution kernel determines the size of the receptive field covered when it slides on the input. Smaller convolution kernels can capture detailed information. Considering the characteristics of extreme values of remote sensing images, that is, areas with pixel values of 0 or 25, it is necessary to find the locations where the pixel values of extreme values of 0 or 255 appear in the valid data area. This embodiment proposes a convolution kernel design scheme dedicated to inspection, using a 1x1 convolution kernel to extract image features, which can ensure that the size of the feature map generated after convolution is consistent with the original feature size, so that the image detail information can be captured more completely.
[0075] The 1x1 convolution kernel is actually a special convolution kernel that does not expand in space and only performs convolution operations in the channel dimension. Therefore, its calculation formula is relatively simple. For each pixel point (i, j, k) of the output feature map, the calculation process is:
[0076] .
[0077] in, k is the channel index of the output feature map, ( i , j , k ) represents the coordinate point in the remote sensing image data ( i , j ) is the pixel value of the kth channel after convolution. This means that the value of each pixel in the output feature map is calculated by weighted summing the input feature map in the channel dimension. C is the number of channels of the input feature map, c is the channel index of the input feature map.
[0078] like Figure 3 As shown in the figure, taking a multispectral image as an example, it contains three channels, representing the pixel values of the three primary colors of RGB. The input is (3, 3, 3), which represents three channels, each with a width of 3 and a height of 3. Assume that there is only one convolution kernel, the convolution kernel channel is 1, the convolution kernel size of each channel is still 1x1, padding=0, and stride=1. The convolution process is as follows Figure 3As shown in the figure, the leftmost is the input feature map matrix, with 3 channels and a sliding stride of 1 each time. The pixel value of each channel is convolved with the value of the corresponding convolution kernel channel, so each channel will correspond to an output convolution result. The three convolution results are accumulated and summed at the corresponding positions to obtain the final convolution result. Here, the convolution output result channel has only 1 because there is only 1 convolution kernel. The final convolution result is the comprehensive information result of each channel of the original image.
[0079] Then the network structure is optimized: Under the premise of ensuring the performance of the model, by designing and adjusting the structure of the convolutional neural network, the inference order of the model is optimized, and the model parameters are reduced, the complexity of the model is reduced as much as possible to improve the performance, efficiency and generalization ability of the network, making it more suitable for parallel computing on the GPU and improving the model calculation speed.
[0080] In some embodiments, when executing step 102, the specific steps may be as follows:
[0081] According to the coordinate data in the remote sensing image data, based on the number of feature map channels, feature map channel index and convolution kernel type in the inspection convolution kernel model, the remote sensing image data is convolved to obtain feature maps on each channel of the remote sensing image data.
[0082] Based on the feature map, the extreme point positioning algorithm is executed on the feature map of each channel using the GPU parallel computing method to obtain the extreme value image in the remote sensing image data; the extreme value image is a local maximum value or a local minimum value.
[0083] The data format of the extreme value image is converted to obtain the position and number of each extreme value point in the remote sensing image data.
[0084] Among them, according to the coordinate data in the remote sensing image data, based on the number of feature map channels, feature map channel index and convolution kernel type in the inspection convolution kernel model, the remote sensing image data is convolved to obtain the feature map on each channel of the remote sensing image data, specifically including:
[0085] According to the formula , and obtain the feature mapping diagram on each channel of the remote sensing image data.
[0086] in, k is the channel index of the output feature map, ( i , j , k ) represents the coordinate point in the remote sensing image data ( i , j ) is the pixel value of the kth channel after convolution, C is the number of channels of the input feature map,c is the channel index of the input feature map.
[0087] The extreme point location algorithm specifically includes:
[0088] Step 301, select a convolution kernel.
[0089] Step 302, slide the convolution kernel on the remote sensing image, perform weighted summation on the elements of each channel, and output the result of the weighted summation to a new image.
[0090] Step 303, searching for a local maximum or minimum value as an extreme point in the new image; the extreme point includes a local maximum value and a local minimum value; the local maximum value is when the pixel values of the three bands are all maximum values and the convolution result is equal to a specific value; the local minimum value is when the pixel values of the three bands are all minimum values and the convolution result is 0.
