Asteroid multi-stage surface feature recognition and surface feature analysis method

Abstract

The invention provides an asteroid multi-stage surface feature recognition and surface feature analysis method, which comprises the following steps of: predicting a detection frame of an asteroid, a meteorite crater and an artificial marker for a long-distance image of the asteroid through natural language prompt; aiming at a prediction result, adopting a hierarchical filtering optimization method to carry out multi-target accurate detection; aiming at a filtering result, fusing a segmentation method of detection box prompt and a segmentation method of automatic mask generation, and drawing a remote sensing semantic map to identify a region of interest; for the recognized region of interest, based on a sampling-first classification strategy, quickly marking a sand image in a detailed examination section image of the region; and aiming at the marked sand image, identifying a sampling weathered layer through granularity analysis, and supporting the construction of a global sampling map. According to the invention, a complete technical chain from long-distance perception to detailed examination section fine analysis is covered, and the ground feature intelligent identification and analysis capability in an asteroid exploration task is improved.

Description

Technical Field

[0001] The present invention belongs to the field of asteroid surface feature identification, and in particular relates to an asteroid multi-stage feature identification and surface feature analysis method. Background Art

[0002] Asteroid surface feature recognition is a remote sensing and computer vision technology that aims to automatically identify and analyze the features of asteroid surfaces by processing and analyzing asteroid surface images. This technology relies on high-resolution image data and can provide scientists with accurate classification information of features such as craters, rocks, sand and other surface features during asteroid exploration missions, thereby providing basic data support for subsequent exploration, sampling and landing missions.

[0003] In previous exploration missions, traditional methods of object recognition mainly rely on manual or semi-automatic techniques, which are inefficient and difficult to meet the real-time and high-accuracy requirements in deep space environments. With the advancement of deep learning technology, researchers have achieved accurate detection and segmentation of boulders on the surface of asteroids through convolutional neural networks combined with a large amount of labeled data. However, these methods are highly dependent on labeled data, especially in cross-celestial missions, and the generalization ability of the model is often limited due to differences in terrain features and image resolution.

[0004] Large-scale pre-trained models (Grounding DINO and SAM, etc.) have powerful detection and segmentation capabilities. They use the prior knowledge of pre-trained models to reduce dependence on labeled data and significantly improve the generalization ability of the model. However, in the special environment of asteroid exploration, there are still some challenges, such as confusion between different types of objects, especially in craters and rocky areas, where blurred semantic boundaries make detection difficult. In addition, rocky areas are densely distributed and numerous, and the detection boxes generated by text prompts often cannot fully cover them, resulting in insufficient recall. Therefore, the improvement and optimization of asteroid surface object recognition is still of great significance.

[0005] The deficiencies of existing technologies for intelligent identification and analysis of asteroid surface objects are mainly reflected in the following aspects: (1) The number of targets to be identified is large, the categories are rich, and the distance and near-distance scales vary significantly. A single detection or segmentation method is difficult to meet all requirements. (2) Without relying on large-scale annotated data, confusion between different types of objects is easy to occur, and the recall of densely distributed rocks is insufficient, resulting in reduced identification accuracy. (3) The identification of sampleable weathering layers still relies on manual labor, and the intelligent detection process needs to be further improved. Summary of the invention

[0006] In order to overcome the shortcomings of the existing technology, the present invention provides a multi-stage asteroid object identification and surface feature analysis method, designs a global perception technology for long-distance object identification and a surface feature analysis technology for detailed investigation, realizes the gradual analysis of asteroid surface object features, and ultimately supports the construction of a global collectability map. Through the organic combination of long-distance global perception and close-range fine analysis, the framework can identify scientific targets and assist the probe in actively perceiving and selecting sampleable areas.

[0007] In order to achieve the above object, the present invention adopts the following technical scheme:

[0008] A method for multi-stage asteroid feature identification and surface feature analysis, comprising the following steps:

[0009] Step 1: Predict the detection frames of asteroids, craters, and artificial landmarks for long-distance asteroid images through natural language prompts;

[0010] Step 2: Based on the prediction results, a hierarchical filtering optimization method is used to perform multi-target precise detection;

[0011] Step 3: Fuse the segmentation method of the detection box prompt and the segmentation method of automatic mask generation to draw a long-range perception semantic map to identify the region of interest;

[0012] Step 4: for the identified region of interest, based on a sampling-first classification strategy, quickly mark the sand image in the detailed inspection segment image of the region of interest;

[0013] Step 5: Based on the marked sand image, identify the sampleable weathered layer through particle size analysis and construct a global mineability map.

