Deep Learning-Based Karst Landform Image Analysis Method and System

Abstract

The present invention provides a method and system for analyzing karst landform images based on deep learning, which relates to the technical field of deep learning. By extracting the basic semantic features of karst landform images and using a pre-learned karst landform image analysis model, the candidate karst landform semantic features corresponding to each semantic space are accurately determined. These candidate karst landform semantic features are loaded into the corresponding sub-models to generate sub-landform description feature parameters, and then a detailed karst landform description knowledge chain is constructed. This method not only significantly improves the accuracy and efficiency of karst landform image recognition, but also effectively clusters the images through the karst landform description knowledge chain, providing a data analysis means for scientific research exploration, geological disaster prediction, environmental protection and other fields of karst landforms.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning. Specifically, it relates to a method and system for analyzing karst landform images based on deep learning. Background Art

[0002] As a unique and complex landform type on the Earth's surface, the formation and evolution process of karst landforms involve various geological processes, which are of great significance for geological scientific research, the development and protection of natural resources, and the prediction and prevention of geological disasters. However, due to the complexity and diversity of karst landforms, traditional image analysis methods are often difficult to accurately and efficiently identify and classify them.

[0003] In the past, the image analysis of karst landforms mainly relied on manual visual interpretation and simple image processing techniques. These methods are not only time-consuming and laborious, but also greatly affected by subjective factors, making it difficult to ensure the objectivity and accuracy of the analysis results. With the continuous development of remote sensing technology and image processing technology, although some image analysis methods based on pixels or simple features have emerged, these methods are still unable to cope when dealing with complex karst landform images and cannot fully explore and utilize the deep semantic information in the images. Summary of the Invention

[0004] In view of the problems mentioned above, in combination with the first aspect of the present invention, embodiments of the present invention provide a method for analyzing karst landform images based on deep learning. The method includes:

[0005] Obtain a karst landform image to be recognized, and extract the basic karst landform semantic features corresponding to the karst landform image;

[0006] Based on the basic karst landform semantic features, determine the candidate karst landform semantic features corresponding to each semantic space of a karst landform image analysis model that has completed parameter learning in advance;

[0007] Load the candidate karst landform semantic features corresponding to each semantic space into the karst landform image analysis sub-models existing in the semantic space corresponding to the karst landform image analysis model, so as to determine a karst landform description knowledge chain based on the sub-landform description feature parameters generated by each karst landform image analysis sub-model;

[0008] Obtain the karst landform description knowledge chain generated by the karst landform image analysis model, and cluster the karst landform image based on the karst landform description knowledge chain.

[0009] In another aspect, an embodiment of the present invention further provides a karst landform image analysis system based on deep learning, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0010] Based on the above aspects, the embodiments of the present application extract the basic semantic features of karst landform images and use a pre-learned karst landform image analysis model to accurately determine the candidate karst landform semantic features corresponding to each semantic space. These candidate karst landform semantic features are loaded into the corresponding sub-models to generate sub-landform description feature parameters, and then a detailed karst landform description knowledge chain is constructed. This method not only significantly improves the accuracy and efficiency of karst landform image recognition, but also effectively clusters the images through the karst landform description knowledge chain, providing a data analysis means for scientific research exploration, geological disaster prediction, environmental protection and other fields of karst landforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 is a schematic execution flow diagram of the karst landform image analysis method based on deep learning provided by an embodiment of the present invention.

[0012] Figure 2 is a schematic hardware architecture diagram of the karst landform image analysis system based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0013] The present invention will be specifically described below in conjunction with the accompanying drawings of the specification. Figure 1 is a schematic flow diagram of the karst landform image analysis method based on deep learning provided by an embodiment of the present invention. The karst landform image analysis method based on deep learning will be introduced in detail below.

[0014] Step S110: Obtain a karst landform image to be subjected to image recognition, and extract the corresponding basic karst landform semantic features of the karst landform image.

[0015] In this embodiment, for the current project on karst landform research in a certain area, karst landform images to be subjected to image recognition can be obtained from the aerial photography image library of the area. These karst landform images cover karst landform areas under different angles and different lighting conditions.

[0016] First, decompose the karst landform image to generate multiple target landform feature blocks. For example, in a large-scale karst landform image, according to the natural structure of the landform, such as obvious topographic features like peaks, valleys, and karst caves, the karst landform image may be decomposed into several smaller regions, and these regions become the target landform feature blocks.

[0017] Then, for each target landform feature block, extract its corresponding block semantic features. For example, for a target landform feature block containing a karst cave entrance, first perform unified processing on its image format. Assume the original image is in a certain special aerial photography format and convert it to a common image format, such as JPEG format. Then perform standardization processing on the image size, adjusting the length and width of the image to a preset standard size, such as 500×500 pixels. Next, perform normalization processing so that the pixel values of the image are within a specific range for better subsequent algorithm processing.

[0018] After that, use a multi-scale image analysis algorithm to construct a multi-scale image representation. For example, determine the scale range to be used from 1 / 2 to 1 / 8 of the original size. For each scale in the scale range, obtain scale feature images at different scales through downsampling (here, the average pooling method is used, that is, calculate the average pixel value of each small region to reduce the number of pixels) and upsampling (using the bilinear interpolation method, calculate new pixel values based on the surrounding pixel values to increase the image size) algorithms. Then combine these scale feature images at different scales to generate the multi-scale image representation of this target landform feature block.

[0019] Next, use a local feature description algorithm to construct a Difference of Gaussian (DoG) pyramid at each scale of the multi-scale image representation to detect local extreme points. For the local extreme points of the rock texture around the detected karst cave entrance, calculate the corresponding orientation histogram. For example, with this local extreme point as the center, take the pixel gray value changes in a certain surrounding range (such as a 10×10 pixel range) to calculate the orientation histogram. Statistically analyze the pixels in the preset surrounding range according to a predefined orientation interval (such as dividing 0 - 360 degrees into 8 intervals) to obtain a histogram representing the orientation distribution, and this histogram can describe the texture features around this local extreme point.

[0020] Combine the orientation histograms of all local extreme points to generate the local feature description of the target landform feature block. For example, the texture features of the rock around the karst cave entrance are reflected through this local feature description.

[0021] Adopt a sorting algorithm based on feature importance to sort according to the importance of local features in describing the target geomorphic feature block. For example, for an image of a karst cave entrance, those local features reflecting the texture of the rock edge may be more important than some local features in other smooth areas. After removing the local features with importance lower than the set importance (a threshold determined based on previous experience or experiments), similar local features among the remaining local features are fused through a feature fusion algorithm to generate an optimized local feature description.

[0022] Then, adopt a shape-based global feature extraction method to extract shape features from the multi-scale image representation. For the target geomorphic feature block of the karst cave entrance, its edge curve is obtained through an edge detection algorithm, and then the edge curve is described by features, such as calculating the curvature of the curve. At the same time, the shape features of the region are described by calculating parameters such as the area, perimeter, and aspect ratio of this block.

[0023] Adopt a color-based global feature extraction method to extract color features from the multi-scale image representation. For example, count the color histogram of the rocks around the karst cave entrance, that is, count the distribution frequencies of different colors, and calculate color moments, such as parameters like color mean, variance, and skewness to describe color features.

[0024] Combine the shape features and color features to generate a global feature description of the target geomorphic feature block. This global feature description can reflect the overall features of the target geomorphic feature block.

[0025] Map the global feature description to a high-dimensional feature space, and convert the low-dimensional global feature description into a high-dimensional feature representation through a non-linear mapping function (such as a mapping function based on a neural network).

[0026] Finally, based on the relative importance of the global features and local features in describing the target geomorphic feature block, adopt a feature fusion algorithm to fuse the global feature description and the local feature description to obtain the fused block semantic features. Integrate the fused block semantic features of all target geomorphic feature blocks to generate the corresponding basic karst geomorphic semantic features of the karst geomorphic image.

[0027] Step S120, determine the candidate karst geomorphic semantic features corresponding to each semantic space of the karst geomorphic image analysis model that has completed parameter learning based on the basic karst geomorphic semantic features.

[0028] In this embodiment, a karst geomorphic image analysis model that has completed parameter learning is configured. This karst geomorphic image analysis model has multiple semantic spaces, and each semantic space is used to process different feature aspects of the karst geomorphology.

