Multi-scale star catalogue impact crater detection method and system and medium
By constructing a multi-scale impact crater detection model and sliding window slice processing technology, the problems of multi-scale recognition and sample imbalance in planet surface impact craters are solved, the recognition accuracy and efficiency are improved, and more accurate data is provided for lunar exploration and geological research. support.
Patent Information
- Application Number
- CN202411142614.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-08-20
AI Technical Summary
The prior art has poor multi-scale identification of impact craters on planet surfaces, and uneven sample samples lead to insufficient recognition accuracy and low efficiency, especially when identifying small impact craters.
Using multi-scale detection algorithm, data augmentation and image processing technology, the impact craters of different diameter ranges are trained by constructing multiple YOLOv5-based impact crater detection sub-models, and combining sliding windows and slice amplification processing to improve the feature extraction of small impact craters. accuracy and recognition efficiency.
It improves the accuracy and efficiency of identification of impact craters of different sizes on the planet's surface, overcomes the problems of insufficient recognition accuracy, unbalanced sample and poor multi-scale recognition effects, and provides more accurate and comprehensive data support for lunar scientific research.
Smart Images

Figure CN118968339A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of remote sensing image processing, and in particular relates to a multi-scale star catalog impact crater detection method, system and medium. Background Art
[0002] In the field of planetary exploration and geological research, the identification and extraction of impact craters on the planetary surface has always been an important research direction. Impact craters are not only one of the most significant geomorphic features on the planetary surface, but also carry rich geological information, which is of great significance for understanding the planetary evolution history, geological structure and surface processes.
[0003] Traditional crater identification methods mainly rely on manual identification and visual interpretation, that is, researchers manually mark and record the location, size and other relevant parameters of craters by directly observing and analyzing images or photos of the planet's surface. Although this method can meet research needs to a certain extent, it requires a lot of manpower and time, and the identification results are often subjective and inconsistent, making it difficult to ensure the quality and accuracy of the data.
[0004] To overcome this limitation, existing technologies usually use computer vision and image processing technologies, such as image analysis, data fusion, machine learning, and training deep learning network models, to process and analyze images of planetary surfaces, thereby realizing automatic recognition and extraction of impact craters. Although these methods have achieved certain results and improved the efficiency and accuracy of impact crater recognition to a certain extent, there are still the following problems:
[0005] 1. Multi-scale recognition is not effective: The impact craters on the surface of the planet vary in size, with diameters ranging from a few meters to hundreds of kilometers. The traditional single-scale recognition method is difficult to adapt to impact craters of different sizes at the same time, especially when identifying small impact craters.
[0006] 2. Sample imbalance problem: There are relatively few large impact craters on the surface of the planet, but a large number of small impact craters. This sample imbalance causes the machine learning model to tend to identify large, distinctive impact craters during training, while ignoring or misjudging small impact craters.
[0007] 3. Insufficient recognition accuracy: Traditional recognition methods, especially those based on simple image processing technology, often have difficulty in achieving high-precision recognition when dealing with small impact craters with unclear features or irregular shapes. This may result in a large number of small impact craters being missed or misidentified.
[0008] 4. Inefficiency: Methods that rely on manual recognition or simple automation are inefficient when processing large-scale planetary surface data and cannot meet the requirements of modern planetary exploration missions for data processing speed and accuracy.
[0009] Therefore, there is an urgent need to develop a detection method that can simultaneously identify impact craters of multiple scales on the surface of a planet, in order to solve the problem that existing detection methods cannot simultaneously identify impact craters of different sizes on the surface of a planet, especially the lack of accuracy and robustness when identifying small impact craters. Summary of the invention
[0010] The purpose of the present invention is to solve the above-mentioned technical problems and provide a multi-scale star catalog impact crater detection method, system and medium. By adopting a multi-scale detection algorithm, data enhancement and image processing technology, the problems of insufficient recognition accuracy, sample imbalance and poor multi-scale recognition effect faced by the prior art in identifying impact craters are overcome, the accuracy and recognition efficiency of feature extraction of small impact craters on the surface of the planet are improved, more accurate and comprehensive data support is provided for lunar scientific research, and the in-depth development of lunar exploration and geological research is promoted.