[0014] The beneficial effects of the present invention compared with the prior art are:

[0015] (1) This paper proposes a set of feasibility-driven asteroid multi-stage feature identification and surface feature analysis technology frameworks, covering a complete technology chain from long-distance perception to close-range fine analysis. The framework focuses on improving detection efficiency and sampling success rate, and through the coordinated application of key technologies, it realizes intelligent identification and analysis of asteroid surface features.

[0016] (2) The present invention optimizes the application of large-scale training models in deep space environments, solves the problems of multi-target overlap and semantic ambiguity based on a hierarchical filtering optimization method, significantly improves the adaptability of the model in deep space scenes, and achieves accurate detection of scientific targets.

[0017] (3) The present invention takes into account the computational and communication costs of the detector and proposes a sampling-priority classification strategy to complete rapid screening of the region locally on the detector. The screening results are used as the input of the fine segmentation task, concentrating computing resources on high-value target areas.

[0018] In summary, the method adopted by the present invention is simple in principle and can achieve the purpose of multi-stage asteroid object identification and surface feature analysis under the input of long-distance and detailed survey segment images. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flow chart of a method for multi-stage object identification and surface feature analysis of an asteroid. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of the present invention.

[0021] like Figure 1 As shown, a method for multi-stage asteroid object recognition and surface feature analysis of the present invention comprises the following steps:

[0022] Step 1: Predict the detection frames of asteroids, craters and artificial landmarks for asteroid long-distance images through natural language prompts, including:

[0023] The input of the long-range perception phase is the long-range image of asteroids. The Grounding DINO model based on the dual encoder-single decoder architecture is used to detect designated targets such as asteroids, craters, and artificial landmarks based on natural language cues.

[0024] Firstly, the asteroid long-distance image is input into the model, and the image and text features are extracted using the image backbone and text backbone, respectively;

[0025] Subsequently, the extracted image and text features are cross-modally fused through the feature enhancement module to achieve alignment of image and text features and obtain cross-modal features. The cross-modal features extract query information from image features by the language-guided query selection module so that the decoder can further perform target detection and frame refinement. Through multimodal learning technology, image and text features are mapped to a unified feature space. For long-distance images of asteroids, "Whiteball, Craterson asteroid." is selected as the text prompt, and a series of detection frames of asteroids, craters and artificial markers are predicted by the model.

[0026] Step 2: Based on the prediction results, a hierarchical filtering optimization method is used to perform multi-target precise detection, including:

[0027] In long-distance images, asteroids, craters, and artificial landmarks have significant uncertainties in their semantic features and spatial distribution. To solve the problems of multi-target overlap and semantic ambiguity, a triple filtering strategy of semantic confidence, area restriction, and spatial overlap is proposed:

[0028] First, select the detection boxes with the highest confidence for asteroids and artificial landmarks as the recognition results:

[0029] ,

[0030] ,

[0031] in, and is the index number of the asteroid detection frame and the artificial landmark detection frame, and They represent the index numbers of the successfully detected asteroid detection frame and artificial landmark detection frame, respectively. and Represent the confidence of the asteroid and artificial landmark target detection frames, respectively. and Represents the set of asteroid and artificial landmark detection frame index numbers respectively;

[0032] Reduce target category conflicts through area restrictions and confidence multi-level filtering, and calculate the area of ​​the asteroid based on the detection frame As a reference area, select a target set with a crater area less than one-third of the asteroid frame area. :

[0033] ,

[0034] in, Indicates the index number of the crater detection box, represents the confidence of the crater target detection box, Represents the maximum confidence value of the crater target detection box, It means two conditions are met at the same time;

[0035] Finally, label overlap detection is performed to determine the spatial relationship between craters and artificial landmarks, and to remove overlapping crater targets:

[0036] ,

[0037] in, and Represents the boundary coordinates of the corresponding detection box, is the index number of the crater detection box that passes the area filter. If If the value is true, then the target Will be eliminated.

