Automatic classification method and system for lunar impact craters

Abstract

This application provides an automatic classification method and system for lunar impact craters, applicable to the field of computer image processing technology. This method addresses the technical problems of low accuracy and low robustness in lunar impact crater classification in related technologies. It includes: adaptively setting dual threshold parameters for an edge detection algorithm corresponding to the image signal-to-noise ratio (SNR) to obtain an adaptive edge detection algorithm for adaptive edge detection of a remote sensing single-channel grayscale image; performing gradient detection on the remote sensing single-channel grayscale image based on a preset first-order differential operator to obtain a fused edge image; adaptively generating fusion weights based on the image SNR; extracting features from the fused edge image and the remote sensing single-channel grayscale image respectively to obtain multi-dimensional feature data; and weighting and summing the normalized multi-dimensional features according to the dimensional weights adaptively generated based on the image SNR to obtain a comprehensive freshness score, thereby improving the robustness and accuracy of lunar impact crater classification.

Description

Technical Field

[0001] This application relates to the field of computer image processing technology, and more particularly to the interdisciplinary field of planetary science remote sensing and computer vision, and to a method for classifying remote sensing images, and more specifically to an automatic classification method and system for lunar impact craters. Background Technology

[0002] The classification of impact crater morphology is a fundamental task in lunar science research and is of great significance to multiple research fields.

[0003] Traditional impact crater classification mainly relies on visual interpretation by experts based on high-resolution remote sensing images. This method is inefficient, highly subjective, and has limited accuracy.

[0004] Currently, there are methods for crater classification based on computer vision. However, these methods mainly focus on the classification of impact crater morphology, such as simple craters, complex craters, and ring basins. There is limited research on binary classification of fresh and degraded craters. Even when such classification is possible, it suffers from insufficient discriminative power due to its reliance on single features. For example, lunar dust cover alters spectral properties rather than geometric morphology, making edge detection-based methods unable to distinguish between "degraded craters with clear edges but covered by lunar dust." Furthermore, existing methods do not consider the differences in contribution of different features, directly superimposing features of different dimensions such as edge density and roundness. This leads to high-value features dominating classification decisions, marginalizing other important features, resulting in low accuracy and robustness in classification. Summary of the Invention

[0005] In view of the above problems, this application provides an automatic lunar impact crater classification method and system that can improve feature attention, improve the classification accuracy and robustness of lunar impact craters.

[0006] According to the first aspect of this application, an automatic classification method for lunar impact craters is provided, comprising: preprocessing pre-acquired lunar remote sensing images to obtain a remote sensing single-channel grayscale image; calculating the image signal-to-noise ratio (SNR) of the remote sensing single-channel grayscale image; adaptively setting dual threshold parameters of an edge detection algorithm corresponding to the image SNR based on the image SNR to obtain an adaptive edge detection algorithm; performing adaptive edge detection on the remote sensing single-channel grayscale image based on the adaptive edge detection algorithm to obtain a first edge image, a second edge image, and a third edge image of three scales of interest respectively; performing gradient detection on the remote sensing single-channel grayscale image based on a preset first-order differential operator to obtain a gradient edge image; adaptively generating fusion weights for the first edge image, the second edge image, the third edge image, and the gradient edge image based on the image SNR; performing a weighted summation of the first edge image, the second edge image, the third edge image, and the gradient edge image based on the fusion weights to obtain a fused edge image; and performing fusion... Feature extraction was performed on edge images and remote sensing single-channel grayscale images to obtain multidimensional feature data, including basic features and specific features. Basic features included edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, and local contrast. Specific features included crater edge sharpness attenuation rate, radiometric texture anisotropy index, and crater bottom-crater wall gradient discontinuity index. The multidimensional feature data was normalized to obtain normalized multidimensional features. The normalized multidimensional features were weighted and summed according to the dimension weights adaptively generated based on the image signal-to-noise ratio to obtain a freshness comprehensive score. The categories of lunar impact craters in the lunar remote sensing image were determined based on the freshness comprehensive score and a preset judgment threshold, and the corresponding classification labels were output. The classification labels were verified on multiple scales based on a preset multi-scale verification algorithm to obtain the classification confidence result. The classification confidence result indicated that the multi-scale verification passed, and the classification label was used as the target crater category in the lunar remote sensing image.

[0007] The second aspect of this application provides an automatic classification system for lunar impact craters, comprising: a preprocessing module for preprocessing pre-acquired lunar remote sensing images to obtain a single-channel grayscale image; an edge extraction module for calculating the image signal-to-noise ratio (SNR) of the single-channel grayscale image, adaptively setting dual threshold parameters of an edge detection algorithm corresponding to the SNR to obtain an adaptive edge detection algorithm, and performing adaptive edge detection on the single-channel grayscale image based on the adaptive edge detection algorithm to obtain a first edge image, a second edge image, and a third edge image of three different scales; a fusion calculation module for performing gradient detection on the single-channel grayscale image based on a preset first-order differential operator to obtain a gradient edge image, and adaptively generating fusion weights for the first edge image, the second edge image, the third edge image, and the gradient edge image based on the image SNR, and performing a weighted summation of the first edge image, the second edge image, the third edge image, and the gradient edge image based on the fusion weights to obtain a fused edge image; and a feature extraction module. The system comprises three modules: a module for extracting features from the fused edge image and the remote sensing single-channel grayscale image, resulting in multi-dimensional feature data. This multi-dimensional feature data includes basic features and specific features. Basic features include edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, and local contrast. Specific features include crater edge sharpness attenuation rate, radiation texture anisotropy index, and crater bottom-crater wall gradient discontinuity index. A scoring calculation module normalizes the multi-dimensional feature data to obtain normalized multi-dimensional features. The normalized multi-dimensional features are then weighted and summed based on dimensional weights adaptively generated according to the image signal-to-noise ratio to obtain a comprehensive freshness score. A category output module determines the category of lunar impact craters in the lunar remote sensing image based on the comprehensive freshness score and a preset threshold, and outputs the corresponding classification label. The classification label is then validated at multiple scales using a preset multi-scale validation algorithm to obtain a classification confidence result. If the classification confidence result indicates that the multi-scale validation has passed, the classification label is used as the target crater category in the lunar remote sensing image.

[0008] According to the automatic lunar impact crater classification method and system of this application, the image signal-to-noise ratio (SNR) of a remote sensing single-channel grayscale image is calculated. Based on the SNR, dual threshold parameters of an edge detection algorithm corresponding to the SNR are adaptively set to obtain an adaptive edge detection algorithm. Adaptive edge detection is then performed on the remote sensing single-channel grayscale image based on this algorithm, resulting in first, second, and third edge images of three different scales. Gradient detection is then performed based on a preset first-order differential operator to obtain gradient edge images. This parallel operation of the edge detection algorithm and the first-order differential operator compensates for the omissions in edge detection by a single algorithm / operator, improving the robustness and feature coverage of the fused edge image. Feature extraction is then performed on the fused edge image to obtain multi-dimensional feature data, which includes basic features and specific features. The system involves nine features, ensuring the presence of both features sensitive to overall shape and those sensitive to local damage, thus guaranteeing the comprehensiveness of the feature data. This more comprehensively highlights the distinction between fresh and degraded impact craters. Furthermore, by using dimensional weights corresponding to the image signal-to-noise ratio to weighted summation of normalized multi-dimensional features, a comprehensive freshness score is obtained. This effectively eliminates dimensional differences in feature data, achieving efficient integration of multi-source information. During the integration process, the contribution of each multi-dimensional feature data is differentiated, distinguishing the degree of attention paid to different feature data, thereby improving the accuracy and scientific rigor of the overall impact crater classification. In addition, the system outputs not only the category of lunar impact craters but also the classification confidence level, providing a direct display of the classification's credibility and enhancing the level of intelligent decision-making support for lunar research. Attached Figure Description

[0009] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0010] Figure 1 This illustration schematically depicts an application scenario of the automatic lunar impact crater classification method, system, and device according to embodiments of this application.