[0091] Step 304: binarize the new image and set non-extreme value points as background values to obtain a mask.
[0092] Step 305: Use the mask to filter the remote sensing image and determine the position of the extreme point in the remote sensing image data.
[0093] Specifically, the extreme value detection method based on the convolution kernel can identify the local maximum or local minimum in the image through convolution operation, thereby detecting significant features or abnormalities in the image. This algorithm not only considers the characteristics of a single pixel, but also considers the relationship between surrounding pixels, thereby improving the accuracy of detection. The specific algorithm steps are as follows: Figure 4 shown.
[0094] 1) Select a suitable convolution kernel, usually a small two-dimensional convolution kernel, a 1x1 kernel. The convolution kernel slides over the image and performs convolution operations with the image to highlight local features in the image.
[0095] 2) Apply the selected convolution kernel to the image. By sliding the convolution kernel on the image, according to the convolution kernel processing principle, the element values in each channel are weighted summed, and the result of the convolution operation is output to a new image, and the output image is controlled to a band. Each pixel value represents a local feature in the original image that matches the shape of the convolution kernel.
[0096] 3) Find the local maximum value (the value after the convolution operation is 765 when the pixel values of the three bands are all 255) or the local minimum value (the value after the convolution operation is 0 when the pixel values of the three bands are all 0) as the extreme point in the result image of the convolution operation. This can be achieved by performing a binarization process on the image, and then setting the value of the non-image extreme value t1 in the single-band image of the convolution kernel processing result to the background value, any non-image extreme value constant t2, to obtain a mask. Then use the mask to screen the extreme value t1 in the image, and then realize the extreme value inspection.
[0097] in Figure 4 W in the convolution kernel represents the weight (the number of input images * input channels * convolution kernel size 1*1); ini:1: indicates that the output convolution result matrix channel is 1; B: pixel value of the convolution result; t1: image extreme value; X: pixel value of the convolution result; t2: any non-image extreme value.
[0098] After determining the location of the extreme points in the remote sensing image data, it also includes:
[0099] Correct the pixel values of the convolution results, including:
[0100] For positions where the pixel value is equal to 0, it is corrected to a pixel value of 1.
[0101] For positions where the pixel value is equal to 255, it is corrected to the pixel value 254.
[0102] Specifically, once an extreme point is detected according to the extreme point location algorithm, it will be corrected to improve data quality. During the correction process, the pixel value of 0 is changed to 1, and the pixel value of 255 is changed to 254, which minimizes the impact on the original image.
[0103] The data format of the extreme value image is converted to obtain the position and quantity of each extreme value point in the remote sensing image data, specifically including:
[0104] The extreme value image is converted from raster to vector data format, and the CGCS2000 geographic coordinate system is selected to perform projection coordinate transformation to obtain the coordinates of each extreme point of the extreme value image.
[0105] According to the coordinates of each extreme point, the SHP data format is used to display the position and number of each extreme point in the remote sensing image data.
[0106] Specifically, in order to facilitate batch inspection and intuitive display of output results, the extracted extreme value images need to be converted from raster to vector data format, and the CGCS2000 geographic coordinate system is selected to perform projection coordinate transformation, and finally the converted coordinates are output to the output results in SHP data format. This embodiment provides an intuitive result display interface, and the final output result is a vector data result in SHP data format, so that users can easily view the location and number of extreme value points. In addition, this embodiment also generates a detailed extreme value point report, which includes the number, location and recommended treatment measures of extreme value points.
[0107] Specifically, this application provides an embodiment for illustration:
[0108] The following uses the No. 3 stereo satellite image of a city as a data source to quickly extract the image extreme value position, and the present invention is described in detail. The specific implementation steps are as follows:
[0109] Step 1) Data preparation: Using the Gaofen-1 multispectral satellite image as the data source, the original remote sensing image data is obtained and preprocessed using the data preparation steps. The processed data is loaded into ArcGIS software for symbolization, and the background color is displayed using RGB (0, 0, 0). Figure 5 As shown in the figure, the area with pixel value (0, 0, 0) within the image range is the extreme value area that needs to be checked.