[0038] Step 3: Fuse the segmentation method of the detection box prompt and the segmentation method of automatic mask generation to draw a long-range perception semantic map to identify the region of interest, including;

[0039] The segmentation method of the detection box prompt uses the detection box generated in step 2 as a prompt condition to input the SAM model, and extracts global features of the image through a visual transformer. The extracted features are mapped to a high-dimensional embedding space to form a global feature representation of the whole image; the prompt box is embedded in the same high-dimensional embedding space, and the target area in the box is located in the global feature through the attention mechanism:

[0040] ,

[0041] in, represents the segmentation mask of the target, represents the detection box prompt of the target, img represents the long-distance image of the asteroid, t corresponds to the target index of the asteroid, crater, and artificial landmark, and the mask set of the semantic target is expressed as:

[0042] ,

[0043] The automatic mask generation segmentation method uses the automatic_mask_generator function of SAM, without the need for detection box prompts, and directly generates all possible ground object segmentation masks by globally analyzing the image content. :

[0044] ,

[0045] The automatically generated mask contains all possible target areas, which are screened in combination with the mask of the key target. , calculate its mask set with the key target For the area with high mask overlap that does not belong to an independent rock region, its index k is included in the set to be eliminated. , the mask after removal is the segmentation result of the rock area, expressed as:

[0046] ,

[0047] After completing the segmentation of various types of land features, a multi-level semantic result overlay method is applied to construct a long-distance perception semantic map and identify areas of interest.

[0048] Step 4: for the identified region of interest, based on a sampling-first classification strategy, quickly marking the sand image in the detailed inspection segment image of the region of interest, including:

[0049] After the detailed inspection segment images of the area of ​​interest are standardized, they are passed into the ResNet network to perform image-level classification tasks and output category scores, which include hard surfaces, rock accumulation areas, and sandy areas. They are converted into probability distributions through the Softmax function, and the category with the largest probability is selected as the final prediction result.

[0050] Step 5: For the marked sand image, identify the sampleable weathered layer through particle size analysis and construct a global mineability map, including:

[0051] SAM is used to perform global segmentation on the marked sand image to generate the initial masks of all objects, which are then merged to form an overall mask, denoted as ;

[0052] Using image spatial resolution , unit: mm / pixel, calculate the pixel scale of the particles, the pixel size corresponding to 5cm is approximately:

[0053] ,

[0054] Morphological corrosion is used to remove particles smaller than 5 cm in size. The kernel size of the corrosion operation is , forming the eroded mask :

[0055] ,

[0056] in, Represents a morphological erosion operation;

[0057] The eroded mask is processed by dilation operation, and the size of the dilation kernel is set according to the camera resolution and particle spacing. , the expanded mask is expressed as :

[0058] ,

[0059] in, represents the morphological dilation operation;

[0060] The morphological erosion operation is performed again, and then the mask is inverted to obtain the distribution area of ​​the sampleable weathering layer. The global collectability map is constructed by combining the orbital information with the long-distance perception semantic map for registration.

[0061] It can be seen that the present invention can effectively improve the accuracy of object recognition and surface feature analysis at long distances and in the detailed investigation stage. In the long-distance perception stage, the hierarchical filtering optimization method solves the problems of semantic ambiguity and multi-label overlap, and realizes efficient screening and contour extraction of objects; in the detailed investigation stage, the sampling area priority classification strategy is combined with the sampleable weathering layer area analysis method, which not only optimizes the resource allocation of the detector, but also provides high-confidence regional support for the sampling task, realizing high-precision multi-stage object recognition and surface feature analysis of asteroids.

[0062] Contents not described in detail in the specification of the present invention belong to the common knowledge of professionals in the field. Although the above illustrative specific embodiments of the present invention are described to facilitate the understanding of the present invention by those skilled in the art, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A multi-stage asteroid feature identification and surface feature analysis method, characterized in that: The following steps are involved: Step 1: Predict the detection frames of asteroids, craters, and artificial landmarks for long-distance asteroid images through natural language prompts; Step 2: Based on the prediction results, a hierarchical filtering optimization method is used to perform multi-target precise detection; Step 3: Fuse the segmentation method of the detection box prompt and the segmentation method of automatic mask generation to draw a long-range perception semantic map to identify the region of interest; Step 4: for the identified region of interest, based on a sampling-first classification strategy, quickly mark the sand image in the detailed inspection segment image of the region of interest; Step 5: For the marked sand image, identify the sampleable weathered layer through particle size analysis and construct a global mineability map.