[0029] First, perform dimensional normalization on the semantic features of the basic karst landform to generate the target karst landform semantic features. For example, the semantic features of the basic karst landform may include multi-dimensional information such as texture features, shape features, and color features at different scales. By using a specific normalization algorithm (such as mapping the value of each dimension to between 0 and 1), the target karst landform semantic features are obtained.

[0030] According to the predefined semantic space of the karst landform image analysis model, determine the feature dimensions or feature combinations corresponding to each semantic space. For example, one semantic space may correspond to the dimension related to the terrain undulation features of the karst landform, and another semantic space may correspond to the dimension related to the water body distribution features of the karst landform. And determine the correspondence between each feature item in the target karst landform semantic features and the semantic space according to the pre-established feature mapping table. For example, a feature item related to the terrain height difference in the target karst landform semantic features will be mapped to the semantic space related to the terrain undulation features according to the feature mapping table.

[0031] According to the feature mapping table, load each feature item in the target karst landform semantic features into the corresponding semantic space of the karst landform image analysis model to generate the candidate karst landform semantic features corresponding to each semantic space. For example, load the feature item related to the cave shape into the candidate karst landform semantic features corresponding to the semantic space dedicated to processing cave shape features.

[0032] Step S130, load the candidate karst landform semantic features corresponding to each semantic space into the karst landform image analysis sub-model existing in the semantic space corresponding to the karst landform image analysis model, so as to determine the karst landform description knowledge chain based on the sub-landform description feature parameters generated by each karst landform image analysis sub-model.

[0033] Suppose the karst landform image analysis model contains multiple karst landform image analysis sub-models, and these karst landform image analysis sub-models respectively process the features of different semantic spaces.

[0034] First, identify the different semantic spaces in the karst landform image analysis model and the corresponding karst landform image analysis sub-models. For example, there is a semantic space regarding the vegetation coverage features of the karst landform, and the corresponding karst landform image analysis sub-model is a model dedicated to analyzing features such as vegetation texture and color. For each semantic space, determine the position and functional features of this semantic space in the overall architecture of the karst landform image analysis model, and establish a mapping relationship table. For example, the semantic space of vegetation coverage features is in the part of the model architecture for analyzing surface features, and its function is to provide feature descriptions related to vegetation. In the mapping relationship table, this semantic space is an entry, and the corresponding karst landform image analysis sub-model is the associated content.

[0035] For each semantic space, the candidate karst landform semantic features corresponding to that semantic space are loaded into the corresponding karst landform image analysis sub-model batch by batch or all at once according to the input interface specifications of the corresponding karst landform image analysis sub-model. For example, for the candidate karst landform semantic features corresponding to the semantic space of vegetation coverage features, the data may be loaded in batches according to the requirements of the vegetation analysis sub-model. Each karst landform image analysis sub-model processes the loaded candidate karst landform semantic features to generate sub-landform description feature parameters corresponding to the input candidate karst landform semantic features. For example, after the vegetation analysis sub-model processes the candidate karst landform semantic features of vegetation coverage features, sub-landform description feature parameters such as vegetation type and vegetation density are generated, and these parameters reflect the feature descriptions of karst landforms by the karst landform image analysis sub-model in different semantic spaces.

[0036] Arrange the sub-landform description feature parameters according to the order of semantic spaces or the predefined logical order of the karst landform image analysis sub-model. For example, first arrange the sub-landform description feature parameters related to terrain features, and then arrange the sub-landform description feature parameters related to vegetation features, etc. Analyze the potential relationships between the sub-landform description feature parameters to identify the existing target correlation points.

[0037] Construct a feature identification label for each sub-landform description feature parameter. For example, for the sub-landform description feature parameter representing the height of a mountain peak, its feature identification label may be "mountain peak - height". Through the predefined karst landform semantic knowledge base, map the feature identification label to a further semantic description, and attach the mapped implementation semantic description to the corresponding sub-landform description feature parameter. For example, "mountain peak - height" may be mapped to "the vertical height of the mountain peak in karst landforms, which has an important impact on the undulation degree of the landform".

[0038] Calculate the semantic similarity between each sub-landform description feature parameter based on the basic semantic distance calculation strategy, and group the sub-landform description feature parameters according to the semantic similarity to obtain an initial grouping sequence. For example, the sub-landform description feature parameters related to the height of a mountain peak and the slope of a mountain peak may be grouped into the same group because of semantic similarity, as they are both related to the terrain features of the mountain peak.

[0039] Optimize the initial grouping sequence to obtain the optimized target grouping sequence. Check the number of sub-geomorphic description feature parameters within each initial group. If the number of sub-geomorphic description feature parameters within any one initial group is greater than the first set number (assumed to be 5), then re-use the advanced semantic distance calculation strategy to calculate the semantic similarity between the sub-geomorphic description feature parameters within this initial group to further subdivide this initial group. For example, if there are 8 sub-geomorphic description feature parameters related to mountain peaks in a group, through the advanced semantic distance calculation strategy, it may be subdivided into more detailed groups related to mountain peak height, mountain peak slope, mountain peak surface area, etc. If the number of sub-geomorphic description feature parameters within any one initial group is less than the second set number (assumed to be 2), then merge this initial group with the adjacent group, and then re-evaluate the semantic similarity of the sub-geomorphic description feature parameters within the merged group. If it meets the preset semantic similarity requirements, then retain the merged group, otherwise maintain the original group status.

[0040] For each group in the optimized target grouping sequence, analyze the relative positions of the geomorphic features corresponding to the sub-geomorphic description feature parameters within the group in the karst geomorphic image. For example, for a group containing mountain peak height and mountain peak slope, if their corresponding geomorphic features are adjacent in the image, then after marking these sub-geomorphic description feature parameters as sub-geomorphic description feature parameters with spatial association, add the corresponding spatial association information to the corresponding group to generate a group combined with spatial relationships.

[0041] Analyze the feature dependence relationships between the sub-geomorphic description feature parameters within each group combined with spatial relationships. Within each group, if the geomorphic feature corresponding to one sub-geomorphic description feature parameter is a dependence condition for the geomorphic feature corresponding to another sub-geomorphic description feature parameter, then mark them as sub-geomorphic description feature parameters with feature dependence relationships, and add the corresponding feature dependence relationship information to the group to generate a group combined with feature dependence relationships. For example, in a group containing the size of the cave entrance and the length of the internal passage of the cave, the size of the cave entrance may be a dependence condition for the length of the internal passage of the cave (a larger entrance may mean a relatively longer internal passage), then mark their feature dependence relationship.

[0042] In each group combined with feature dependence relationships, search for the sub-geomorphic description feature parameter with the most association relationships and determine it as the target association point. For example, in a group containing multiple sub-geomorphic description feature parameters related to mountain peaks, the mountain peak height may be determined as the target association point because it has more semantic similarity associations, spatial relationship associations, and feature dependence relationship associations with other parameters.

[0043] Take each target associated point as a node in the target association structure, and for each target associated point, connect the relevant sub-geomorphic description feature parameters to this target associated point according to the association relationship between this target associated point and other sub-geomorphic description feature parameters, so as to construct the target association structure. For example, taking the mountain peak height as the target associated point, connect the sub-geomorphic description feature parameters related to it, such as the mountain peak slope and the mountain peak surface area, to this node.

[0044] Determine the starting point of the karst landform description knowledge chain, and the starting point is the sub-geomorphic description feature parameter related to the most significant feature of the karst landform. For example, if in the entire karst landform image, a huge sinkhole is the most significant feature, then the sub-geomorphic description feature parameter related to the sinkhole depth or diameter may be determined as the starting point.

[0045] According to the pre-determined construction rules, gradually connect other sub-geomorphic description feature parameters to the karst landform description knowledge chain based on the target association structure until all sub-geomorphic description feature parameters are incorporated into the karst landform description knowledge chain, and generate the final karst landform description knowledge chain. For example, starting from this starting point of the sinkhole depth, connect the sub-geomorphic description feature parameters related to the terrain and vegetation around the sinkhole in sequence according to the target association structure, and gradually construct a complete karst landform description knowledge chain.

[0046] Step S140, obtain the karst landform description knowledge chain generated by the karst landform image analysis model, and cluster the karst landform image based on the karst landform description knowledge chain.

[0047] In this embodiment, the karst landform description knowledge chain includes the relationships between various features in the karst landform image and detailed landform descriptions.