[0011] Figure 2 A flowchart illustrating an automatic lunar impact crater classification method according to an embodiment of this application is shown schematically.

[0012] Figure 3 This schematic diagram illustrates the data flow involved in the automatic classification method for lunar impact craters according to an embodiment of this application.

[0013] Figure 4 This illustration schematically shows the operation flow diagram of the automatic classification method for lunar impact craters according to embodiments of this application, including cases where multi-scale verification passes and fails.

[0014] Figure 5 This schematic diagram illustrates the structural block diagram of an automatic lunar impact crater classification system according to an embodiment of this application;

[0015] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing an automatic lunar impact crater classification method according to an embodiment of this application. Detailed Implementation

[0016] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0017] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0018] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0019] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0020] Impact crater morphology classification is fundamental to lunar science research and is of great significance to multiple research fields. In geochronological studies, the density distribution of fresh impact craters effectively reflects the exposure age of the lunar surface, providing crucial evidence for establishing a lunar geological timeline. From the perspective of lunar regolith evolution, the degree of degradation of impact craters can serve as an important indicator of the thickness and formation history of the lunar regolith. In impact history reconstruction, statistical analysis of impact craters with different morphologies can reveal the patterns of impact flux changes experienced by the lunar surface, thereby reconstructing the early impact environment evolution of the Moon. Furthermore, at the engineering application level, impact crater morphology classification is crucial for risk assessment of lunar landing sites. By identifying the integrity of impact craters, safety assurance can be provided for lander site selection and lunar surface exploration activities, ensuring the smooth implementation of engineering missions.

[0021] Lunar impact craters are bowl-shaped or basin-shaped landforms formed by high-speed impacts from small celestial bodies such as meteorites and comets. Their morphological characteristics record the geological evolution history of the lunar surface. Based on the degree of morphological preservation, impact craters can be divided into fresh impact craters, which are relatively new and have clearly defined morphological features, and degraded impact craters, which have undergone long-term alteration and have blurred morphological features. Fresh impact craters typically exhibit the following characteristics: sharp rim outlines, steep crater walls, distinct crater floor shadows, near-perfect circular shapes, and high grayscale contrast between the rim and floor. Degraded impact craters, on the other hand, are characterized by blunted rims, collapsed crater walls, infilled crater floors, irregular shapes, and reduced grayscale contrast due to the effects of space weathering, solar wind bombardment, micrometeorite erosion, cosmic ray radiation, tectonic activity, and subsequent impacts.

[0022] Based on this, an automatic classification method and system for lunar impact craters is proposed in the embodiments of this application.

[0023] Figure 1 The diagram illustrates an application scenario of the automatic lunar impact crater classification method according to an embodiment of this application.

[0024] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0025] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0026] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0027] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0028] It should be noted that the automatic lunar impact crater classification method provided in this application embodiment can be executed by server 105. Correspondingly, the automatic lunar impact crater classification method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and is capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or server 105.

[0029] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0030] The following will be based on Figure 1 The described scene, through Figures 2-4 The automatic classification method for lunar impact craters according to the embodiments of the application is described in detail.

[0031] Figure 2 A flowchart illustrating an automatic classification method for lunar impact craters according to an embodiment of this application is shown.

[0032] like Figure 2 As shown, the automatic classification method for lunar impact craters in this embodiment includes operations S210 to S260, which specifically include the following:

[0033] In operation S210, the pre-acquired lunar remote sensing image is preprocessed to obtain a single-channel grayscale image. In operation S220, the image signal-to-noise ratio (SNR) of the single-channel grayscale image is calculated. Based on the SNR, the dual threshold parameters of the edge detection algorithm corresponding to the SNR are adaptively set to obtain an adaptive edge detection algorithm. Adaptive edge detection is then performed on the single-channel grayscale image based on the adaptive edge detection algorithm to obtain first, second, and third edge images of three different scales. In operation S230, gradient detection is performed on the single-channel grayscale image based on a preset first-order differential operator to obtain a gradient edge image. Based on the image SNR, fusion weights are adaptively generated for the first, second, and third edge images and the gradient edge image. The first, second, and third edge images and the gradient edge image are then weighted and summed based on the fusion weights to obtain a fused edge image. In operation S240, the fused edge image and the remote sensing image are processed together. Feature extraction is performed on the single-channel grayscale image to obtain multi-dimensional feature data, which includes basic features and specific features. Basic features include edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, and local contrast. Specific features include crater edge sharpness attenuation rate, radiation texture anisotropy index, and crater bottom-crater wall gradient discontinuity index. In operation S250, the multi-dimensional feature data is normalized to obtain normalized multi-dimensional features. The normalized multi-dimensional features are weighted and summed according to the dimension weights adaptively generated based on the image signal-to-noise ratio to obtain a comprehensive freshness score. In operation S260, the category of lunar impact craters in the lunar remote sensing image is determined based on the comprehensive freshness score and a preset judgment threshold, and the classification label corresponding to the category is output. The classification label is verified on multiple scales based on a preset multi-scale verification algorithm to obtain the classification confidence result. In response to the classification confidence result indicating that the multi-scale verification is passed, the classification label is used as the target crater category of the lunar remote sensing image.

[0034] As an example, the first step is to batch read lunar remote sensing images. These images can be in any format, such as TIFF (Tag Image File Format) or IMG (Image). Preprocessing is then performed on the lunar remote sensing images, including grayscale conversion and linear stretching, to obtain a single-channel grayscale image. Next, the signal-to-noise ratio (SNR) of the single-channel grayscale image is calculated using the following formula:

[0035] ; Represented by a base of 10 The logarithm of ;

[0036] in, This represents the average grayscale value of the image signal region in a single-channel grayscale image from remote sensing. Let $\mathbf{x}$ be the standard deviation of the grayscale values ​​in the background noise region of the remote sensing single-channel grayscale image. Then, based on this image signal-to-noise ratio (SNR), corresponding dual-threshold parameters can be adaptively set to obtain an adaptive edge detection algorithm. Thus, by performing Canny edge detection with different parameter settings on remote sensing single-channel grayscale images with different SNRs, we obtain first, second, and third edge images at three different scales. That is, Canny edge detection is performed at three scales of the remote sensing single-channel grayscale image, with the Gaussian filter standard deviations being:

[0037] To obtain the first edge image respectively Second edge image Third edge image .