[0011] To achieve the above object, the technical solution adopted by the present invention is:
[0012] A multi-scale star catalog impact crater detection method comprises the following steps:
[0013] S1: obtaining star catalog image data containing impact craters, segmenting the obtained star catalog image at different scales, and obtaining star catalog image blocks of different scales; star catalog image blocks of different scales respectively contain impact crater targets of different diameter ranges;
[0014] S2: Build a multi-scale crater detection model, which includes multiple crater detection sub-models based on YOLOv5;
[0015] S3: Multiple crater detection sub-models are trained using star catalog image blocks of different scales to obtain multiple sub-models for detecting and identifying crater targets of different diameter ranges;
[0016] S4: obtaining a star catalog image to be detected, dividing it into image blocks of different scales, and inputting them into sub-models of corresponding scale ranges respectively, to obtain all candidate bounding boxes of impact crater targets in the star catalog image to be detected;
[0017] S5: Sort all candidate bounding boxes of each crater target based on their areas, calculate the IoU values of the candidate bounding box with the largest area and other candidate bounding boxes, and select the candidate bounding box with the largest area from all candidate bounding boxes whose IoU values exceed a predetermined IoU threshold as the final detection result of the crater target.
[0018] In some embodiments of the present invention, S5 specifically includes the following steps:
[0019] S51: Define a predetermined IoU threshold τ and a candidate bounding box list B for each crater target:
[0020] B={B 1 ,B 2 ,…,B N};
[0021] Among them, each candidate bounding box contains location information;
[0022] S52: sorting the candidate bounding boxes in list B based on the areas of the candidate bounding boxes;
[0023] S53: Select the candidate bounding box B with the largest current area in list B i , i∈[1,N];
[0024] Calculate the other candidate bounding boxes in list B and B i The IoU value of all i The candidate bounding boxes whose IoU values exceed the predetermined IoU threshold τ are added to a predetermined empty set G;
[0025] After the calculation is completed, if the set G is not empty, in B i The candidate bounding box B{max} with the largest area selected from the set G is added to a predetermined result list D;
[0026] S54: Remove the candidate bounding box with the largest current area and all candidate bounding boxes in set G from list B;
[0027] S55: repeat steps S53-S54 until all candidate bounding boxes in list B are removed;
[0028] S56: Output the candidate bounding boxes in list D as the final detection result of the crater target.
[0029] In some embodiments of the present invention, in step S1, the method of segmenting each star catalog image into different scales specifically includes the following steps:
[0030] Set three sliding windows of different sizes;
[0031] Three sliding windows of different sizes start from the upper left corner of each star catalog image and slide from left to right and from top to bottom. During the sliding process, star catalog image blocks of corresponding window sizes are cropped from the image.
[0032] In some embodiments of the present invention, the three sliding windows of different sizes are sliding windows of 320×320 pixels, 640×640 pixels, and 1280×1280 pixels.
[0033] In some embodiments of the present invention, the following steps are also included:
[0034] Setting a first diameter threshold and a second diameter threshold; the second diameter threshold is greater than the first diameter threshold;
[0035] The image blocks cropped by the sliding windows of three different sizes respectively include crater targets with diameters below the first diameter threshold, crater targets between the first diameter threshold and the second diameter threshold, and crater targets above the second diameter threshold.
[0036] In some embodiments of the present invention, the first diameter threshold is 50 pixels, and the second diameter threshold is 100 pixels.
[0037] In some embodiments of the present invention, the following steps are also included:
[0038] Setting a predetermined scale threshold;
[0039] In step S1, after obtaining star catalog image blocks of different scales, for image blocks with side lengths below a predetermined scale threshold, they are divided into multiple sub-regions of the same size, each sub-region is enlarged respectively, and the enlarged image blocks are segmented again based on the division of each sub-region.
[0040] In some embodiments of the present invention, the value range of the predetermined scale threshold is 320 pixels-640 pixels.
[0041] Some embodiments of the present invention further provide a multi-scale star catalog impact crater detection system, characterized in that it includes:
[0042] at least one processor;
[0043] at least one memory for storing at least one program;
[0044] When at least one program is executed by at least one processor, the at least one processor implements a multi-scale star catalog impact crater detection method.
[0045] Some embodiments of the present invention further provide a storage medium storing a program executable by a processor, wherein the program executable by the processor is used to implement a multi-scale star catalog impact crater detection method when executed by the processor.