[0110] Step 2) Design and train the inspection convolution kernel model. Design a 1x1 convolution kernel model. On the basis of ensuring the original image characteristics, namely resolution, perform dimensionality reduction processing on the original image, that is, three channels become one channel. Use GPU for accelerated calculation so that the model can learn the characteristics and rules of extreme value detection.
[0111] Step 3) This experiment uses the TensorFlow framework to implement GPU-accelerated parallel computing, which significantly improves the efficiency of image extremum detection. The premise of the experiment is to ensure that CUDA and CUDNN are correctly installed. These two technologies are necessary conditions for realizing NVIDIA GPU-accelerated computing. During the model building phase, pay special attention to deploying the model to the GPU to perform training tasks to fully utilize the computing power of the GPU. If there are multiple GPUs in the experimental environment, users can independently select the GPU device to run the model and flexibly configure the use strategy of GPU resources. Specifically, the training process can be limited to the first GPU execution by setting, and the dynamic growth function of GPU memory can be enabled to adapt to changes in memory requirements during training.
[0112] Step 4) Extreme value check: This embodiment provides an intuitive extreme value check interface, such as Figure 6As shown, users can easily and batch-wise check image extreme values, use GPU for raster calculations, and quickly and effectively detect the location and type of extreme values in images.
[0113] Step 5) Output the results. The final output result is vector data in SHP data format. Fig. 9 As shown, users can easily and intuitively view the quantity and quickly locate the extreme point position. Figure 7 , 8 As shown in Figure 10, the output extreme value spatial position and the image can be completely overlapped, which facilitates the analysis and interpretation of the detected extreme points to determine whether they represent significant features or abnormalities in the image. The meaning and significance of the extreme points can be further analyzed by combining image context information and domain knowledge.
[0114] Finally, the extreme value check efficiency is compared. This experiment tests and compares the efficiency of ArcGIS extreme value check method, numpy array method, and fully connected neural network algorithm. The specific check efficiency is shown below. The output of the convolutional neural network method is divided into numpy.ndarray and TIF image. Both are tested in this experiment.
[0115] Among them, the numpy array method, the fully connected neural network algorithm, and the convolutional neural network method output are numpy.ndarray, and the efficiency is shown in Table 1:
[0116] Table 1 Output is numpy.ndarray Check efficiency comparison
[0117]
[0118] Among them, the numpy array method, the fully connected neural network method, and the convolutional neural network method output the TIF image inspection efficiency as shown in Table 2:
[0119] Table 2 Comparison of inspection efficiency when output as TIF image
[0120]
[0121] Research conclusion: By using the inspection convolution kernel model and GPU accelerated computing, it is possible to quickly inspect and locate extreme values in remote sensing images and greatly improve detection efficiency.
[0122] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Fig.11As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the position and number of each extreme point in the remote sensing image data. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a remote sensing image extreme value inspection method based on deep learning is implemented.
[0123] Those skilled in the art will understand that Fig.11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0124] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.
[0125] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.
[0126] 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 used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0127] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0128] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.
[0129] In summary, this application has the following beneficial effects:
[0130] 1) Compared with the traditional threshold or rule-based extreme value detection method, this application introduces a detection convolution kernel model and uses GPU accelerated computing technology to achieve efficient and rapid detection of extreme values in remote sensing images, thereby greatly improving the detection efficiency.
[0131] 2) Significantly improved quality accuracy: Thanks to the use of deep learning models, this application has demonstrated higher accuracy and reliability in the extreme value inspection process, effectively improving the quality level of the test results.
[0132] 3) Widely enhanced applicability: The method proposed in this application has strong versatility and wide applicability, and can flexibly cope with various types and resolutions of remote sensing images, providing a more convenient and effective solution for practical applications.