[0038] Then, based on the first-order differential operator Sobel, Sobel gradient detection is performed on the remote sensing single-channel grayscale image, and binarization is performed using the median of the gradient as a threshold to obtain the gradient edge image. Based on the aforementioned SNR, the fusion weights are adaptively determined and weighted fusion is performed to obtain the fused edge image. For example, the fusion weights can take different values ​​depending on the signal-to-noise ratio of the image:

[0039] when hour:

[0040] ;

[0041] when hour:

[0042] ;

[0043] when hour, ;

[0044] Adaptive threshold binarization is performed on the fused edge intensity map, and the threshold is... ,in, The mean value of the blending edge strength. The standard deviation of the fused edge intensity is used to obtain the fused edge image, which then lays the data foundation for subsequent quantization of features to obtain multidimensional feature data.

[0045] After obtaining the fused edge image, features are extracted from both the fused edge image and the remote sensing single-channel grayscale image to obtain multidimensional feature data. This multidimensional feature data includes basic features and specific features. Basic features include edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, and local contrast. Specific features include pit edge sharpness attenuation rate, radiation texture anisotropy index, and pit bottom-pit wall gradient discontinuity index. Specifically, for basic morphological features:

[0046] Edge pixel ratio is obtained by calculating the ratio of the number of edge pixels to the total number of pixels in the fused edge image; fused edge density is obtained by dividing the fused edge image into an 8×8 grid and calculating the ratio of the standard deviation to the mean of the edge pixel density in each grid; maximum circularity is obtained by detecting all circular structures through the Hough transform detection mechanism, calculating the ratio of the perimeter of the edge pixels on the circumference of each circle to the theoretical circumference, and taking the maximum value; maximum circle ratio is obtained by calculating the ratio of the area of ​​the detected largest circle to the area of ​​the total region; average gradient is obtained by calculating the average gradient magnitude of the original grayscale image in the fused edge region; local contrast is obtained by calculating the grayscale difference inside and outside the fused edge region.

[0047] For specific features, namely impact crater features, the crater edge sharpness attenuation rate is determined by selecting N sampling points (N≥24) along the circumference of the main circle in the remote sensing single-channel grayscale image, and then performing an analysis on each sampling point along the normal direction. Within a pixel range (L≥8), a gradient sequence is sampled. An exponential decay model is fitted to the gradient outside the circle, and the average decay coefficient of N sampling points is calculated. The radiometric texture anisotropy index is obtained by extracting an annular region with a radius from r to 2r (r is the radius of the main circle) from the remote sensing single-channel grayscale image, dividing the region into M sectors (M≥24), and counting the number of edge pixels within each sector. Based on this and by performing proportional calculations, the gradient discontinuity index of the pit bottom to the pit wall is obtained by sampling gradient sequences from the center of the circle outward in the remote sensing single-channel grayscale image along K radial directions (K≥8), detecting the jump amplitude ΔG at the gradient abrupt change point, and calculating the average jump amplitude in K directions.

[0048] The aforementioned multidimensional feature data is then normalized to obtain normalized multidimensional features. These features are then weighted and summed based on dimension weights adaptively generated according to the image signal-to-noise ratio (SNR) to obtain a comprehensive freshness score. The weighted combination data, including the dimension weights, varies depending on the image SNR, such as:

[0049] The nine features are normalized to the [0, 1] interval using a pre-defined min-max method; feature weights are adaptively determined based on the image quality of the remote sensing single-channel grayscale image.

[0050] when At that time, the weights of the basic features are as follows:

[0051] ;

[0052] The specific feature weights are as follows: ;

[0053] when At that time, the weights of the basic features are as follows:

[0054] ;

[0055] Specialized feature weights ;

[0056] when At that time, the weights of the basic features are as follows:

[0057] ;

[0058] Specialized feature weights ;

[0059] The weighted sum of the scores can be used directly as the overall freshness score, or it can be used as an initial score. The initial score, plus any bonus items, becomes the final overall freshness score. ,in .

[0060] The categories of lunar impact craters in lunar remote sensing images are determined based on the overall freshness score and a preset judgment threshold, and the corresponding classification labels are output. For example, when the overall freshness score F ≥ the judgment threshold... ( Range of values When the freshness score is [value missing], it is considered a fresh impact crater. At that time, it was determined to be a degraded impact crater.

[0061] This example addresses the technical problems of existing automatic classification methods, such as single feature set, imperfect fusion mechanism, low classification efficiency, and low classification accuracy. It constructs a systematic quantitative evaluation system for impact crater morphology. By extracting multi-dimensional feature data and performing adaptive weighted summation, it eliminates dimensional differences and distinguishes the degree of attention of different feature data. This enables binary classification decisions for impact crater classification, outputting the target crater category for fresh and degraded impact craters. This ensures the existence of feature data sensitive to both overall shape and local damage, thus guaranteeing the comprehensiveness of the feature data and more comprehensively highlighting the differences between fresh and degraded impact craters. The differences between the two are identified, and the normalized multidimensional features are weighted and summed based on the dimensional weights adaptively generated according to the image signal-to-noise ratio to obtain a comprehensive freshness score. This allows for the targeted elimination of dimensional differences in feature data, achieving effective integration of multi-source information. Furthermore, the contribution of each multidimensional feature data is differentiated during the integration process, distinguishing the degree of attention paid to different feature data, thereby improving the accuracy and scientific rigor of the overall impact crater classification. Multi-scale verification is introduced to obtain classification confidence results. If the classification confidence results represent the multi-scale verification as passing, the classification label is used as the target crater category of the lunar remote sensing image, thus improving the accuracy of classification.

[0062] In this embodiment, the pre-acquired lunar remote sensing image is preprocessed to obtain a single-channel grayscale image, including: performing grayscale processing on the lunar remote sensing influence using a preset weighted average algorithm to obtain a single-channel grayscale image; and performing linear stretching processing on the single-channel grayscale image for contrast enhancement to obtain a single-channel grayscale image.

[0063] As an example, this preprocessing includes grayscale conversion and linear stretching. Grayscale conversion involves using a weighted average method to convert the lunar remote sensing image (RGB image, Red, Green, Blue, or mixed red-green-blue pixel image) into a single-channel grayscale image. The formula for calculating the single-channel grayscale image is as follows:

[0064] ;

[0065] In this context, R represents red pixels, G represents green pixels, and B represents blue pixels.

[0066] Linear stretching: Employs 2% saturation stretching to enhance contrast, mapping grayscale values ​​to... The interval is used to obtain a single-channel grayscale image from remote sensing.

[0067] Based on this, image standardization processing was completed to provide a data foundation for subsequent quantitative feature analysis, so as to improve the gray-scale contrast between the edge and the bottom of the crater, enhance the distinction between gray levels, and thus improve the accuracy of subsequent impact crater classification.

[0068] In this embodiment, in response to an image signal-to-noise ratio greater than a preset second signal-to-noise threshold, the dual threshold parameters are set to a first dual threshold parameter having a first low threshold and a first high threshold with the longest numerical range; in response to an image signal-to-noise ratio less than the second signal-to-noise threshold but greater than the preset first signal-to-noise threshold, the dual threshold parameters are set to a second dual threshold parameter having a second low threshold and a second high threshold with a medium numerical range; in response to an image signal-to-noise ratio less than the first signal-to-noise threshold, the dual threshold parameters are set to a third dual threshold parameter having a third low threshold and a third high threshold with the shortest numerical range.