[0048] Cluster the karst landform image based on this karst landform description knowledge chain. For example, if the karst landform description knowledge chain contains different types of karst landform features, such as a group of sub-geomorphic description feature parameters with sinkholes as the main feature, and another group of sub-geomorphic description feature parameters with a cluster of karst caves as the main feature, etc. The karst landform image can be clustered according to these different feature combinations. Images with similar karst landform description knowledge chains (i.e., similar landform feature combinations and relationships) are grouped into one category. For example, those karst landform images that mainly feature large sinkholes and sparse vegetation are clustered together, while those karst landform images that mainly feature a dense cluster of karst caves and abundant underground rivers are clustered into another category. In this way, through clustering based on the karst landform description knowledge chain, the karst landform image can be better classified and analyzed, which helps to further study the karst landform, such as studying the distribution laws and formation mechanisms of different types of karst landforms.

[0049] Based on the above steps, in the embodiment of the present application, by extracting the basic semantic features of the karst landform image and using the pre-trained karst landform image analysis model, the candidate karst landform semantic features corresponding to each semantic space are accurately determined. These candidate karst landform semantic features are loaded into the corresponding sub-models to generate sub-landform description feature parameters, and then a detailed karst landform description knowledge chain is constructed. This method not only significantly improves the accuracy and efficiency of karst landform image recognition, but also effectively clusters the images through the karst landform description knowledge chain, providing a data analysis means for scientific research exploration, geological disaster prediction, environmental protection and other fields of karst landforms.

[0050] In a possible implementation manner, step S110 specifically includes:

[0051] Step S111, decompose the karst landform image to generate a plurality of target landform feature blocks.

[0052] Step S112, extract the block semantic features corresponding to each target landform feature block, and determine the basic karst landform semantic features corresponding to the karst landform image based on the block semantic features corresponding to each target landform feature block.

[0053] In a possible implementation manner, after step S111, the method further includes:

[0054] Step S1111, obtain a sequence of karst landform feature screening maps. When there is no landform feature block in the plurality of target landform feature blocks that matches the landform feature blocks in the sequence of karst landform feature screening maps, then screen out the karst landform image.

[0055] Step S1112, when there is any one of the plurality of target landform feature blocks that matches the landform feature blocks in the sequence of karst landform feature screening maps, then jump to step S112.

[0056] In this embodiment, for the process of extracting the basic karst landform semantic features corresponding to the karst landform image, for a karst landform image from a specific area, this karst landform image contains rich karst landform information, such as mountain peaks of various shapes, karst caves, and surface signs of underground rivers. First, decompose this karst landform image to generate multiple target landform feature blocks. For example, based on obvious natural boundaries such as the contour of the mountain peak, the entrance of the karst cave, and the meandering trend of the underground river on the surface, the image is segmented into different target landform feature blocks. One target landform feature block may be a single mountain peak area, and another may be an area containing the entrance of the karst cave and its surrounding rock structures. Next, extract the corresponding block semantic features for each target landform feature block, and determine the basic karst landform semantic features corresponding to the karst landform image based on the block semantic features corresponding to each target landform feature block.

[0057] Among them, after decomposing the karst landform image to generate multiple target landform feature blocks, a sequence of karst landform feature screening maps will be obtained. This sequence of karst landform feature screening maps contains some specific landform feature patterns, and these landform feature patterns may be landform features that were previously determined not to require further analysis or do not conform to the current research focus. For example, this sequence of screening maps may contain some small landform feature patterns that have been fully studied and have little relevance to the current research purpose, such as image patterns of some isolated and very small corrosion pits. Search among the multiple target landform feature blocks. If there is no landform feature block that matches the landform feature patterns in the sequence of karst landform feature screening maps, then this karst landform image is screened out because this karst landform image may not contain the karst landform features of interest in this embodiment or the features it contains have been excluded from the research scope. If there is any landform feature block that matches the landform feature patterns in the sequence of karst landform feature screening maps among the multiple target landform feature blocks, then jump to the step of extracting the corresponding block semantic features for each target landform feature block and continue with the subsequent analysis and processing.

[0058] In a possible implementation manner, step S112 specifically includes:

[0059] Step S1121, successively perform unified processing of the image format, standardized processing of the image size, and normalization processing on each target landform feature block to generate the preprocessed target landform feature blocks.

[0060] In this embodiment, taking the target geomorphic feature block containing the mountain area mentioned above as an example, assume that the format of the originally collected image is a special format that is only applicable to specific acquisition devices, and this format may have compatibility issues in subsequent processing algorithms. Through the unified processing of the image format, it is converted into a general format that is widely supported by image processing algorithms, such as the TIFF format. The standardization processing of the image size is to ensure the consistency of different target geomorphic feature blocks in terms of size during the subsequent processing, facilitating unified algorithm operations. For the target geomorphic feature block of the mountain area, if its original size is large and irregular, for example, 800 pixels in length and 600 pixels in width, it is adjusted to a preset standard size, such as 400 pixels in both length and width. The normalization processing is to map the pixel values of the image to a specific interval, such as normalizing the pixel values to between 0 and 1, which helps to improve the accuracy and efficiency of subsequent algorithm processing. After this series of preprocessing operations, the preprocessed target geomorphic feature block suitable for subsequent analysis is obtained.

[0061] Step S1122, using a multi-scale image analysis algorithm, construct a multi-scale image representation of the preprocessed target geomorphic feature block. Specifically, first determine the scale range to be used. For each scale in the scale range, obtain scale feature images at different scales through downsampling and upsampling algorithms. Among them, mean pooling method is used for downsampling, and bilinear interpolation method is used for upsampling. Then, combine the scale feature images at different scales to generate the multi-scale image representation.

[0062] Taking the target geomorphic feature block containing the entrance of the karst cave and its surrounding rock structures as an example, first determine the scale range to be used. According to the experience and requirements of karst landform image analysis, it is determined that the scale range from 1 / 3 to 1 / 9 of the original size is appropriate. For each scale in this scale range, obtain scale feature images at different scales through downsampling and upsampling algorithms. Mean pooling method is used for downsampling. For example, at a certain scale, divide the target geomorphic feature block into several small regions, calculate the average value of the pixels in each small region as the pixel value after downsampling, thereby reducing the number of pixels in the image. Bilinear interpolation method is used for upsampling. When it is necessary to restore the low-resolution downsampled image to a higher resolution, calculate the new pixel value through bilinear interpolation according to the values of the surrounding pixels, thereby obtaining scale feature images at different scales. Finally, combine these scale feature images at different scales to generate the multi-scale image representation of the target geomorphic feature block of the entrance of the karst cave and its surrounding rock structures. This multi-scale image representation can comprehensively reflect the feature information of the target geomorphic feature block at different scales, which helps to extract more accurate features subsequently.

[0063] Step S1123: Using a local feature description algorithm, at each scale of the multi-scale image representation, construct a Difference of Gaussian (DoG) pyramid to detect local extreme points. For each detected local extreme point, calculate the corresponding orientation histogram to describe the texture features around the local extreme point. The calculation of the orientation histogram is based on the pixel gray value changes within a preset range around the local extreme point. The pixels within the preset range are statistically analyzed according to predefined orientation intervals to obtain a histogram representing the orientation distribution.

[0064] Step S1124: Combine the orientation histograms of all local extreme points to generate the local feature description of the target geomorphic feature block, which is used to reflect the local texture and structural features of the target geomorphic feature block.

[0065] In this embodiment, a local feature description algorithm is used to construct a Difference of Gaussian (DoG) pyramid at each scale of the multi-scale image representation to detect local extreme points. For the target geomorphic feature block containing surface signs of an underground river, construct a Difference of Gaussian (DoG) pyramid at each scale of its multi-scale image representation. During the construction process, convolve the image with Gaussian kernel functions of different scales to detect local extreme points. For each detected local extreme point, calculate the corresponding orientation histogram to describe the texture features around the local extreme point. For example, for the rock texture around the underground river, taking a detected local extreme point as the center, select the pixel gray value changes within a preset range (such as a 10×10 pixel range) around it to calculate the orientation histogram. Statistically analyze according to predefined orientation intervals (such as dividing 0 - 360 degrees into 8 intervals), and classify the pixels within this range into the corresponding intervals according to the direction of their gray value changes to obtain a histogram representing the orientation distribution. This orientation histogram can describe in detail the texture features around the local extreme point, such as the trend and variation of the rock texture around the underground river. Combine the orientation histograms of all local extreme points to generate the local feature description of the surface signs of the underground river, which effectively reflects the local texture and structural features of the target geomorphic feature block.