[0046] The beneficial effects of the present invention are:
[0047] 1. The present invention constructs multiple crater detection sub-models and uses image blocks of different scales for training to obtain multiple sub-models for detecting and identifying crater targets with different diameter ranges, so that the trained multi-scale crater detection model can simultaneously detect craters with different diameter ranges, thereby improving the accuracy of crater recognition of various sizes of the model;
[0048] 2. The present invention adopts slicing, magnification, and re-slicing processing for images containing small-scale impact craters, thereby improving the clarity of small-scale impact crater features in the image, thereby improving the model's ability to extract features for small-scale impact craters;
[0049] 3. The present invention evaluates the overlap and confidence between different candidate bounding boxes, and preferentially retains the prediction results with high confidence and large area, so as to obtain more accurate and complete crater recognition results. While maintaining the detection accuracy, it reduces redundant predictions and improves the recognition efficiency.
[0050] 4. The present invention overcomes the problems of insufficient recognition accuracy, sample imbalance and poor multi-scale recognition effect faced by the prior art in identifying craters of different scales in the star catalog by constructing a multi-scale crater detection model. In particular, it improves the accuracy and efficiency of feature extraction and recognition of small craters on the planet's surface, provides more accurate and comprehensive data support for lunar scientific research, and promotes the in-depth development of lunar exploration and geological research. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0052] Figure 1 It is a flowchart of a multi-scale star catalog impact crater detection method;
[0053] Figure 2 Schematic diagram of the star catalog image data processing process in this embodiment;
[0054] Figure 3 Schematic diagram of the sliding window cropping process;
[0055] Figure 4 This is a schematic diagram of the process of enlarging and segmenting a small-scale image;
[0056] Figure 5 This is a schematic diagram of redundant crater detection results of the multi-scale crater detection model;
[0057] Among them, (a), (b), (c), and (d) are the redundant detection results of the same impact crater in different slices;
[0058] Figure 6 The present invention is a flow chart of a redundant impact crater processing method in the prior art;
[0059] Figure 7 A flow chart of the redundant impact crater processing method provided in this embodiment;
[0060] Figure 8 A statistical graph showing the results of the detection method provided in this embodiment identifying 500,000 impact craters in the Chang'e 4 landing area;
[0061] Fig. 9 A schematic diagram of the detection results of identifying multi-scale impact craters using the detection method provided in this embodiment. DETAILED DESCRIPTION
[0062] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.
[0063] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0064] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.
[0065] The technical solution of the present invention is described in detail below in conjunction with specific embodiments and the accompanying drawings.
[0066] As attached Figure 1 -Attached Fig. 9 As shown, in an illustrative embodiment of a multi-scale star catalog impact crater detection method, system and medium of the present invention, the method is used to detect and identify impact craters of different scales on the moon, and the method includes the following steps.
[0067] S1: Data acquisition and preprocessing.
[0068] First, we collected high-definition image data of the lunar surface collected by lunar probes such as Chang'e 4 and organized them into a data set. Based on the image data in the data set, we conducted a detailed study and scale division of the impact craters on the lunar surface, paying special attention to the characteristics and distribution patterns of small impact craters in the images.
[0069] Obtain multiple high-resolution star images containing impact craters from the data set, and use the sliding window method to segment the acquired star images at different scales to obtain star image blocks of different scales, such as Figure 3 As shown in the figure, the star catalog image blocks of different scales contain impact crater targets of different diameter ranges.
[0070] After obtaining star catalog image blocks of different scales, they are made into training sets and test sets without spatial intersection, and both the training set and the test set contain star catalog image blocks of various scales.
[0071] In some embodiments of the present invention, in step S1, the method of segmenting each star catalog image into different scales specifically includes the following steps.
[0072] Three sliding windows of different sizes are set, namely a small-scale window, a medium-scale window and a large-scale window. Specifically, in this embodiment, the sizes of the three sliding windows of different sizes are 320×320 pixels, 640×640 pixels and 1280×1280 pixels, respectively.
[0073] Three sliding windows of different sizes start from the upper left corner of each star catalog image and slide from left to right and from top to bottom. During the sliding process, star catalog image blocks of corresponding window sizes are cropped from the image, namely small-scale image blocks, medium-scale image blocks and large-scale image blocks. The sizes of the three-scale image blocks are 320×320 pixels, 640×640 pixels and 1280×1280 pixels, respectively.
[0074] Some embodiments of the present invention further include the following steps.
[0075] A first diameter threshold and a second diameter threshold are set, and the second diameter threshold is greater than the first diameter threshold. In this embodiment, the first diameter threshold is 50 pixels, and the second diameter threshold is 100 pixels.