[0069] As an example, based on the calculations above The value adaptively sets the dual threshold parameters of the edge detection algorithm. This edge detection algorithm can be the Canny algorithm (a multi-level edge detection algorithm), which adaptively adjusts the dual threshold parameters of the Canny algorithm based on the image signal-to-noise ratio. More specifically, for example:

[0070] when When setting a low threshold for the Canny algorithm. Canny algorithm's high threshold ;

[0071] when When, set ;

[0072] when When, set .

[0073] Thus, Canny edge detection with different parameter settings is performed on remote sensing single-channel grayscale images with different signal-to-noise ratios, resulting in first, second, and third edge images focusing on three scales respectively. Specifically, Canny edge detection is performed on the three scales of the remote sensing single-channel grayscale image using a dual threshold adapted to the image. The Gaussian filter standard deviations are as follows: To obtain the first edge image respectively Second edge image Third edge image This improves the targeting and effectiveness of edge detection, thereby enhancing the effectiveness and accuracy of impact crater classification while increasing the accuracy of edge image detection.

[0074] In this embodiment, adaptive edge detection is performed on a remote sensing single-channel grayscale image based on an adaptive edge detection algorithm to obtain a first edge image, a second edge image, and a third edge image focusing on three scales respectively. This includes: performing detailed-scale edge detection on the remote sensing single-channel grayscale image using an edge detection algorithm with a first double threshold parameter, a second double threshold parameter, or a third threshold parameter to obtain a first edge image. The first edge image is a detailed-scale image used to capture the crater edge details and radial patterns; performing medium-scale edge detection on the remote sensing single-channel grayscale image to obtain a second edge image. The second edge image is a medium-scale image used to capture the edge structure of the crater; and performing coarse-scale edge detection on the remote sensing single-channel grayscale image to obtain a third edge image. The third edge image is a coarse-scale image used to preserve the main edge and crater edge contour of the crater.

[0075] As an example, the signal-to-noise ratio (SNR) of the image is calculated as follows: a 20×20 pixel region inside the impact crater is selected as the signal region; 10×10 pixel regions at each of the four corners of the image are selected as the noise region; signal mean: Noise standard deviation: ; Image quality is moderate, when When, set That is, by using an edge detection algorithm with a second double threshold parameter to perform detailed-scale edge detection on a remote sensing single-channel grayscale image, a first edge image, a second edge image, and a third edge image are obtained. More specifically, the detailed scale of the first edge image is... After the process of "Gaussian filtering, gradient calculation, non-maximum suppression, and double thresholding," a detailed-scale image with 458,932 edge pixels can be obtained, which can capture subtle textures such as pit edge details and radial patterns; if the second edge image is of medium scale... The number of edge pixels can be 312567, resulting in a medium-scale image. Its characteristic is that it can filter out some noise while preserving the main edge structure; however, the third edge image is at a coarse scale. Its edge pixel count can be 187234, resulting in a coarse-scale image that retains only the strongest subject edges and has a clear pit edge outline.

[0076] Based on this, edge images at detailed, intermediate, and coarse scales can be derived specifically based on edge detection, thereby improving the overall feature coverage of edge images and enhancing the robustness and accuracy of subsequent lunar impact crater classification.

[0077] Figure 3 The diagram illustrates the data flow involved in the automatic classification method for lunar impact craters according to an embodiment of this application.

[0078] exist Figure 3In the illustrated embodiment, the multidimensional feature data is normalized to obtain normalized multidimensional features. The normalized multidimensional features are then weighted and summed according to the dimensional weights adaptively generated based on the image signal-to-noise ratio to obtain a comprehensive freshness score. This includes normalizing the edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, local contrast, pit edge sharpness attenuation rate, radial texture anisotropy index, and pit bottom-pit wall gradient discontinuity index to obtain normalized multidimensional features. These normalized multidimensional features include normalized values ​​for edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, local contrast, pit edge sharpness attenuation rate, radial texture anisotropy index, and pit bottom-pit wall gradient discontinuity index. The following data are used to determine the normalized values: average gradient, local contrast, pit edge sharpness attenuation rate, anisotropy index, and discontinuity index. Weighted combination data corresponding to the image signal-to-noise ratio are then determined. Based on the dimensional weights corresponding to each normalized multidimensional feature in the weighted combination data, the normalized values ​​for edge pixel proportion, fused edge density, maximum circularity, maximum circularity ratio, average gradient, local contrast, pit edge sharpness attenuation rate, anisotropy index, and discontinuity index are weighted and summed to obtain the freshness score. The freshness score is then summed with the pre-acquired reward items to obtain the comprehensive freshness score.

[0079] In a specific example, the proportion of edge pixels After normalization, the normalized value of the edge pixel proportion is... Edge density fusion involves dividing the image into an 8×8 grid and calculating the edge pixel density of each grid: mean density: Density standard deviation: The average density Normalized fusion edge density normalization value Maximum roundness Three circular structures were detected using the Hough detection mechanism: Circle 1: Center (2048, 2035), radius 24px, circumference edge pixels 126, theoretical circumference 151px; circularity = 126 / 151 = 0.834; Circle 2: Center (2080, 2050), radius 8px, circularity 0.612; Circle 3: Center (2010, 2020), radius 6px, circularity 0.523; Normalized maximum circularity value Maximum circle ratio The area of ​​the largest circle (circle 1) is: The total area is: ; Normalized maximum circle ratio Average gradient Calculate the gradient magnitude and mean of the original grayscale image in the blending edge region (377,645 pixels): Normalized average gradient normalization value Local contrast Mean value within the edge region: Mean value outside the marginal region: ; Normalized local contrast value .

[0080] For impact crater-specific features, the crater edge sharpness attenuation rate Its main circle (circle 1): center (2048, 2035), radius 24px; 36 sampling points are selected along the circumference, and for each point: the gradient is sampled in the range of [-10, +10]px along the normal direction; the model is fitted to the gradient outside the circle (d≥0): the least squares method is used to fit the model. In a more specific example, the fitting result for the sampling point (θ=0°) can be: Average attenuation coefficient of 36 sampling points: Normalized pit edge sharpness attenuation rate normalized value The physical explanation is as follows: The crater edge exhibits rapid gradient decay and sharp boundaries, consistent with the characteristics of a fresh impact crater; the radial texture shows anisotropy index Its annular region has an inner radius of 24px and an outer radius of 48px; it is divided into 36 10° sectors, and the number of edge pixels of each sector is counted.

[0081] sector angle ;

[0082] edge pixels ;

[0083] Statistic: , ;

[0084] ;

[0085] Normalized heterosex index normalized value The physical explanation is as follows: The radial patterns exhibit a directional distribution, consistent with the characteristics of a fresh impact crater;

[0086] Pit bottom-pit wall gradient discontinuity index Along 8 radial directions Sampling from the center outwards; in a specific example, direction gradient sequence:

[0087] distance ;

[0088] gradient ;

[0089] Gradient abrupt change point detection: The maximum value of the second derivative appears at... Place;

[0090] Mutation magnitude: ;

[0091] ;

[0092] Normalized discontinuous exponent normalization value The physical explanation is as follows: There is a clear gradient change at the junction of the crater bottom and the crater wall, which is consistent with the bowl-shaped structure of a fresh impact crater.