[0133] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0134] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A remote sensing image extreme value inspection method based on deep learning, characterized in that: The remote sensing image extreme value inspection method comprises: Acquire remote sensing image data of the target area; The remote sensing image data is input into the trained inspection convolution kernel model, and the position and number of each extreme point in the remote sensing image data are obtained by using a GPU accelerated calculation method; Among them, the training process of checking the convolution kernel model is: Obtain sample remote sensing image data; Performing target marking on the sample remote sensing image data to obtain marked sample remote sensing image data; the target includes the position and number of each extreme point in the sample remote sensing image data; Constructing a remote sensing image data set according to the labeled sample remote sensing image data; the remote sensing image data set includes a training set, a test set and a validation set divided according to a set ratio; Based on the remote sensing image data set, the inspection convolution kernel model is trained to obtain a trained inspection convolution kernel model; wherein the inspection convolution kernel model is a convolution kernel model that uses a 1x1 convolution kernel for superposition calculation; The remote sensing image data is input into the trained inspection convolution kernel model, and the GPU accelerated calculation method is used to obtain the position and number of each extreme point in the remote sensing image data, specifically including: According to the coordinate data in the remote sensing image data, based on the number of feature map channels, feature map channel index and convolution kernel type in the inspection convolution kernel model, the remote sensing image data is convolved to obtain a feature map on each channel of the remote sensing image data; Based on the feature map, using a GPU parallel computing method, an extreme point positioning algorithm is executed on the feature map of each channel to obtain an extreme value image in the remote sensing image data; the extreme value image is a local maximum value or a local minimum value; Performing data format conversion on the extreme value image to obtain the position and quantity of each extreme value point in the remote sensing image data; Among them, according to the coordinate data in the remote sensing image data, based on the number of feature map channels, feature map channel index and convolution kernel type in the inspection convolution kernel model, the remote sensing image data is convolved to obtain the feature map on each channel of the remote sensing image data, specifically including: According to the formula , obtaining a feature map on each channel of the remote sensing image data; in, k is the channel index of the output feature map, ( i , j , k ) represents the coordinate point in the remote sensing image data ( i , j ) k The pixel value of the channel after convolution, C is the number of channels of the input feature map, c is the channel index of the input feature map.
2. The remote sensing image extreme value inspection method based on deep learning according to claim 1, characterized in that: The extreme point positioning algorithm specifically includes: Select the convolution kernel; Sliding the convolution kernel on the remote sensing image, weighted summing the elements of each channel, and outputting the weighted summing result to a new image; Find the local maximum or minimum value in the new image as an extreme point; the extreme point includes a local maximum value and a local minimum value; the local maximum value is when the pixel values of the three bands are all maximum values and the convolution result is equal to a specific value; the local minimum value is when the pixel values of the three bands are all minimum values and the convolution result is 0; Binarization is performed on the new image, and non-extreme value points are set as background values to obtain a mask; Use masks to filter remote sensing images and determine the locations of extreme points in remote sensing image data.
3. The remote sensing image extreme value inspection method based on deep learning according to claim 2 is characterized in that: After determining the location of the extreme points in the remote sensing image data, it also includes: Correct the pixel values of the convolution results, including: For the position where the pixel value is equal to 0, it is corrected to the pixel value 1; For positions where the pixel value is equal to 255, it is corrected to the pixel value 254.
4. The remote sensing image extreme value inspection method based on deep learning according to claim 1 is characterized in that: The data format of the extreme value image is converted to obtain the position and number of each extreme value point in the remote sensing image data, specifically including: The extreme value image is converted from a raster to a vector data format, and the CGCS2000 geographic coordinate system is selected to perform a projection coordinate transformation to obtain the coordinates of each extreme point of the extreme value image; According to the coordinates of each extreme point, the SHP data format is used to display the position and number of each extreme point in the remote sensing image data.
5. The remote sensing image extreme value inspection method based on deep learning according to claim 1 is characterized in that: After obtaining the remote sensing image data of the target area, it also includes: The acquired remote sensing image data is preprocessed; the data preprocessing includes: image correction, radiation correction, geometric correction, atmospheric correction, denoising and smoothing.
6. The remote sensing image extreme value inspection method based on deep learning according to claim 1 is characterized in that: The remote sensing image data includes satellite images and drone images.
7. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a remote sensing image extreme value inspection method based on deep learning as described in any one of claims 1 to 6.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements a remote sensing image extreme value detection method based on deep learning as described in any one of claims 1-6.