[0076] The image blocks cropped by the sliding windows of three different sizes respectively include crater targets with diameters below the first diameter threshold, crater targets between the first diameter threshold and the second diameter threshold, and crater targets above the second diameter threshold.
[0077] Specifically, an image block of 320×320 pixels contains at least crater targets with a diameter of less than 50 pixels, an image block of 640×640 pixels contains at least crater targets with a diameter between 50 pixels and 100 pixels, and an image block of 1280×1280 pixels contains at least crater targets with a diameter of more than 100 pixels. It should be noted that, in addition to crater targets of corresponding diameters, image blocks of each scale may also contain crater targets of other diameters.
[0078] In terms of image feature extraction, in order to solve the problem of the lack of sufficient feature information when the prior art recognizes small-scale targets, in some embodiments of the present invention, for smaller-scale image blocks, a "Slice-Enlarge-Slice" (SES) strategy is adopted, and the strategy further includes the following steps.
[0079] A predetermined scale threshold is set. In this embodiment, the predetermined scale threshold ranges from 320 pixels to 640 pixels.
[0080] In step S1, after obtaining the star catalog image blocks of different scales, for the image blocks whose side length is below the predetermined scale threshold, the image blocks are divided into a plurality of sub-regions of the same size, each sub-region is enlarged respectively, and each sub-region is enlarged to an image block of the same scale as the original image block before the enlargement, and the enlarged image block is divided again based on the division of each sub-region, and divided into a plurality of image blocks of the same scale as the original image block before the enlargement. For example, as shown in the attached Figure 4 As shown, in this embodiment, a "slice-enlarge-slice" strategy is adopted for the image block of 320×320 pixels, and the acquired image block of 320×320 pixels is upsampled twice and enlarged as a whole to an image block of 1280×1280 pixels. The enlarged image block is divided into regions of 320×320 pixels and cropped into 16 image blocks of 320×320 pixels.
[0081] S2: Construct a multi-scale crater detection model, which includes three YOLOv5-based crater detection sub-models with the same structure and independent of each other.
[0082] S3: The three crater detection sub-models are trained using the training set data to obtain three sub-models for detecting and identifying crater targets of different diameter ranges. In this embodiment, they are YOLOv5-L sub-model, YOLOv5-M sub-model and YOLOv5-S sub-model, respectively. The three sub-models are responsible for detecting large-scale (more than 100 pixels), medium-scale (50-100 pixels) and small-scale (less than 50 pixels) crater targets in the detection area. Each sub-model adjusts the weights through training to optimize for craters of a specific scale.
[0083] After the sub-model training is completed, the test set data is used to test the detection and recognition effects of the three sub-models.
[0084] S4: Obtain the star catalog image to be detected, divide it into image blocks of 320×320 pixels, 640×640 pixels, and 1280×1280 pixels, and input them into the YOLOv5-S sub-model, YOLOv5-M sub-model, and YOLOv5-L sub-model respectively to obtain all candidate bounding boxes of the impact crater targets in the star catalog image to be detected, as shown in the attached figure. Figure 4 It should be noted that multiple candidate bounding boxes of the same impact crater target can be obtained by annotating one sub-model or by annotating multiple sub-models separately.
[0085] S5: Sort all candidate bounding boxes of each crater target based on their areas, calculate the IoU values of the candidate bounding box with the largest area and other candidate bounding boxes, and select the candidate bounding box with the largest area from all candidate bounding boxes whose IoU values exceed a predetermined IoU threshold as the final detection result of the candidate bounding box.
[0086] It should be noted that YOLOv5 uses the traditional non-maximum suppression algorithm (NMS) to select the prediction box B with the highest confidence score from all candidate bounding boxes. 1 as the baseline, and then remove the 1 All other bounding boxes whose intersection over union (IoU) exceeds a predefined threshold. Figure 6 As shown in the figure, the best detected target in the target area is not necessarily the target with the highest confidence score. Therefore, in some specific scenarios, the traditional NMS algorithm cannot select the optimal candidate bounding box.