[0093] Thus, the normalized nine-dimensional feature vector is:

[0094] ;

[0095] In this example, because (Medium quality) This data uses a weighted combination with a medium quality weighting configuration. The dimensional weights in this weighted combination correspond one-to-one with the normalized values ​​for edge pixel proportion, blended edge density, maximum circularity, maximum circle ratio, average gradient, local contrast, pit edge sharpness attenuation rate, anisotropy index, and discontinuity index, as detailed below:

[0096] Basic feature weights: ;

[0097] Specialized feature weights: ;

[0098] Complete weight vector: ;

[0099] The rating, or freshness rating, can be calculated as follows:

[0100] ;

[0101] The contribution analysis of each item is as follows:

[0102] Basic feature contribution:

[0103] ;

[0104] Specialized feature contribution: .

[0105] In this embodiment, the freshness comprehensive score includes the above-mentioned score and preset reward items. Thus, the freshness comprehensive score can not only ensure the existence of multi-dimensional feature data that is sensitive to the overall shape, but also the existence of reward items related to circle recognition, thereby ensuring the comprehensiveness of feature data and more comprehensively enhancing the distinction between fresh impact craters and degraded impact craters.

[0106] In this embodiment, obtaining the reward item includes: using median filtering to enhance the morphological features of the fused edge image to obtain a feature-enhanced image; performing circle detection on the feature-enhanced image based on a preset circle detection mechanism to obtain undetermined circles, and classifying the undetermined circles to obtain the number of valid circles including valid circles among the undetermined circles, where valid circles are undetermined circles with a circularity greater than a preset circle threshold; and performing a product operation based on the number of valid circles and a preset single circle scoring coefficient to obtain the reward item.

[0107] As an example, after obtaining the fused edge image, a 5×5 window median filter is used to smooth the image and remove salt-and-pepper noise, providing high-quality input for subsequent circle detection. In this example, when detecting impact crater morphology, the circle detection mechanism is the Hough detection mechanism, that is, the Hough detection mechanism is introduced as a morphological verification. If a circular structure that conforms to the impact crater scale is detected, such as a radius of 6-45 pixels, it is regarded as a circle to be determined. Considering the accuracy of circle detection, in this example, circles to be determined with a circularity greater than a circle threshold are regarded as valid circles. This circle threshold can be... Assuming a score of 0.6, for Circle 1: 0.834 ≥ 0.6; for Circle 2: 0.612 ≥ 0.6; and for Circle 3: 0.523 < 0.6, it can be seen that Circle 1 and Circle 2 are valid circles. Based on the number of valid circles, bonus items can be added to the freshness score. For example, a single valid circle can receive a bonus score of 0.1. If there are multiple circular structures, the bonus item is obtained by multiplying the number of valid circles by the scoring coefficient of a single circle. This scoring coefficient of a single circle is also the bonus weight, which can be 0.1. For example, if the number of circles is 2, the bonus item is 0.2.

[0108] Based on this, we strengthen the typical circular features, namely the discrimination contribution of the effective circle, and reduce the misclassification rate of elliptical or irregular pits.

[0109] In this embodiment, in response to a freshness score greater than or equal to a threshold, the lunar impact crater in the lunar remote sensing image is classified as a fresh impact crater; in response to a freshness score less than the threshold, the lunar impact crater in the lunar remote sensing image is classified as a degraded impact crater.

[0110] In a specific example, the threshold for this determination is 0.5. In this example, when F ≥ 0.5, it is determined to be a fresh impact crater; when F < 0.5, it is determined to be a degraded impact crater.

[0111] In this example, the threshold of 0.5 is not generated out of thin air. In this example, the freshness comprehensive score is obtained through a pre-trained freshness comprehensive scoring model. This freshness comprehensive scoring model compares the freshness comprehensive score with the threshold. When training the freshness comprehensive scoring model, the threshold [0.3, 0.7] is traversed on the validation set with a step size of 0.05. The threshold corresponding to the maximum value of the training comprehensive score output by the freshness comprehensive scoring model during the training phase is selected as the threshold. For example, if the maximum value of the training comprehensive score F1 = 0.89, the corresponding threshold is 0.5.

[0112] Based on this, a classification decision is made based on the judgment threshold, outputting the category of the lunar impact crater and the corresponding classification label, such as fresh or degraded. Simultaneously, the above-mentioned freshness score, comprehensive freshness score, and quantified multidimensional feature data can also be output. In this way, the binary classification of lunar impact craters is completed with high precision, improving the accuracy and robustness of lunar impact crater freshness classification, and providing reliable data support for lunar soil thickness inversion, geochronological research, and lunar surface morphology mapping.

[0113] In this embodiment, the classification confidence result is the ratio obtained by dividing the difference between the freshness comprehensive score and the judgment threshold by the judgment threshold. In response to the classification confidence result indicating that the multi-scale verification failed, the lunar remote sensing image is scaled by a preset ratio to obtain a reduced image and a magnified image. The freshness comprehensive scores corresponding to the reduced image and the magnified image are obtained respectively, serving as the reduced comprehensive score and the magnified comprehensive score. The reduced comprehensive score, the magnified comprehensive score, and the freshness comprehensive score corresponding to the lunar remote sensing image are weighted and summed according to preset comprehensive score weights to obtain a confidence comprehensive score. The confidence label of the category corresponding to the confidence comprehensive score is used as the target crater category of the lunar remote sensing image.

[0114] As an example, while outputting the classification label, it also outputs the classification confidence result obtained by calculating the distance between the freshness comprehensive score and the decision threshold. This distance calculation is the ratio obtained by dividing the difference between the two by the decision threshold. Based on this ratio, it can intuitively represent the difference between the freshness comprehensive score and the training comprehensive score output by the freshness comprehensive score model during the training phase, and thus intuitively reflect the level of confidence of its classification.

[0115] Figure 4 The diagram illustrates the operation flow of the automatic classification method for lunar impact craters according to embodiments of this application, including cases where multi-scale verification passes and fails.

[0116] exist Figure 4In the example shown, if the confidence level is low, multi-scale verification is triggered, such as scaling the lunar remote sensing image by 0.5x and 2x, repeating the steps in the example above to obtain: a 0.5x scaled image (2048×2048px): , , ,score ; 2x scaling of the image (8192×8192px): , , ,score The weighted average confidence score ;

[0117] A confidence score of ≥0.5 indicates a fresh impact crater.

[0118] Therefore, by introducing multi-scale verification to ensure the accuracy of classification labels, if the classification confidence result of the multi-scale verification fails the verification, a more rigorous comprehensive confidence score is obtained by weighted averaging, thereby further improving the accuracy and robustness of impact crater classification.

[0119] In this embodiment, a pre-trained freshness comprehensive scoring model is used to normalize multidimensional feature data to obtain normalized multidimensional features. The normalized multidimensional features are then weighted and summed according to dimensional weights adaptively generated based on image signal-to-noise ratio to obtain a comprehensive freshness score. The pre-training of the freshness comprehensive scoring model includes: repeatedly training a pre-constructed freshness comprehensive scoring architecture using pre-built samples, and iteratively optimizing the classification and weight parameters of the freshness comprehensive scoring architecture using a controlled variable method until the improvement of the objective function of the freshness comprehensive scoring architecture is less than a pre-set training threshold. The freshness comprehensive scoring architecture learned in the last training iteration is then used as the freshness comprehensive scoring model for the application stage, and the classification and weight parameters of the last trained freshness comprehensive scoring architecture are used as the classification and weight parameters of the freshness comprehensive scoring model. Specifically, when optimizing the classification parameters, the weight parameters are fixed; when optimizing the weight parameters, the classification parameters are fixed.