[0066] Step S1125: Adopt a sorting algorithm based on feature importance to sort the local features in the local feature description according to their importance in describing the target geomorphic feature block. After removing the local features with importance lower than the set importance, fuse the similar local features among the remaining local features through a feature fusion algorithm to generate an optimized local feature description, where the evaluation of feature importance is determined based on the occurrence frequency of local features in the entire local feature description and their correlation with other local features.

[0067] Taking the target geomorphic feature block containing the mountain peak area as an example, in its local feature description, those local features related to the contour shape of the mountain peak, such as the local features corresponding to the sharp shape at the top of the mountain peak, may have relatively high importance when describing the mountain peak area. While some local features caused by noise or local minor changes may be relatively unimportant. The evaluation of feature importance is determined based on the occurrence frequency of the local feature in the entire local feature description and its correlation with other local features. For example, if a certain local feature appears frequently in the local feature description and has a strong correlation with other local features that can reflect the main features of the mountain peak, then the importance of this local feature is relatively high. After removing the local features with importance lower than the set importance (this set importance is a threshold determined based on a large amount of analysis and experience of karst geomorphic images), the similar local features among the remaining local features are fused through a feature fusion algorithm to generate an optimized local feature description. For example, some local features describing the similar directions of rock textures in the mountain peak area are fused to obtain a more representative optimized local feature description.

[0068] Step S1126: Use a shape-based global feature extraction method to extract shape features from the multi-scale image representation. The shape features include contour features and regional shape features. Among them, the contour features obtain the edge curve of the target geomorphic feature block through an edge detection algorithm, and perform feature description on the edge curve. The regional shape features are described by calculating the area, perimeter, and aspect ratio parameters of the target geomorphic feature block.

[0069] In this embodiment, for the target geomorphic feature block containing the cave entrance and its surrounding rock structures, the shape features include contour features and regional shape features. The contour features obtain the edge curve of the target geomorphic feature block through an edge detection algorithm. For example, the Canny edge detection algorithm is used to accurately detect the edge curve of the cave entrance, and then feature description is performed on the edge curve. Features such as the curvature of the edge curve and the length of the curve can be calculated, and these features can reflect the shape characteristics of the cave entrance. The regional shape features are described by calculating parameters such as the area, perimeter, and aspect ratio of the target geomorphic feature block. For example, calculate the area, perimeter, and aspect ratio occupied by the cave entrance and its surrounding rock structures, and these parameters can describe the shape characteristics of this area as a whole.

[0070] Step S1127: Use a color-based global feature extraction method to extract color features from the multi-scale image representation. The color features include color histograms and color moments. The color histogram statistically represents the distribution frequency of different colors in the target geomorphic feature block, and the color moment describes the color features by calculating the color mean, variance, and skewness parameters of the target geomorphic feature block.

[0071] Taking the target geomorphic feature block containing the mountain peak area as an example, the color features include color histogram and color moment. The color histogram is to statistically analyze the distribution frequencies of different colors in the target geomorphic feature block. For example, the proportion of the number of pixels with different gray values or different color channels (if it is a color image) in the mountain peak area to the total number of pixels is statistically analyzed to obtain the color histogram. The color moment is to describe the color features by calculating the color mean, variance, and skewness parameters of the target geomorphic feature block. Calculating the mean of the colors in the mountain peak area can understand the average color level of this area, the variance can reflect the degree of color dispersion, and the skewness can reflect the asymmetry of the color distribution.

[0072] Step S1128: Combine the shape feature and the color feature to generate a global feature description of the target geomorphic feature block, where the global feature description is used to reflect the overall feature of the target geomorphic feature block.

[0073] For the target geomorphic feature block containing the surface signs of an underground river, combine the previously extracted shape features (such as the shape features of the contour curve and the area, perimeter, etc. of the area) and color features (color histogram and color moment) of the area around the underground river. This global feature description can comprehensively reflect the overall features of the target geomorphic feature block, such as the overall shape and color distribution and other comprehensive features of the underground river surface sign area.

[0074] Step S1129: Map the global feature description to a high-dimensional feature space, convert the low-dimensional global feature description into a high-dimensional feature representation through a non-linear mapping function, and based on the relative importance of the global feature and the local feature in describing the target geomorphic feature block, use a feature fusion algorithm to fuse the global feature description and the local feature description to obtain the fused block semantic feature, and integrate the fused block semantic features of all target geomorphic feature blocks to generate the basic karst geomorphic semantic feature corresponding to the karst geomorphic image.

[0075] In this embodiment, taking the target geomorphic feature block containing the entrance of a karst cave and its surrounding rock structure as an example, use a pre-determined non-linear mapping function (such as a mapping function based on a neural network) to convert the originally low-dimensional global feature description (a feature description composed of shape features and color features, etc.) into a high-dimensional feature representation. This high-dimensional feature representation can more richly represent the feature information of the target geomorphic feature block, which helps to better distinguish different target geomorphic feature blocks in subsequent analysis.

[0076] Furthermore, taking the target geomorphic feature block containing the mountain peak area as an example, if it is found through evaluation that the global features are highly important in describing the overall shape and color distribution of the mountain peak, while the local features are more crucial in describing the local texture and special structures of the mountain peak, according to the pre-determined fusion algorithm (such as the weighted average fusion algorithm), the global feature description and the local feature description are fused according to their relative importance. After fusing the global feature description (including shape and color features) and the local feature description (such as the texture and structure features of the local mountain peak) of the mountain peak area, the fused block semantic feature is obtained, which can comprehensively reflect various feature information of the mountain peak area.

[0077] Finally, in a karst geomorphic image, there are multiple target geomorphic feature blocks, including the mountain peak area, the cave entrance area, the surface indication area of the underground river, etc. The fused block semantic features of these areas are obtained respectively. The fused block semantic features of these different target geomorphic feature blocks are integrated. The integration process may include operations such as combining and arranging different features according to certain rules, and finally generating the basic karst geomorphic semantic feature that can comprehensively reflect the features of the entire karst geomorphic image. This basic karst geomorphic semantic feature covers various geomorphic feature information in the karst geomorphic image and provides a basis for subsequent analysis and processing.

[0078] In a possible implementation manner, step S120 includes:

[0079] Step S121, performing dimensional normalization processing on the basic karst geomorphic semantic feature to generate the target karst geomorphic semantic feature.

[0080] In this embodiment, taking the basic karst geomorphic semantic feature extracted and integrated from the karst geomorphic image before as an example, these basic karst geomorphic semantic features may contain various feature information from different target geomorphic feature blocks, such as the shape, texture, and color features of the mountain peak area, the relevant features of the cave entrance area, and the features of the surface indication area of the underground river, etc. These features may have different value ranges and data distributions in terms of dimensions. The dimensional normalization processing aims to perform unified standardization operations on these features of different dimensions, so that each feature dimension is within a predetermined standard range. For example, for the feature dimension representing the height of the mountain peak, its original numerical value may vary within a relatively large numerical interval, and through dimensional normalization processing, it is mapped to the numerical range between 0 and 1; for the feature dimension representing the average color of the cave entrance, a similar normalization operation is also performed. In this way, each feature dimension in the entire basic karst geomorphic semantic feature is processed, and finally the target karst geomorphic semantic feature is generated. This target karst geomorphic semantic feature has consistency in terms of dimensions and is convenient for subsequent interaction operations with the karst geomorphic image analysis model.

[0081] Step S122: Determine the feature dimension or feature combination corresponding to each semantic space according to the predefined semantic space of the karst landform image analysis model, and determine the corresponding relationship between each feature item in the target karst landform semantic feature and the semantic space according to the pre-established feature mapping table.

[0082] Next, assume that the karst landform image analysis model has multiple predefined semantic spaces. One semantic space is defined as a space dedicated to processing features related to the terrain undulation of karst landforms. The feature dimensions corresponding to this semantic space may include feature dimensions such as the height of mountain peaks, slopes, and the depth of valleys, or combinations of these feature dimensions; another semantic space may be a space dedicated to processing features related to water bodies in karst landforms, and the corresponding feature dimensions may be feature dimensions such as the width of underground rivers and the area of water bodies in karst caves, or combinations of them. The pre-established feature mapping table details the mapping relationship between each feature item in the target karst landform semantic feature and these semantic spaces. For example, in the target karst landform semantic feature, the feature item of mountain peak height will be mapped to the semantic space for processing features related to terrain undulation according to the feature mapping table; while the feature item of the area of water bodies in karst caves will be mapped to the semantic space for processing features related to water bodies. In this way, the corresponding relationship between each feature item and the semantic space is accurately determined.