[0087] In the present invention, a region-prioritized non-maximum suppression algorithm (NMS_AF) is provided, the core idea of which is that when processing IoU overlap, not only the confidence but also the area of the candidate bounding box is considered. When the IoU values of multiple candidate bounding boxes and a candidate bounding box with the highest confidence exceed a predetermined IoU threshold, the candidate bounding boxes exceeding the threshold are grouped into a group, and the candidate bounding box with the largest area is selected from the group as the representative of the group, that is, as the final detection result of the crater target. In practical applications, the suppression threshold can be dynamically adjusted according to the score, overlap and positional relationship of the candidate bounding box to ensure that the algorithm selects the optimal candidate bounding box as the final detection result, reducing false detection and missed detection.
[0088] In some embodiments of the present invention, S5 specifically includes the following steps.
[0089] S51: Define a predetermined IoU threshold τ and a candidate bounding box list B for each crater target:
[0090] B={B 1 ,B 2 ,…,BN};
[0091] Among them, each candidate bounding box contains location information.
[0092] S52: Sort the candidate bounding boxes in list B based on their areas.
[0093] S53: Select the candidate bounding box B with the largest current area in list B i , i∈[1,N].
[0094] Calculate the other candidate bounding boxes in list B and B i The IoU value of all i The candidate bounding boxes whose IoU values exceed the predetermined IoU threshold τ are added to a predetermined empty set G.
[0095] After the calculation is completed, if the set G is not empty, in B i The candidate bounding box B{max} with the largest area in the set G is added to the result list D.
[0096] S54: Remove the candidate bounding box with the largest current area and all candidate bounding boxes in set G from list B.
[0097] S55: Repeat steps S53-S54 until all candidate bounding boxes in list B are removed.
[0098] S56: Output the candidate bounding boxes in list D as the final detection result of the crater target.
[0099] The present invention constructs a multi-scale crater detection model and introduces a region-priority non-maximum suppression algorithm (NMS_AF), which not only improves the accuracy and efficiency of identifying crater targets of different scales, but also overcomes the recognition difficulties in the prior art caused by the scarcity of small crater features and uneven sample distribution by adopting a "slice-zoom-slice" strategy.
[0100] The beneficial effects of the present invention are described below in conjunction with a specific embodiment:
[0101] In a specific embodiment of identifying 500,000 impact craters in the Chang'e-4 landing area, the detection results of the multi-scale impact crater detection model and its sub-models are evaluated as shown in the attached figure. Figure 8 As shown, attached Figure 8 The statistical data of the evaluation are shown in Table 1, and the crater recognition effect is shown in Appendix. Fig. 9 Based on the attached Figure 8 , Attachment Fig. 9As can be seen from Table 1, in terms of recognition accuracy, the detection method YOLOv5-Fused provided by the present invention identified about 500,000 craters in the Chang'e 4 landing area, and the identifiable diameter of the craters was as small as 4m. The Precision (precision) of the multi-scale crater detection model was 0.985, the Recall (recall rate) was 0.943, and the F1-Score (comprehensive evaluation index) was 0.964, showing excellent performance.
[0102] Table 1 Evaluation table of detection results of multi-scale crater detection model and its sub-models
[0103]
[0104] It should be noted that in traditional crater detection methods, the values of evaluation indicators such as precision, recall rate and F1 score are usually between 0.85 and 0.90, while the detection method provided by the present invention improves each indicator by 5%-10%, which means that when identifying craters of different scales, especially small craters, the detection method provided by the present invention can more accurately separate crater targets of different scales from complex backgrounds, thereby reducing false alarms and missed alarms.
[0105] Some embodiments of the present invention further provide a multi-scale star catalog impact crater detection system, characterized in that it includes:
[0106] at least one processor;
[0107] at least one memory for storing at least one program;
[0108] When at least one program is executed by at least one processor, the at least one processor implements a multi-scale star catalog impact crater detection method provided by the present invention.
[0109] Some embodiments of the present invention further provide a storage medium storing a program executable by a processor, wherein the program executable by the processor is used to implement a multi-scale star catalog impact crater detection method provided by the present invention when executed by the processor.
[0110] Finally, it should be noted that: the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0111] The above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solution of the present invention, which should be included in the scope of the technical solution for protection of the present invention.