[0120] As an example, a feedback mechanism is established to achieve adaptive parameter optimization. When training the freshness comprehensive scoring architecture, 200 randomly selected samples can be used, such as 100 fresh samples and 100 degraded samples. Then, based on the preset expert annotations and the output freshness comprehensive score and classification labels during training, the accuracy, precision, and recall of the freshness comprehensive scoring architecture are calculated. Iterative optimization is then performed using a controlled variable method, such as fixing the weights (i.e., fixing the weight parameters) and adjusting the threshold parameters (i.e., adjusting the classification parameters). The search range of the threshold parameters can be set to an initial value ±20%. Alternatively, the threshold can be fixed (i.e., fixing the classification parameters), and the coefficients of the weight parameters can be adjusted. For example, the constraint condition can be set to the sum of weights = 1. For the objective function, select the optimal combination of classification parameters and weight parameters, and iterate repeatedly until... If the improvement is less than 1%, the iteration stops, and the classification parameters and weight parameters of the freshness comprehensive scoring architecture learned in the last training session are used as the classification parameters and weight parameters of the freshness comprehensive scoring model. In this example, the classification parameters in this training process are related to the judgment threshold in the above-mentioned application process. That is, the optimal classification parameters generated during the training process are used as the judgment threshold for the freshness comprehensive scoring model, and the weight parameters are the dimensional weights corresponding to the image signal-to-noise ratio of different categories.

[0121] In this way, a closed loop of iterative optimization and verification is established to avoid the decline in generalization ability caused by parameter solidification, improve the generalization ability of the freshness comprehensive scoring model in the application stage, and improve the accuracy and robustness of impact crater binary classification.

[0122] As described above, the automatic lunar impact crater classification method of this application calculates the image signal-to-noise ratio (SNR) of a single-channel grayscale image from remote sensing. Based on the SNR, it adaptively sets dual threshold parameters for an edge detection algorithm corresponding to the SNR, obtaining an adaptive edge detection algorithm. Then, it performs adaptive edge detection on the single-channel grayscale image from remote sensing, obtaining first, second, and third edge images of three different scales. Finally, it performs gradient detection based on a preset first-order differential operator to obtain gradient edge images. Therefore, the edge detection algorithm and the first-order differential operator operate in parallel, which can compensate for the missed detections of a single algorithm / operator, improving the robustness and feature coverage of the fused edge image. Feature extraction is then performed on the fused edge image to obtain multi-dimensional feature data, which includes basic features and specific features. The system employs nine features, ensuring the presence of both features sensitive to overall shape and those sensitive to local damage. This comprehensiveness enhances the distinction between fresh and degraded impact craters. Furthermore, by weighting the normalized multi-dimensional features according to the image signal-to-noise ratio, a comprehensive freshness score is obtained. This effectively eliminates dimensional differences in feature data, enabling efficient integration of multi-source information. During integration, the system differentiates the contribution of each multi-dimensional feature, distinguishing the degree of attention given to different features, thereby improving the accuracy and scientific rigor of the overall impact crater classification. In addition to outputting the lunar impact crater category, the system also outputs the classification confidence level, providing a clear visual representation of the classification's credibility and enhancing the level of intelligent decision-making support for lunar research.

[0123] like Figure 5As shown, this application also provides an automatic lunar impact crater classification system 500, which includes: a preprocessing module 510 for preprocessing pre-acquired lunar remote sensing images to obtain a remote sensing single-channel grayscale image; an edge extraction module 520 for calculating the image signal-to-noise ratio (SNR) of the remote sensing single-channel grayscale image, adaptively setting dual threshold parameters of the edge detection algorithm corresponding to the image SNR to obtain an adaptive edge detection algorithm, and performing adaptive edge detection on the remote sensing single-channel grayscale image based on the adaptive edge detection algorithm to obtain a first edge image, a second edge image, and a third edge image focusing on three scales respectively; a fusion calculation module 530 for performing gradient detection on the remote sensing single-channel grayscale image based on a preset first-order differential operator to obtain a gradient edge image, and adaptively generating fusion weights for the first edge image, the second edge image, the third edge image, and the gradient edge image based on the image SNR, and performing a weighted summation on the first edge image, the second edge image, the third edge image, and the gradient edge image based on the fusion weights to obtain a fused edge image; feature extraction; and other functions. Module 540 is used to extract features from the fused edge image and the remote sensing single-channel grayscale image to obtain multi-dimensional feature data. The multi-dimensional feature data includes basic features and special features. The basic features include edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, and local contrast. The special features include crater edge sharpness attenuation rate, radiation texture anisotropy index, and crater bottom-crater wall gradient discontinuity index. The scoring calculation module 550 is used to normalize the multi-dimensional feature data to obtain normalized multi-dimensional features. The normalized multi-dimensional features are weighted and summed according to the dimension weights adaptively generated based on the image signal-to-noise ratio to obtain a comprehensive freshness score. The category output module 560 is used to determine the category of lunar impact craters in the lunar remote sensing image based on the comprehensive freshness score and a preset judgment threshold, and output the classification label corresponding to the category. The classification label is verified on multiple scales based on a preset multi-scale verification algorithm to obtain the classification confidence result. In response to the classification confidence result indicating that the multi-scale verification has passed, the classification label is used as the target crater category of the lunar remote sensing image.

[0124] In this embodiment, the preprocessing module 510 includes: a grayscale processing unit, used to perform grayscale processing on the lunar remote sensing influence using a preset weighted average algorithm to obtain a single-channel grayscale image; and an enhancement processing unit, used to perform linear stretching processing on the remote sensing single-channel grayscale image with respect to contrast enhancement to obtain a remote sensing single-channel grayscale image.

[0125] The edge extraction module 520, in response to an image signal-to-noise ratio greater than a preset second signal-to-noise threshold, sets the dual threshold parameters to a first dual threshold parameter having a first low threshold and a first high threshold having the longest numerical range; in response to an image signal-to-noise ratio less than the second signal-to-noise threshold but greater than the preset first signal-to-noise threshold, sets the dual threshold parameters to a second dual threshold parameter having a second low threshold and a second high threshold having a medium numerical range; in response to an image signal-to-noise ratio less than the first signal-to-noise threshold, sets the dual threshold parameters to a third dual threshold parameter having a third low threshold and a third high threshold having the shortest numerical range.

[0126] The edge extraction module 520 performs adaptive edge detection on a remote sensing single-channel grayscale image based on an adaptive edge detection algorithm, obtaining a first edge image, a second edge image, and a third edge image at three different scales. This includes: performing detailed-scale edge detection on the remote sensing single-channel grayscale image using an edge detection algorithm with a first double threshold parameter, a second double threshold parameter, or a third threshold parameter to obtain a first edge image, which is a detailed-scale image used to capture the crater's edge details and radial patterns; performing medium-scale edge detection on the remote sensing single-channel grayscale image to obtain a second edge image, which is a medium-scale image used to capture the crater's edge structure; and performing coarse-scale edge detection on the remote sensing single-channel grayscale image to obtain a third edge image, which is a coarse-scale image used to preserve the main edge and crater edge contour of the crater.

