An image column-level feature analysis method and device, electronic equipment and medium

Abstract

The application provides an image column-level feature analysis method and device, electronic equipment and medium. The image column-level feature analysis method comprises the following steps: acquiring a B-scan image to be analyzed; dividing the B-scan image to be analyzed into multiple column image blocks according to a preset width; extracting column-level features of each of the image blocks based on a pre-trained Segmenter model; and analyzing the column-level features to identify underground structure change information corresponding to the B-scan image to be analyzed. In this application, each column of information is regarded as an independent image block, and feature analysis is realized under the guidance of self-attention. This method can more comprehensively understand the changes and features of underground structures. Compared with traditional methods, the application can provide more accurate, real-time and adaptive underground radar image analysis results, bringing new technical breakthroughs to the field of underground detection.

Description

Technical Field

[0001] This application relates to the field of image processing, and more specifically, to a method, apparatus, electronic device, and medium for image column-level feature analysis. Background Technology

[0002] In existing technologies, processing underground radar (GPR) B-scan images often employs traditional image processing techniques or deep learning methods. Some methods use models such as convolutional neural networks (CNNs) to extract and analyze features from the entire image. Other methods may employ rule-based feature engineering to extract and classify specific information within the image.

[0003] However, these methods often struggle to effectively process the independent information of each column in a GPR B-scan image, resulting in the inability to capture the correlation and features between columns in certain situations.

[0004] Therefore, how to solve the above problems is an urgent issue that needs to be addressed. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and medium for image column-level feature analysis, aiming to improve the above-mentioned problems.

[0006] Firstly, this application provides an image column-level feature analysis method, which includes:

[0007] Obtain the B-scan image to be analyzed;

[0008] The B-scan image to be analyzed is divided into multiple image blocks according to a preset width;

[0009] The column-level features of each image block are extracted based on the pre-trained Segmenter model;

[0010] The column-level features are analyzed to identify the underground structural change information corresponding to the B-scan image to be analyzed.

[0011] In one possible embodiment, dividing the B-scan image to be analyzed into multiple image blocks according to a preset width includes:

[0012] The B-scan image to be analyzed is vertically divided into multiple columns of image blocks with a width of 1 pixel. The height of each column of image blocks is equal to the height of the B-scan image to be analyzed.

[0013] In one possible embodiment, the pre-trained Segmenter model includes: an image patch unit, a position encoding unit, and a multi-head self-attention mechanism unit, wherein the image patch unit, the position encoding unit, and the multi-head self-attention mechanism unit are connected in sequence;

[0014] The image block unit is used to input at least one column of the image blocks;

[0015] The position encoding unit is used to perform position encoding on the image block input by the image block unit;

[0016] The multi-head self-attention mechanism unit is used to perform column-level feature extraction on the position-encoded image blocks to capture the column-level information and inter-column relationships of the image blocks.

[0017] In one possible embodiment, the analysis of the column-level features to identify the subsurface structural change information corresponding to the B-scan image to be analyzed includes:

[0018] Obtain all the column-level features and multiple representative target column-level features;

[0019] Multiple target column-level features are input into a classifier to obtain multiple column-level classification results; wherein each column-level classification result is the classification result of each pixel in the image;

[0020] Based on the column-level classification results, the subsurface structural change information corresponding to the B-scan image to be analyzed is identified.

[0021] In one possible embodiment, obtaining all the column-level features involves multiple representative target column-level features, including:

[0022] The average value of all the column-level features is taken to obtain multiple representative target column-level features.

[0023] In one possible embodiment, obtaining all the column-level features involves multiple representative target column-level features, including:

[0024] Global pooling is performed on all the column-level features to obtain multiple representative target column-level features.

[0025] In one possible embodiment, identifying the subsurface structure change information corresponding to the B-scan image to be analyzed based on the column-level classification results includes:

[0026] The column-level classification results are stitched together to obtain a column-processed image;

[0027] The subsurface structural change information corresponding to the B-scan image to be analyzed is obtained by identifying the column processing image.

[0028] Secondly, this application also provides an image column-level feature analysis device, the device comprising:

[0029] The acquisition module is used to acquire the B-scan image to be analyzed.

[0030] The image processing module is used to divide the B-scan image to be analyzed into multiple image blocks according to a preset width;

[0031] The feature extraction module is used to extract column-level features of each image block based on the pre-trained Segmenter model;

[0032] The analysis module is used to analyze the column-level features and identify the underground structural change information corresponding to the B-scan image to be analyzed.

[0033] Thirdly, this application also provides an electronic device, comprising:

[0034] Memory, used to store executable instructions;

[0035] The processor, when executing executable instructions stored in the memory, implements the image column-level feature analysis method as described in any of the first aspects.

[0036] Fourthly, this application also provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program, when run by a processing device, performs the steps of the image column-level feature analysis method as described in any one of the first aspects.