Claims
1. A multi-scale star catalog impact crater detection method, It is characterized in that The following steps are involved: S1: acquiring star catalog image data containing impact craters, segmenting the acquired star catalog image at different scales, and obtaining data sets of star catalog image blocks at different scales; the star catalog image blocks at different scales respectively contain impact crater targets at different diameter ranges; S2: Build a multi-scale crater detection model, which includes multiple crater detection sub-models based on YOLOv5; S3: Multiple crater detection sub-models are trained using data sets of star catalog image blocks of different scales to obtain multiple sub-models for detecting and identifying crater targets of different diameter ranges; S4: obtaining a star catalog image to be detected, dividing it into image blocks of different scales, and inputting them into sub-models of corresponding scale ranges respectively, to obtain all candidate bounding boxes of each impact crater target in the star catalog image to be detected; S5: Sort all candidate bounding boxes of each crater target based on their areas, calculate the IoU values of the candidate bounding box with the largest area and other candidate bounding boxes, and select the candidate bounding box with the largest area from all candidate bounding boxes whose IoU values exceed a predetermined IoU threshold as the final detection result of the crater target.
2. The multi-scale star catalog impact crater detection method according to claim 1, It is characterized in that The S5 specifically includes the following steps: S51: Define a predetermined IoU threshold τ and a candidate bounding box list B for each crater target: B={B 1 ,B 2 ,…,B N }; Among them, each candidate bounding box contains location information; S52: sorting the candidate bounding boxes in list B based on the areas of the candidate bounding boxes; S53: Select the candidate bounding box B with the largest current area in list B i , i∈[1,N]; Calculate the other candidate bounding boxes in list B and B i The IoU value of all i The candidate bounding boxes whose IoU values exceed the predetermined IoU threshold τ are added to a predetermined empty set G; After the calculation is completed, if the set G is not empty, in B i The candidate bounding box B{max} with the largest area selected from the set G is added to the result list D; S54: Remove the candidate bounding box with the largest current area and all candidate bounding boxes in set G from list B; S55: repeat steps S53-S54 until all candidate bounding boxes in list B are removed; S56: Output the candidate bounding boxes in the list D as the final detection result of the crater target.
3. The multi-scale star catalog impact crater detection method according to claim 1, It is characterized in that In step S1, the method of segmenting each star catalog image into different scales specifically includes the following steps: Set three sliding windows of different sizes; Three sliding windows of different sizes start from the upper left corner of each star catalog image and slide from left to right and from top to bottom. During the sliding process, star catalog image blocks of corresponding window sizes are cropped from the image.
4. The multi-scale star catalog impact crater detection method according to claim 3, It is characterized in that The three sliding windows of different sizes are sliding windows of 320×320 pixels, 640×640 pixels, and 1280×1280 pixels respectively.
5. The multi-scale star catalog impact crater detection method according to any one of claims 3 or 4, It is characterized in that The following steps are also included: Setting a first diameter threshold and a second diameter threshold; wherein the second diameter threshold is greater than the first diameter threshold; The image blocks cropped by the sliding windows of three different sizes respectively include crater targets with diameters below the first diameter threshold, crater targets between the first diameter threshold and the second diameter threshold, and crater targets above the second diameter threshold.
6. The multi-scale star catalog impact crater detection method according to claim 5, It is characterized in that The first diameter threshold is 50 pixels, and the second diameter threshold is 100 pixels.
7. The multi-scale star catalog impact crater detection method according to claim 1, It is characterized in that The following steps are also included: Setting a predetermined scale threshold; In step S1, after obtaining star catalog image blocks of different scales, for image blocks whose side lengths are below the predetermined scale threshold, they are divided into multiple sub-regions of the same size, each sub-region is enlarged respectively, and the enlarged image blocks are segmented again based on the division of each sub-region.
8. The multi-scale star catalog impact crater detection method according to claim 7, It is characterized in that The predetermined scale threshold value ranges from 320 pixels to 640 pixels.
9. A multi-scale star catalog impact crater detection system, It is characterized in that include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the multi-scale star catalog impact crater detection method as described in any one of claims 1-8.
10. A storage medium storing a program executable by a processor, Features: The processor-executable program is used to implement the multi-scale star catalog impact crater detection method as described in any one of claims 1-8 when executed by the processor.
Citation Information
Patent Citations
End-to-end impact crater detection and identification method based on fully convolutional neural network structure
CN108734219A
Remote sensing image multi-scale target detection method based on data fusion and feature selection
CN113449666A
Method and device for automatically extracting moon impact crater based on machine learning
CN115272769A
Meteorite crater detection method based on image enhancement and YOLOv5 improvement
CN115661622A
Moon and south pole impact crater detection method based on improved TransUnet network
CN116309348A