[0127] The scoring calculation module 550 normalizes the multidimensional feature data to obtain normalized multidimensional features. It then performs a weighted summation of these normalized multidimensional features based on dimensional weights adaptively generated according to the image signal-to-noise ratio. This includes normalizing edge pixel proportion, fused edge density, maximum circularity, maximum circularity ratio, average gradient, local contrast, pit edge sharpness attenuation rate, radial texture anisotropy index, and pit bottom-pit wall gradient discontinuity index. The normalized multidimensional features include normalized values ​​for edge pixel proportion, fused edge density, maximum circularity, maximum circularity ratio, average gradient, local contrast, pit edge sharpness attenuation rate, anisotropy index, and discontinuity index. Finally, it determines the weighted combination data corresponding to the image signal-to-noise ratio and calculates the weighted combination data based on its relationship with each normalized multidimensional feature. The corresponding dimensional weights are used to weight and sum the normalized values ​​of edge pixel proportion, fused edge density, maximum circularity, maximum circle ratio, average gradient, local contrast, pit edge sharpness attenuation rate, anisotropy index, and discontinuity index to obtain a freshness score. The freshness score is then summed with the pre-acquired reward items to obtain a comprehensive freshness score. The reward items include: using median filtering to enhance the morphological features of the fused edge image to obtain a feature-enhanced image; performing circle detection on the feature-enhanced image based on a preset circle detection mechanism to obtain undetermined circles, classifying the undetermined circles to obtain the number of valid circles that include valid circles (valid circles are those with a circularity greater than a preset circle threshold); and multiplying the number of valid circles with a preset individual circle scoring coefficient to obtain the reward item.

[0128] The category output module 560 determines the category of lunar impact craters in the lunar remote sensing image as fresh impact craters when the comprehensive freshness score is greater than or equal to the judgment threshold; and determines the category of lunar impact craters in the lunar remote sensing image as degraded impact craters when the comprehensive freshness score is less than the judgment threshold. The classification confidence result involved is the ratio obtained by dividing the difference between the comprehensive freshness score and the judgment threshold by the judgment threshold. In response to the classification confidence result indicating that the multi-scale verification failed, the lunar remote sensing image is scaled by a preset ratio to obtain a reduced image and a magnified image. The comprehensive freshness scores corresponding to the reduced image and the magnified image are obtained respectively as the reduced comprehensive score and the magnified comprehensive score. The reduced comprehensive score, the magnified comprehensive score, and the comprehensive freshness score corresponding to the lunar remote sensing image are weighted and summed according to the preset comprehensive score weight to obtain the confidence comprehensive score. The confidence label of the category corresponding to the confidence comprehensive score is used as the target crater category of the lunar remote sensing image.

[0129] The output module 560 of this category normalizes the multidimensional feature data using a pre-trained freshness comprehensive scoring model to obtain normalized multidimensional features. It then performs a weighted summation of the normalized multidimensional features based on dimension weights adaptively generated according to the image signal-to-noise ratio to obtain the freshness comprehensive score. The pre-training of the freshness comprehensive scoring model includes: repeatedly training a pre-constructed freshness comprehensive scoring architecture using pre-built samples, and iteratively optimizing the classification and weight parameters of the freshness comprehensive scoring architecture using a controlled variable method until the improvement of the objective function of the freshness comprehensive scoring architecture is less than a pre-set training threshold. The freshness comprehensive scoring architecture learned in the last training iteration is then used as the freshness comprehensive scoring model for the application stage, and its classification and weight parameters are used as the classification and weight parameters of the freshness comprehensive scoring model. Specifically, when optimizing the classification parameters, the weight parameters are fixed; when optimizing the weight parameters, the classification parameters are fixed.

[0130] According to embodiments of this application, the preprocessing module 510, edge extraction module 520, fusion calculation module 530, feature extraction module 540, scoring calculation module 550, and category output module 560 can be implemented in one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the preprocessing module 510, edge extraction module 520, fusion calculation module 530, feature extraction module 540, scoring calculation module 550, and category output module 560 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the preprocessing module 510, edge extraction module 520, fusion calculation module 530, feature extraction module 540, scoring calculation module 550, and category output module 560 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0131] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing an automatic lunar impact crater classification method according to an embodiment of this application.

[0132] like Figure 6As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor, and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0133] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0134] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0135] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0136] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0137] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the automatic lunar impact crater classification method provided in the embodiments of this application.

[0138] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0139] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0140] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0141] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0142] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0143] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0144] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A method for automatic classification of lunar impact craters, characterized in that, include: Preprocessing is performed on the pre-acquired lunar remote sensing images to obtain a single-channel grayscale image; The signal-to-noise ratio (SNR) of the remote sensing single-channel grayscale image is calculated. Based on the SNR, the dual-threshold parameters of the edge detection algorithm corresponding to the SNR are adaptively set to obtain an adaptive edge detection algorithm. Adaptive edge detection is then performed on the remote sensing single-channel grayscale image based on the adaptive edge detection algorithm to obtain a first edge image, a second edge image, and a third edge image at three different scales. Specifically, in response to the SNR being greater than a preset second SNR threshold, the dual-threshold parameters are set to a first dual-threshold parameter having a first low threshold and a first high threshold with the longest numerical range. In response to the SNR being less than the second SNR threshold but greater than the preset first SNR threshold, the dual-threshold parameters are set to a second dual-threshold parameter having a second low threshold and a second high threshold with a medium numerical range. In response to the SNR being less than the first SNR threshold, the dual-threshold parameters are set to a third dual-threshold parameter having a third low threshold and a third high threshold with the shortest numerical range. Gradient detection is performed on the remote sensing single-channel grayscale image based on a preset first-order differential operator to obtain a gradient edge image. Then, based on the image signal-to-noise ratio, fusion weights are adaptively generated for the first edge image, the second edge image, the third edge image, and the gradient edge image. The first edge image, the second edge image, the third edge image, and the gradient edge image are weighted and summed based on these fusion weights to obtain a fused edge image. The first edge image is a detail-scale image used to capture the crater's edge details and radial patterns; the second edge image is a medium-scale image used to capture the crater's edge structure; and the third edge image is a coarse-scale image used to preserve the crater's main edge and crater edge contour. Feature extraction is performed on the fused edge image and the remote sensing single-channel grayscale image to obtain multidimensional feature data. The multidimensional feature data includes basic features and special features. The basic features include edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, and local contrast. The special features include pit edge sharpness attenuation rate, radiation texture anisotropy index, and pit bottom-pit wall gradient discontinuity index. The multidimensional feature data is normalized to obtain normalized multidimensional features. The normalized multidimensional features are then weighted and summed according to the dimension weights adaptively generated based on the image signal-to-noise ratio to obtain a comprehensive freshness score. The categories of lunar impact craters in the lunar remote sensing image are determined based on the freshness comprehensive score and the preset judgment threshold, and the classification labels corresponding to the categories are output. The classification labels are then subjected to multi-scale verification based on the preset multi-scale verification algorithm to obtain the classification confidence result. In response to the classification confidence result indicating that the multi-scale verification has passed, the classification label is used as the target crater category of the lunar remote sensing image.