[0083] Step S123: According to the feature mapping table, load each feature item in the target karst landform semantic feature into the corresponding semantic space of the karst landform image analysis model to generate candidate karst landform semantic features corresponding to each semantic space.

[0084] Finally, for the semantic space for processing features related to terrain undulation, load the feature items related to the height of mountain peaks, slopes, and the depth of valleys, etc. in the target karst landform semantic feature into this semantic space according to the feature mapping table. These loaded feature items constitute the candidate karst landform semantic features corresponding to this semantic space; for the semantic space for processing features related to water bodies, load the feature items related to the width of underground rivers and the area of water bodies in karst caves, etc. in the target karst landform semantic feature, so as to generate the candidate karst landform semantic features corresponding to this semantic space. For other semantic spaces in the karst landform image analysis model, in the same way, load the corresponding feature items into their respective semantic spaces according to the feature mapping table, and finally generate candidate karst landform semantic features corresponding to each semantic space. These candidate karst landform semantic features will serve as the basis for subsequent operations and will be further loaded into the karst landform image analysis sub-model existing in the semantic space corresponding to the karst landform image analysis model for processing, thus promoting the progress of the entire karst landform image analysis process.

[0085] In a possible implementation manner, before the step S130, the method further includes:

[0086] Step A110: Obtain a sample karst landform image, and extract the basic sample karst landform semantic features corresponding to the sample karst landform image.

[0087] In this embodiment, these sample karst landform images are selected and represent karst landform situations of different types and different characteristics. For example, multiple sample images are selected from a specific karst landform research area. Some of the images mainly show large-scale peak forest landforms, some images focus on presenting the complex internal structure of karst caves, and some images reflect the interaction between surface rivers and karst landforms. These sample karst landform images are processed to extract the corresponding basic sample karst landform semantic features. Taking the sample image of the peak forest landform as an example, operate according to the method for extracting basic karst landform semantic features from karst landform images described above. First, decompose the sample image of the peak forest landform to generate multiple target landform feature blocks, which may include areas such as a single mountain peak and valleys in a mountain peak group. Then, perform unified processing of the image format, standardized processing of the image size, and normalization processing on each target landform feature block in sequence to generate the preprocessed target landform feature block. After that, use a multi-scale image analysis algorithm to construct its multi-scale image representation, use a local feature description algorithm to obtain a local feature description, use a sorting algorithm based on feature importance to perform feature screening and fusion to obtain an optimized local feature description, then use a global feature extraction method based on shape and color to obtain a global feature description, map the global feature description to a high-dimensional feature space, and finally fuse based on the relative importance of the global feature and the local feature to obtain the fused block semantic feature. Integrate the fused block semantic features of all target landform feature blocks to obtain the basic sample karst landform semantic features corresponding to the sample image of the peak forest landform.

[0088] Step A120: Determine the input sample karst landform semantic features corresponding to each semantic space of the karst landform image analysis model based on the basic sample karst landform semantic features.

[0089] Suppose the karst landform image analysis model has specialized processing for the semantic spaces of terrain structure, vegetation cover, water body distribution, etc. For the basic example karst landform semantic features of the peak forest landform example image, the features related to the height and slope of the peaks, etc. will be determined as the input example karst landform semantic features corresponding to the terrain structure semantic space; the features related to the types and densities of vegetation in the peak forest area will be determined as the input example karst landform semantic features corresponding to the vegetation cover semantic space; and the features related to surface rivers or water bodies in the karst caves will be determined as the input example karst landform semantic features corresponding to the water body distribution semantic space. In this way, according to the definitions and requirements of different semantic spaces, the input example karst landform semantic features corresponding to each semantic space are determined from the basic example karst landform semantic features.

[0090] Step A130, obtain the example karst landform description knowledge chain corresponding to the example karst landform image.

[0091] For the peak forest landform example image mentioned above, its example karst landform description knowledge chain includes the description of the relationships between the various features of the peak forest landform and the overall landform feature description. For example, in this knowledge chain, there may be a certain relationship between the height and slope of the peaks and the vegetation cover situation, and this relationship may be based on the laws obtained from on-site investigations or previous studies. At the same time, the distribution of surface rivers may also have some connection with the formation and evolution of the peaks, and these relationships will be recorded in the example karst landform description knowledge chain. For the example image of the internal structure of a karst cave, its example karst landform description knowledge chain will describe the relationships between the size and shape of the karst cave and the internal rock structure, water body distribution, and possible signs of biological activities, etc.

[0092] Step A140, load the input example karst landform semantic features corresponding to each semantic space into the semantic space corresponding to the karst landform image analysis model, determine the example karst landform description knowledge chain as the output result expected to be learned by the karst landform image analysis model, perform parameter learning on the karst landform image analysis model to generate a target karst landform image analysis model, and the target karst landform image analysis model includes multiple karst landform image analysis sub-models, and the neuron weight parameters in each karst landform image analysis sub-model are different.

[0093] Taking the processing of the semantic space of the terrain structure as an example, the karst landform semantic features of the input samples corresponding to the previously determined semantic space are loaded into the module corresponding to the terrain structure semantic space in the karst landform image analysis model. The same operation is also performed on other semantic spaces such as the vegetation coverage semantic space and the water body distribution semantic space. After the loading is completed, the karst landform image analysis model performs parameter learning with the sample karst landform description knowledge chain as the expected learning output result. During the learning process, the weight parameters of each neuron inside the model will be continuously adjusted. Since the target karst landform image analysis model includes multiple karst landform image analysis sub-models, the weight parameters of the neurons in each karst landform image analysis sub-model are different. For example, in the karst landform image analysis sub-model for processing the terrain structure semantic space, the weight parameters of the neurons will be adjusted according to the corresponding parts of the input sample karst landform semantic features related to the terrain structure and the sample karst landform description knowledge chain, so that this sub-model can better analyze and process the features related to the terrain structure. For other karst landform image analysis sub-models, such as the sub-models for processing vegetation coverage and water body distribution, similar parameter adjustments will also be made according to their respective corresponding input sample karst landform semantic features and the sample karst landform description knowledge chain, and finally a target karst landform image analysis model is generated, which can process and analyze more accurately in the subsequent analysis of karst landform images.

[0094] In a possible implementation manner, step S130 specifically includes:

[0095] Step S131, identifying different semantic spaces in the karst landform image analysis model and the karst landform image analysis sub-models corresponding to the semantic spaces. For each semantic space, determine the position and functional characteristics of the semantic space in the overall architecture of the karst landform image analysis model, and establish a mapping relationship table. Each semantic space in the mapping relationship table is used as an entry, and the corresponding karst landform image analysis sub-model is used as the associated content.

[0096] Step S132, for each semantic space, load the candidate karst landform semantic features corresponding to the semantic space into the corresponding karst landform image analysis sub-model batch by batch or at one time according to the input interface specification of the corresponding karst landform image analysis sub-model. Process the loaded candidate karst landform semantic features through each karst landform image analysis sub-model to generate sub-landform description feature parameters corresponding to the input candidate karst landform semantic features. The sub-landform description feature parameters reflect the feature description of the karst landform by the karst landform image analysis sub-model in different semantic spaces.

[0097] Step S133: Arrange each of the sub-geomorphic description feature parameters in the order of the semantic space or the predefined logical order of the karst landform image analysis sub-model, analyze the potential relationships between the sub-geomorphic description feature parameters to identify the existing target correlation points, and establish a target correlation structure based on the target correlation points.

[0098] Step S134: Determine the starting point of the karst landform description knowledge chain, where the starting point is the sub-geomorphic description feature parameter related to the most significant feature of the karst landform.

[0099] Step S135: According to the pre-determined construction rules, gradually connect the other sub-geomorphic description feature parameters to the karst landform description knowledge chain based on the target correlation structure until all the sub-geomorphic description feature parameters are incorporated into the karst landform description knowledge chain to generate the final karst landform description knowledge chain.

[0100] In a possible implementation manner, Step S133 includes:

[0101] Step S1331: Construct a feature identification label for each sub-geomorphic description feature parameter, and the feature identification label is used to describe the feature of the karst landform represented by the sub-geomorphic description feature parameter.