2. The method for multi-stage asteroid feature identification and surface feature analysis according to claim 1, characterized in that: The step 1 comprises: The asteroid long-distance images are input into the Grounding DINO model based on a dual encoder-single decoder architecture, which uses the image backbone and text backbone to extract image and text features respectively; The extracted image and text features are cross-modally fused through the feature enhancement module to achieve alignment of image and text features and obtain cross-modal features; The cross-modal feature extracts query information from image features by a language-guided query selection module, and obtains a series of detection frames of asteroids, craters and artificial landmarks through a decoder.

3. The method for multi-stage asteroid feature identification and surface feature analysis according to claim 1, characterized in that: The step 2 includes a triple filtering strategy of semantic confidence, area restriction and spatial overlap: First, select the detection boxes with the highest confidence for asteroids and artificial landmarks as the recognition results: ， ， in, and is the index number of the asteroid detection frame and the artificial landmark detection frame, and They represent the index numbers of the successfully detected asteroid detection frame and artificial landmark detection frame, respectively. and Represent the confidence of the asteroid and artificial landmark target detection frames, respectively. and Represents the set of asteroid and artificial landmark detection frame index numbers respectively; Calculate the area of ​​the asteroid based on the detection frame As a reference area, select a target set with a crater area less than one-third of the asteroid frame area. : ， in, Indicates the index number of the crater detection box, represents the confidence of the crater target detection box, Represents the maximum confidence value of the crater target detection box, It means two conditions are met at the same time; Finally, label overlap detection is performed to determine the spatial relationship between craters and artificial landmarks, and to remove overlapping crater targets: ， in, and Represents the boundary coordinates of the corresponding detection box, is the index number of the crater detection box that passes the area filter. If If the value is true, then the target Will be eliminated.

4. The method for multi-stage asteroid feature identification and surface feature analysis according to claim 1, characterized in that: The step 3 comprises: The segmentation method of the detection box prompt uses the detection box generated in step 2 as a prompt condition to input the SAM model, and extracts global features of the image through a visual transformer. The extracted features are mapped to a high-dimensional embedding space to form a global feature representation of the whole image; the prompt box is embedded in the same high-dimensional embedding space, and the target area in the box is located in the global feature through the attention mechanism: ， in, represents the segmentation mask of the target, represents the detection box prompt of the target, img represents the long-distance image of the asteroid, t corresponds to the target index of the asteroid, crater, and artificial landmark, and the mask set of the semantic target is expressed as: ， The automatic mask generation segmentation method uses the automatic_mask_generator function of SAM, without the need for detection box prompts, and directly generates all possible ground object segmentation masks by globally analyzing the image content. : ， For each automatically generated mask , calculate its mask set with the key target For the area with high mask overlap that does not belong to an independent rock region, its index k is included in the set to be eliminated. , the mask after removal is the segmentation result of the rock area, expressed as: ， After completing the segmentation of various types of land features, a multi-level semantic result overlay method is applied to construct a long-distance perception semantic map and identify areas of interest.

5. The method for multi-stage asteroid feature identification and surface feature analysis according to claim 1, characterized in that: The step 4 comprises: After the detailed inspection segment images of the area of ​​interest are standardized, they are passed into the ResNet network to perform image-level classification tasks and output category scores, which include hard surfaces, rock accumulation areas, and sandy areas. They are converted into probability distributions through the Softmax function, and the category with the largest probability is selected as the final prediction result.

6. The method for multi-stage asteroid feature identification and surface feature analysis according to claim 1, characterized in that: The step 5 comprises: SAM is used to perform global segmentation on the marked sand image to generate the initial masks of all objects, which are then merged to form an overall mask, denoted as ; Using image spatial resolution , unit: mm / pixel, calculate the pixel scale of the particles, the pixel size corresponding to 5cm is: ， Morphological corrosion is used to remove particles smaller than 5 cm in size. The kernel size of the corrosion operation is , forming the eroded mask : ， in, Represents a morphological erosion operation; The eroded mask is processed by dilation operation, and the size of the dilation kernel is set according to the camera resolution and particle spacing. , the expanded mask is expressed as : ， in, represents the morphological dilation operation; The morphological erosion operation is performed again, and then the mask is inverted to obtain the distribution area of ​​the sampleable weathering layer. The global collectability map is constructed by combining the orbital information with the long-distance perception semantic map for registration.

Images