2. The method according to claim 1, characterized in that, The preprocessing of the pre-acquired lunar remote sensing images to obtain a single-channel grayscale image includes: The lunar remote sensing effects are processed into grayscale using a preset weighted average algorithm to obtain a single-channel grayscale image; The remote sensing single-channel grayscale image is subjected to linear stretching for contrast enhancement to obtain the remote sensing single-channel grayscale image.

3. The method according to claim 2, characterized in that, The first edge image is obtained by performing detail-scale edge detection on the remote sensing single-channel grayscale image using an edge detection algorithm with a first dual threshold parameter; The second edge image is obtained by performing medium-scale edge detection on the remote sensing single-channel grayscale image using an edge detection algorithm with a second dual threshold parameter; The third edge image is obtained by performing coarse-scale edge detection on the remote sensing single-channel grayscale image using an edge detection algorithm with a third double threshold parameter.

4. The method according to claim 1, characterized in that, The step of normalizing the multidimensional feature data to obtain normalized multidimensional features, and then weighting and summing the normalized multidimensional features according to the dimension weights adaptively generated based on the image signal-to-noise ratio, includes: The edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, local contrast, pit edge sharpness attenuation rate, radial texture anisotropy index, and pit bottom-pit wall gradient discontinuity index are normalized to obtain normalized multidimensional features. The normalized multidimensional features include normalized values ​​for edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, local contrast, pit edge sharpness attenuation rate, anisotropy index, and discontinuity index. Determine the weighted combination data corresponding to the signal-to-noise ratio of the image, and perform a weighted summation on the normalized values ​​of edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, local contrast, pit edge sharpness attenuation rate, anisotropy index, and discontinuity index according to the dimension weights corresponding to each normalized multidimensional feature in the weighted combination data, to obtain the freshness score; The freshness score is summed with the pre-acquired reward items to obtain the overall freshness score.

5. The method according to claim 4, characterized in that, Obtaining the aforementioned reward items includes: The fused edge image is enhanced with morphological features using median filtering to obtain a feature-enhanced image. The feature enhancement image is subjected to circle detection based on a preset circle detection mechanism to obtain undetermined circles. The undetermined circles are then classified to obtain the number of valid circles that include valid circles. The valid circles are undetermined circles whose roundness is greater than a preset circle threshold. The reward is obtained by multiplying the number of valid circles with the preset scoring coefficient for a single circle.

6. The method according to claim 5, characterized in that, In response to the freshness comprehensive score being greater than or equal to the determination threshold, the lunar impact crater in the lunar remote sensing image is determined to be a fresh impact crater. If the overall freshness score is less than the determination threshold, the lunar impact crater in the lunar remote sensing image is classified as a degraded impact crater.

7. The method according to claim 6, characterized in that, The classification confidence score is the ratio obtained by dividing the difference between the overall freshness score and the judgment threshold by the judgment threshold; wherein, In response to the failure of the multi-scale verification of the classification confidence result, the lunar remote sensing image is scaled by a preset ratio to obtain a reduced image and an enlarged image. The freshness comprehensive score corresponding to the reduced image and the enlarged image is obtained respectively, and used as the reduced comprehensive score and the enlarged comprehensive score. The reduced comprehensive score, the enlarged comprehensive score, and the freshness comprehensive score corresponding to the lunar remote sensing image are weighted and summed according to the preset comprehensive scoring weights to obtain a confidence comprehensive score. The confidence label of the category corresponding to the confidence comprehensive score is used as the target crater category of the lunar remote sensing image.

8. The method according to claim 1, characterized in that, The multidimensional feature data is normalized using a pre-trained freshness comprehensive scoring model to obtain normalized multidimensional features. These normalized multidimensional features are then weighted and summed according to dimensional weights adaptively generated based on the image signal-to-noise ratio to obtain a comprehensive freshness score. The pre-training of the freshness comprehensive scoring model includes: The pre-built freshness comprehensive scoring architecture is repeatedly trained using pre-constructed samples. The classification parameters and weight parameters of the freshness comprehensive scoring architecture are iteratively optimized using the controlled variable method until the improvement of the objective function of the freshness comprehensive scoring architecture is less than a pre-set training threshold. Then, the freshness comprehensive scoring architecture learned in the last training is used as the freshness comprehensive scoring model in the application stage, and the classification parameters and weight parameters of the freshness comprehensive scoring architecture learned in the last training are used as the classification parameters and weight parameters of the freshness comprehensive scoring model. Wherein, when optimizing the classification parameters, the weight parameters are fixed, and when optimizing the weight parameters, the classification parameters are fixed.

9. An automatic classification system for lunar impact craters, characterized in that, include: The preprocessing module is used to preprocess the pre-acquired lunar remote sensing images to obtain a single-channel grayscale image. An edge extraction module is used to calculate the image signal-to-noise ratio (SNR) of the remote sensing single-channel grayscale image, adaptively set dual threshold parameters of an edge detection algorithm corresponding to the image SNR to obtain an adaptive edge detection algorithm, and perform adaptive edge detection on the remote sensing single-channel grayscale image based on the adaptive edge detection algorithm to obtain a first edge image, a second edge image, and a third edge image of three scales of interest respectively; wherein, in response to the image SNR being greater than a preset second SNR threshold, the dual threshold parameters are set to a first dual threshold parameter having a first low threshold and a first high threshold with the longest numerical range; in response to the image SNR being less than the second SNR threshold but greater than the preset first SNR threshold, the dual threshold parameters are set to a second dual threshold parameter having a second low threshold and a second high threshold with a medium numerical range; in response to the image SNR being less than the first SNR threshold, the dual threshold parameters are set to a third dual threshold parameter having a third low threshold and a third high threshold with the shortest numerical range; The fusion calculation module is used to perform gradient detection on the remote sensing single-channel grayscale image based on a preset first-order differential operator to obtain a gradient edge image. It then adaptively generates fusion weights for the first edge image, the second edge image, the third edge image, and the gradient edge image based on the image signal-to-noise ratio. Finally, it performs a weighted summation of the first edge image, the second edge image, the third edge image, and the gradient edge image based on the fusion weights to obtain a fused edge image. The first edge image is a detail-scale image used to capture the crater's edge details and radial patterns; the second edge image is a medium-scale image used to capture the crater's edge structure; and the third edge image is a coarse-scale image used to preserve the main edge and crater edge contour of the crater. The feature extraction module is used to extract features from the fused edge image and the remote sensing single-channel grayscale image respectively to obtain multi-dimensional feature data. The multi-dimensional feature data includes basic features and special features. The basic features include edge pixel ratio, fused edge density, maximum circularity, maximum circularity ratio, average gradient, and local contrast. The special features include pit edge sharpness attenuation rate, radiation texture anisotropy index, and pit bottom-pit wall gradient discontinuity index. The scoring calculation module is used to normalize the multidimensional feature data to obtain normalized multidimensional features, and to perform weighted summation of the normalized multidimensional features according to the dimension weights adaptively generated based on the image signal-to-noise ratio to obtain a comprehensive freshness score. The category output module is used to determine the category of lunar impact craters in the lunar remote sensing image based on the freshness comprehensive score and a preset judgment threshold, and output the classification label corresponding to the category. The classification label is then subjected to multi-scale verification based on a preset multi-scale verification algorithm to obtain a classification confidence result. In response to the classification confidence result indicating that the multi-scale verification has passed, the classification label is used as the target crater category of the lunar remote sensing image.

Images