[0102] Step S1332: Through the predefined karst landform semantic knowledge base, map the feature identification label to a further semantic description, and attach the implemented semantic description after mapping to the corresponding sub-geomorphic description feature parameter.

[0103] Step S1333: Calculate the semantic similarity between each sub-geomorphic description feature parameter based on the basic semantic distance calculation strategy, and group the sub-geomorphic description feature parameters according to the semantic similarity to obtain an initial grouping sequence. Each initial grouping in the initial grouping sequence reflects the potential relationship based on semantics between the sub-geomorphic description feature parameters, and the sub-geomorphic description feature parameters within the same initial grouping are relevant at the partial feature level of the karst landform.

[0104] Step S1334: Optimize the initial grouping sequence to obtain an optimized target grouping sequence. Specifically, check the number of sub-geomorphic description feature parameters in each initial group. If the number of sub-geomorphic description feature parameters in any initial group is greater than the first set number, then re - adopt the advanced semantic distance calculation strategy to calculate the semantic similarity between the sub-geomorphic description feature parameters in this initial group to further subdivide this initial group. If the number of sub-geomorphic description feature parameters in any initial group is less than the second set number, then merge this initial group with the adjacent group and re - evaluate the semantic similarity of the sub-geomorphic description feature parameters in the merged group. If it meets the preset semantic similarity requirement, then retain the merged group; otherwise, maintain the original group status, where the second set number is less than the first set number.

[0105] Step S1335: For each group in the optimized target grouping sequence, analyze the relative positions of the geomorphic features corresponding to the sub-geomorphic description feature parameters within the group in the karst geomorphic image. If the geomorphic features corresponding to any two or more sub-geomorphic description feature parameters are adjacent or have an inclusion relationship in the karst geomorphic image, then after marking the any two or more sub-geomorphic description feature parameters as sub-geomorphic description feature parameters with spatial association, add the corresponding spatial association information to the corresponding group to generate a group combined with spatial relationships.

[0106] Step S1336: Analyze the feature dependency relationships between the sub-geomorphic description feature parameters within each group combined with spatial relationships. Within each group, if the geomorphic feature corresponding to one sub-geomorphic description feature parameter is a dependency condition for the geomorphic feature corresponding to another sub-geomorphic description feature parameter, then mark them as sub-geomorphic description feature parameters with feature dependency relationships and add the corresponding feature dependency relationship information to the group to generate a group combined with feature dependency relationships.

[0107] Step S1337: In each group combined with feature dependency relationships, search for the sub-geomorphic description feature parameter with the most association relationships and determine the sub-geomorphic description feature parameter with the most association relationships as the target association point, where the association relationships include semantic similarity association, spatial relationship association, and feature dependency relationship association.

[0108] Step S1338: Take each of the target association points as a node in the target association structure, and for each target association point, connect the relevant sub-geomorphic description feature parameters to this target association point according to the association relationship between this target association point and other sub-geomorphic description feature parameters to construct the target association structure.

[0109] In this embodiment, the karst landform image analysis model includes multiple semantic spaces. For example, there is a semantic space dedicated to processing the topographic features of karst landforms, such as semantic spaces related to features like peak height, valley depth, and slope gradient; there is also a semantic space related to water body features in karst landforms, including the flow rate of underground rivers, the water area in karst caves, etc.; and a semantic space related to vegetation cover, involving the type, density, and distribution of vegetation. Each semantic space has a corresponding karst landform image analysis sub-model. Taking the semantic space of topographic features as an example, its corresponding karst landform image analysis sub-model is a module specifically used to process terrain-related data, and its internal algorithms and structures are optimized for the analysis and processing of topographic features.

[0110] For each semantic space, determine its position and functional characteristics in the overall architecture of the karst landform image analysis model. The semantic space of topographic features may be located in the part of the model architecture for analyzing the basic structure of the landform, and its function is to accurately extract and analyze the terrain-related features of karst landforms. The semantic space of water body features may be under the branch of water resource-related analysis, and its main function is to describe the water body situation in karst landforms. The semantic space of vegetation cover is in the area for analyzing surface coverings and is responsible for processing vegetation-related information. Then, establish a mapping relationship table. In this table, the semantic space of topographic features is taken as an entry, and the corresponding karst landform image analysis sub-model (the module for processing terrain data) is taken as the associated content. Similarly, the semantic space of water body features and its corresponding sub-model, the semantic space of vegetation cover and its corresponding sub-model, etc. are all associated in the mapping relationship table one by one.

[0111] Specifically, for each semantic space, load the candidate karst landform semantic features corresponding to this semantic space according to the input interface specifications of the corresponding karst landform image analysis sub-model. Taking the semantic space of topographic features as an example, if the input interface specifications of its corresponding karst landform image analysis sub-model require data to be arranged in a specific format and order, then process the candidate karst landform semantic features corresponding to the semantic space of topographic features according to this requirement. Suppose the candidate karst landform semantic features include data such as peak height values and valley depth values. It may be necessary to convert these data into a specific data type (such as floating-point numbers) and arrange them in the specified order. For the semantic space of water body features, if the input interface specifications of its corresponding sub-model require data to be input in batches, then load the candidate karst landform semantic features related to water bodies (such as underground river flow rate values, water area values in karst caves, etc.) into the corresponding karst landform image analysis sub-model batch by batch according to this requirement.

[0112] Each karst landform image analysis sub-model processes the loaded candidate karst landform semantic features. The terrain feature sub-model performs complex calculations and analyses on candidate karst landform semantic features such as the height of peaks and the depth of valleys input. For example, it may calculate sub-geomorphic description feature parameters such as the undulation degree of the terrain and the complexity of the terrain according to internal algorithms. These sub-geomorphic description feature parameters reflect the feature descriptions of karst landforms by the karst landform image analysis sub-model in different semantic spaces. After the water body feature sub-model processes candidate karst landform semantic features such as the underground river flow rate and the water body area in a cave, it generates sub-geomorphic description feature parameters related to the water body, such as the dynamic stability of the water body and the interaction degree between the water body and the surrounding landforms.

[0113] Arrange each sub-geomorphic description feature parameter according to the order of the semantic space or the predefined logical order of the karst landform image analysis sub-model. For example, first arrange the sub-geomorphic description feature parameters related to terrain features, such as the undulation degree of the terrain and the slope of the slope surface, then arrange the sub-geomorphic description feature parameters related to water body features, such as the dynamic stability of the water body and the water body area in a cave, and finally arrange the sub-geomorphic description feature parameters related to vegetation cover, such as vegetation density and vegetation type.

[0114] Analyze the potential relationships between each sub-geomorphic description feature parameter to identify the existing target correlation points. First, construct a feature identification label for each sub-geomorphic description feature parameter. For the sub-geomorphic description feature parameter of the undulation degree of the terrain, its feature identification label may be "terrain-undulation degree". Through the predefined karst landform semantic knowledge base, map this feature identification label to a further semantic description. In the karst landform semantic knowledge base, "terrain-undulation degree" may be mapped to "the amplitude of the height change of the terrain surface in karst landforms relative to sea level, reflecting the result of karstification on terrain shaping", and then attach this mapped implementation semantic description to the corresponding sub-geomorphic description feature parameter.

[0115] Calculate the semantic similarity between each sub-geomorphic description feature parameter based on the basic semantic distance calculation strategy. For example, the two sub-geomorphic description feature parameters of the undulation degree of the terrain and the slope of the slope surface may have a high semantic similarity because they are both related to the shape and change of the terrain. According to the semantic similarity, group the sub-geomorphic description feature parameters to obtain an initial grouping sequence. The undulation degree of the terrain and the slope of the slope surface may be grouped into the same group because they are relevant at the terrain feature level of karst landforms. While the dynamic stability of the water body and the vegetation density may be grouped into different groups because they represent different aspects of landform features.

[0116] Optimize the initial grouping sequence. Check the number of sub-geomorphic description feature parameters within each initial grouping. Assume the first set number is 5. If the number of sub-geomorphic description feature parameters within a certain initial grouping is greater than 5, for example, a grouping containing 8 terrain-related sub-geomorphic description feature parameters, re-adopt the advanced semantic distance calculation strategy to calculate the semantic similarity between the sub-geomorphic description feature parameters within this initial grouping, so as to further subdivide this initial grouping. For example, this grouping may be subdivided into more detailed sub-groupings related to peak height, valley depth, slope gradient, etc. If the number of sub-geomorphic description feature parameters within any initial grouping is less than the second set number (assume the second set number is 2), then merge this initial grouping with the adjacent grouping, and re-evaluate the semantic similarity of the sub-geomorphic description feature parameters within the merged grouping. If the preset semantic similarity requirement is met, retain the merged grouping; otherwise, maintain the original grouping status.

[0117] For each grouping in the optimized target grouping sequence, analyze the relative positions of the geomorphic features corresponding to the sub-geomorphic description feature parameters within the group in the karst geomorphic image. For example, in a grouping containing peak height and slope gradient, if in the karst geomorphic image, the geomorphic features corresponding to the peak height and the slope gradient of the slope are adjacent, then after marking the two sub-geomorphic description feature parameters of peak height and slope gradient as sub-geomorphic description feature parameters with spatial association, add the corresponding spatial association information (such as the adjacent relationship) to the corresponding grouping to generate a grouping combined with spatial relationship.

[0118] Analyze the feature dependence relationship between the sub-geomorphic description feature parameters within each grouping combined with spatial relationship. Within a grouping containing the water area in the cave and the shape of the cave, if the shape of the cave is a dependent condition for the water area in the cave (because the shape of the cave will limit the water area), then mark them as sub-geomorphic description feature parameters with feature dependence relationship, and add the corresponding feature dependence relationship information (such as the limiting relationship of the cave shape on the water area) to the grouping to generate a grouping combined with feature dependence relationship.

[0119] In each group of combined feature dependencies, search for the sub-geomorphic description feature parameter with the most associated relationships and determine it as the target association point. For example, in a group containing multiple sub-geomorphic description feature parameters related to mountain peaks (such as mountain peak height, mountain peak slope, mountain peak surface area, etc.), the mountain peak height may be determined as the target association point because it has more semantic similarity associations, spatial relationship associations, and feature dependency associations with other parameters. Take each target association point as a node in the target association structure. For each target association point, connect the relevant sub-geomorphic description feature parameters to this target association point according to the association relationship between this target association point and other sub-geomorphic description feature parameters to construct the target association structure. For example, taking the mountain peak height as the target association point, connect the sub-geomorphic description feature parameters related to it, such as mountain peak slope and mountain peak surface area, to this node.

[0120] Determine the starting point of the karst landform description knowledge chain, which is the sub-geomorphic description feature parameter related to the most significant feature of the karst landform. For example, in a karst landform area, if there is a huge sinkhole, the sub-geomorphic description feature parameter related to the sinkhole depth or diameter may be determined as the starting point. Because the sinkhole is the most significant feature in this karst landform area, constructing the knowledge chain starting from this most significant feature can better reflect the feature relationships of the entire karst landform.

[0121] According to the pre-determined construction rules, gradually connect other sub-geomorphic description feature parameters to the karst landform description knowledge chain based on the target association structure until all sub-geomorphic description feature parameters are incorporated into the karst landform description knowledge chain to generate the final karst landform description knowledge chain. Starting from the starting point related to the sinkhole depth, according to the target association structure, if the sinkhole depth is related to the terrain undulation degree around the sinkhole, connect the sub-geomorphic description feature parameter of the terrain undulation degree to the knowledge chain. Then, if the terrain undulation degree is related to the vegetation distribution to a certain extent, connect the sub-geomorphic description feature parameter related to the vegetation distribution to the knowledge chain, and so on, gradually incorporating all sub-geomorphic description feature parameters into the karst landform description knowledge chain, and finally obtaining a complete knowledge chain that can comprehensively describe the feature relationships of the karst landform.

[0122] Figure 2 The hardware structure diagram of the deep learning-based karst landform image analysis system 100 provided by the embodiment of the present invention for implementing the above deep learning-based karst landform image analysis method is shown, as Figure 2 shown, the deep learning-based karst landform image analysis system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0123] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions for the deep learning-based karst landform image analysis system 100 to execute or use to complete the exemplary methods described in the present invention.

[0124] In a specific implementation process, one or more processors 110 execute the computer-executable instructions stored in the machine-readable storage medium 120, so that the processors 110 can execute the deep learning-based karst landform image analysis method in the above method embodiments. The processors 110, the machine-readable storage medium 120, and the communication unit 140 are connected through a bus 130, and the processors 110 can be used to control the transceiver actions of the communication unit 140.

[0125] For the specific implementation process of the processors 110, reference may be made to the various method embodiments executed by the above deep learning-based karst landform image analysis system 100. Their implementation principles and technical effects are similar, and will not be elaborated herein in this embodiment.

[0126] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the above deep learning-based karst landform image analysis method is implemented.

[0127] It should be noted that, in order to simplify the description of the present invention disclosure and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing, or description thereof.

Claims

1. A method for analyzing karst landform images based on deep learning, characterized in that, The method includes: Obtain a karst landform image to be subjected to image recognition, and extract the basic karst landform semantic features corresponding to the karst landform image; Based on the basic karst landform semantic features, determine the candidate karst landform semantic features corresponding to each semantic space of a pre-completed parameter learning karst landform image analysis model; Load the candidate karst landform semantic features corresponding to each semantic space into the karst landform image analysis sub-model existing in the semantic space corresponding to the karst landform image analysis model, so as to determine the karst landform description knowledge chain based on the sub-landform description feature parameters generated by each karst landform image analysis sub-model; Obtain the karst landform description knowledge chain generated by the karst landform image analysis model, and cluster the karst landform image based on the karst landform description knowledge chain; The step of extracting the basic karst landform semantic features corresponding to the karst landform image specifically includes: Decompose the karst landform image to generate a plurality of target landform feature blocks; Extract the block semantic features corresponding to each target landform feature block, and determine the basic karst landform semantic features corresponding to the karst landform image based on the block semantic features corresponding to each target landform feature block; The step of determining the candidate karst landform semantic features corresponding to each semantic space of a pre-completed parameter learning karst landform image analysis model based on the basic karst landform semantic features includes: Perform dimensional normalization processing on the basic karst landform semantic features to generate target karst landform semantic features; According to the predefined semantic spaces of the karst landform image analysis model, determine the feature dimensions or feature combinations corresponding to each semantic space, and determine the corresponding relationship between each feature item in the target karst landform semantic features and the semantic space according to a pre-established feature mapping table; According to the feature mapping table, load each feature item in the target karst landform semantic features into the corresponding semantic space of the karst landform image analysis model, and generate the candidate karst landform semantic features corresponding to each semantic space.

2. The method for analyzing karst landform images based on deep learning according to claim 1, wherein, After the step of decomposing the karst landform image to generate a plurality of target landform feature blocks, the method further includes: Obtain a karst landform feature screening map sequence. When there is no landform feature block in the plurality of target landform feature blocks that matches the landform feature block in the karst landform feature screening map sequence, then screen out the karst landform image; When there is any one landform feature block in the plurality of target landform feature blocks that matches the landform feature block in the karst landform feature screening map sequence, then jump to the step of extracting the block semantic features corresponding to each target landform feature block.

3. The method for analyzing karst landform images based on deep learning according to claim 1, wherein The step of extracting the block semantic features corresponding to each target landform feature block and determining the basic karst landform semantic features corresponding to the karst landform image based on the block semantic features corresponding to each target landform feature block specifically includes: Perform unified processing of the image format, standardized processing of the image size, and normalization processing on each target landform feature block in sequence to generate a pre-processed target landform feature block; Adopt a multi-scale image analysis algorithm to construct a multi-scale image representation of the preprocessed target geomorphic feature block. Specifically, first determine the scale range to be used. For each scale in the scale range, obtain scale feature images at different scales through downsampling and upsampling algorithms. Among them, mean pooling is used for downsampling, and bilinear interpolation is used for upsampling. Then, combine the scale feature images at different scales to generate the multi-scale image representation; Use a local feature description algorithm. At each scale of the multi-scale image representation, detect local extreme points by constructing a Difference of Gaussian (DoG) pyramid. For each detected local extreme point, calculate the corresponding orientation histogram to describe the texture features around the local extreme point. The calculation of the orientation histogram is based on the pixel gray value changes in a preset range around the local extreme point. The pixels in the surrounding preset range are statistically analyzed according to a predefined orientation interval to obtain a histogram representing the orientation distribution; Combine the orientation histograms of all local extreme points to generate the local feature description of the target geomorphic feature block. The local feature description is used to reflect the local texture and structural features of the target geomorphic feature block; Adopt a sorting algorithm based on feature importance to sort the local features in the local feature description according to their importance in describing the target geomorphic feature block. After removing the local features with importance lower than the set importance, fuse the similar local features among the remaining local features through a feature fusion algorithm to generate an optimized local feature description. Among them, the evaluation of feature importance is determined based on the appearance frequency of local features in the entire local feature description and their correlation with other local features; Adopt a shape-based global feature extraction method to extract shape features from the multi-scale image representation. The shape features include contour features and regional shape features. Among them, the contour features are obtained by an edge detection algorithm to get the edge curve of the target geomorphic feature block, and the edge curve is described. The regional shape features are described by calculating the area, perimeter, aspect ratio parameters of the target geomorphic feature block; Adopt a color-based global feature extraction method to extract color features from the multi-scale image representation. The color features include color histograms and color moments. The color histogram is to statistically analyze the distribution frequencies of different colors in the target geomorphic feature block. The color moment is used to describe the color features by calculating the color mean, variance, skewness parameters of the target geomorphic feature block; Combine the shape features and the color features to generate the global feature description of the target geomorphic feature block. The global feature description is used to reflect the overall features of the target geomorphic feature block; Map the global feature description to a high-dimensional feature space, and convert the low-dimensional global feature description into a high-dimensional feature representation through a non-linear mapping function; Based on the relative importance of global features and local features in describing the target geomorphic feature blocks, a feature fusion algorithm is used to fuse the global feature description and the local feature description to obtain the fused block semantic features; Integrate the fused block semantic features of all target geomorphic feature blocks to generate the basic karst geomorphic semantic features corresponding to the karst geomorphic image.

4. The method for analyzing karst landform images based on deep learning according to claim 1, wherein Before the step of loading the candidate karst geomorphic semantic features corresponding to each semantic space into the semantic space corresponding to the karst geomorphic image analysis model, the method further includes: Obtain a sample karst geomorphic image, and extract the basic sample karst geomorphic semantic features corresponding to the sample karst geomorphic image; Based on the basic sample karst geomorphic semantic features, determine the input sample karst geomorphic semantic features corresponding to each semantic space of the karst geomorphic image analysis model; Obtain the sample karst geomorphic description knowledge chain corresponding to the sample karst geomorphic image; Load the input sample karst geomorphic semantic features corresponding to each semantic space into the semantic space corresponding to the karst geomorphic image analysis model, determine the sample karst geomorphic description knowledge chain as the output result expected to be learned by the karst geomorphic image analysis model, perform parameter learning on the karst geomorphic image analysis model to generate a target karst geomorphic image analysis model, and the target karst geomorphic image analysis model includes multiple karst geomorphic image analysis sub-models, and the neuron weight parameters in each karst geomorphic image analysis sub-model are different.

5. The method for analyzing karst landform images based on deep learning according to claim 1, wherein, The karst geomorphic image analysis sub-models existing in the step of loading the candidate karst geomorphic semantic features corresponding to each semantic space into the semantic space corresponding to the karst geomorphic image analysis model, in order to determine the karst geomorphic description knowledge chain based on the sub-geomorphic description feature parameters generated by each karst geomorphic image analysis sub-model, specifically includes: Identify different semantic spaces in the karst geomorphic image analysis model and the karst geomorphic image analysis sub-models corresponding to the semantic spaces. For each semantic space, determine the position and functional features of the semantic space in the overall architecture of the karst geomorphic image analysis model, and establish a mapping relationship table. Each semantic space in the mapping relationship table is used as an entry, and the corresponding karst geomorphic image analysis sub-model is used as the associated content; For each semantic space, load the candidate karst geomorphic semantic features corresponding to the semantic space into the corresponding karst geomorphic image analysis sub-model batch by batch or at one time according to the input interface specification of the corresponding karst geomorphic image analysis sub-model. Each karst geomorphic image analysis sub-model processes the loaded candidate karst geomorphic semantic features to generate sub-geomorphic description feature parameters corresponding to the input candidate karst geomorphic semantic features, and the sub-geomorphic description feature parameters reflect the feature description of the karst geomorphic image analysis sub-model for the karst geomorphology in different semantic spaces; Arrange each of the sub-geomorphic description feature parameters in the order of the semantic space or the predefined logical order of the karst landform image analysis sub-model, analyze the potential relationships between the sub-geomorphic description feature parameters to identify the existing target correlation points, and establish a target correlation structure based on the target correlation points; Determine the starting point of the karst landform description knowledge chain, where the starting point is the sub-geomorphic description feature parameter related to the most significant feature of the karst landform; According to the pre-determined construction rules, gradually connect other sub-geomorphic description feature parameters to the karst landform description knowledge chain based on the target correlation structure until all sub-geomorphic description feature parameters are incorporated into the karst landform description knowledge chain to generate the final karst landform description knowledge chain.

6. The method for analyzing karst landform images based on deep learning according to claim 5, wherein The step of analyzing the potential relationships between the sub-geomorphic description feature parameters to identify the existing target correlation points and establishing a target correlation structure based on the target correlation points includes: Construct a feature identification label for each sub-geomorphic description feature parameter, where the feature identification label is used to describe the feature of the karst landform represented by the sub-geomorphic description feature parameter; Through the pre-defined karst landform semantic knowledge base, map the feature identification label to a further semantic description, and attach the implemented semantic description of the mapping to the corresponding sub-geomorphic description feature parameter; Calculate the semantic similarity between each sub-geomorphic description feature parameter based on the basic semantic distance calculation strategy, and group the sub-geomorphic description feature parameters according to the semantic similarity to obtain an initial grouping sequence. Each initial grouping in the initial grouping sequence reflects the potential relationship based on semantics between the sub-geomorphic description feature parameters. The sub-geomorphic description feature parameters within the same initial grouping are relevant at the partial feature level of the karst landform; Optimize the initial grouping sequence to obtain an optimized target grouping sequence. Specifically, check the number of sub-geomorphic description feature parameters within each initial grouping. If the number of sub-geomorphic description feature parameters within any one initial grouping is greater than the first set number, then re-use the advanced semantic distance calculation strategy to calculate the semantic similarity between the sub-geomorphic description feature parameters within this initial grouping to further subdivide this initial grouping. If the number of sub-geomorphic description feature parameters within any one initial grouping is less than the second set number, then merge this initial grouping with the adjacent grouping and re-evaluate the semantic similarity of the sub-geomorphic description feature parameters within the merged grouping. If it meets the preset semantic similarity requirements, then retain the merged grouping, otherwise maintain the original grouping state, where the second set number is less than the first set number; For each packet in the optimized target packet sequence, analyze the relative positions of the geomorphic features corresponding to the intra-group sub-geomorphic description feature parameters in the karst geomorphic image. If the geomorphic features corresponding to any two or more sub-geomorphic description feature parameters are adjacent or have an inclusion relationship in the karst geomorphic image, after marking the any two or more sub-geomorphic description feature parameters as sub-geomorphic description feature parameters with spatial association, add the corresponding spatial association information to the corresponding packet to generate a packet combined with spatial relationships; Analyze the feature dependence relationships between the sub-geomorphic description feature parameters within each packet combined with spatial relationships. Within each packet, if the geomorphic feature corresponding to a sub-geomorphic description feature parameter is a dependence condition for the geomorphic feature corresponding to another sub-geomorphic description feature parameter, mark the corresponding sub-geomorphic description feature parameters as having a feature dependence relationship, and add the corresponding feature dependence relationship information to the packet to generate a packet combined with feature dependence relationships; In each packet combined with feature dependence relationships, search for the sub-geomorphic description feature parameter with the most association relationships, and determine the sub-geomorphic description feature parameter with the most association relationships as the target association point, where the association relationships include semantic similarity association, spatial relationship association, and feature dependence relationship association; Take each of the target association points as a node in the target association structure, and for each target association point, connect the relevant sub-geomorphic description feature parameters to the target association point according to the association relationships between the target association point and other sub-geomorphic description feature parameters to construct the target association structure.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store programs, instructions, or codes, and when the programs, instructions, or codes are executed by a processor, to implement the deep learning-based karst geomorphic image analysis method described in any one of claims 1-6 above.

8. A karst landform image analysis system based on deep learning, characterized in that, The deep learning-based karst geomorphic image analysis system includes a processor and a memory. The memory is connected to the processor. The memory is used to store programs, instructions, or codes, and the processor is used to execute the programs, instructions, or codes in the memory to implement the deep learning-based karst geomorphic image analysis method described in any one of claims 1-6 above.

Images