[0037] Beneficial effects:

[0038] The above-described image column-level feature analysis method, apparatus, electronic device, and medium provided in this application acquire a B-scan image to be analyzed; divide the B-scan image to be analyzed into multiple image blocks according to a preset width; extract column-level features of each image block based on a pre-trained Segmenter model; analyze the column-level features to identify the underground structural change information corresponding to the B-scan image to be analyzed, thereby realizing the analysis and extraction of column-level features of underground radar (GPR) B-scan images, thus more accurately capturing structural changes and feature information in the image. This application can provide more accurate, real-time, and adaptive underground radar image analysis results, bringing a new technological breakthrough to the field of underground detection. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the structure of an electronic device provided in the first embodiment of this application;

[0041] Figure 2 A flowchart illustrating an image column-level feature analysis method provided in the second embodiment of this application;

[0042] Figure 3 A schematic diagram comparing the image column-level feature analysis method provided in this application with the prior art of image segmentation;

[0043] Figure 4 This is a schematic diagram of the image stitching and fusion process performed using the image column-level feature analysis method provided in this application.

[0044] Figure 5 This is a schematic diagram of the functional modules of an image column-level feature analysis device provided in the third embodiment of this application. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0046] First Embodiment

[0047] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. In this application, it can be... Figure 1 The schematic diagram shown illustrates an example electronic device 100 for implementing the image column-level feature analysis method of the embodiments of this application.

[0048] like Figure 1 The diagram shows the structure of an electronic device 100. The electronic device 100 includes one or more processors 102, one or more storage devices 104, input devices 106, and output devices 108. These components are interconnected via a bus system and / or other forms of connection mechanisms (not shown). It should be noted that... Figure 1The components and structure of the electronic device 100 shown are merely exemplary and not limiting; the electronic device may have, as needed. Figure 1 The components shown may also have Figure 1 Other components and structures not shown.

[0049] The processor 102 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 100 to perform desired functions.

[0050] It should be understood that the processor 102 in the embodiments of this application may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0051] The storage device 104 may include one or more computer program products, which may include various forms of computer-readable storage media.

[0052] It should be understood that the storage device 104 in the embodiments of this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0053] The computer-readable storage medium may store one or more computer program instructions, which the processor 102 may execute to implement the client functions (implemented by the processor) in the embodiments of this application described below, and / or other desired functions. Various applications and various data may also be stored in the computer-readable storage medium, such as various data used and / or generated by the applications.

[0054] The input device 106 may be a device used by a user to input commands, and may include one or more of the following: keyboard, mouse, microphone, and touch screen.

[0055] Second embodiment:

[0056] Reference Figure 2 The flowchart shown illustrates an image column-level feature analysis method, which specifically includes the following steps:

[0057] Step S201: Obtain the B-scan image to be analyzed.

[0058] Among them, the B-scan image is a ground-penetrating radar (GPR) B-scan image, which can be an image acquired in real time.

[0059] Of course, in practical use, the B-scan image can also be a non-real-time image; no specific limitation is made here.

[0060] Step S202: Divide the B-scan image to be analyzed into multiple image blocks according to a preset width.

[0061] Optionally, the preset width is 1 pixel, where 1 pixel refers to one pixel of the B-scan image to be analyzed.

[0062] It should be understood that this application preferably uses a width of 1 pixel, which can improve the accuracy of feature extraction.

[0063] Of course, in other application scenarios, the preset width can also be 2 pixels or more pixels.

[0064] It should be noted that the height of each image patch is equal to the height of the B-scan image to be analyzed. That is, the image patches are directly cut from the B-scan image to be analyzed column by column, without scaling or cropping, to ensure data reliability.

[0065] like Figure 3 As shown, this application preserves the information characteristics of columns by using a column-level segmentation method compared to the traditional segmentation method, which facilitates the calculation of the temporal relationship between columns.

[0066] Step S203: Extract column-level features for each image block based on the pre-trained Segmenter model.

[0067] The Segmenter model adopts the VIT (Vision Transformer) architecture.

[0068] The pre-trained Segmenter model includes: an image patch unit, a position encoding unit, and a multi-head self-attention mechanism unit, which are connected sequentially. The image patch unit is used to input at least one column of image patches. The position encoding unit is used to encode the position of the image patches input by the image patch unit. The multi-head self-attention mechanism unit is used to perform column-level feature extraction on the position-encoded image patches to capture the column-level information and inter-column relationships of the image patches.

[0069] Specifically:

[0070] (1) Image Patch Unit: To adapt to the input of column-level image features, the input layer of the model is pre-adjusted to accept image patches of appropriate size (one or more columns). In other words, the image patch unit processing method treats one or more columns of the image as individual image patch units as input to the model. Specifically, the functions of the image patch unit are as follows:

[0071] 1. Adapting to Image Column-Level Features: Each column of a GPR B-scan image represents radar data acquired at different times or locations. Treating each column as an independent image patch better reflects the changes in subsurface structures at different locations. This helps the model analyze the characteristics of each column more accurately.

[0072] 2. Independence processing: Each column of information is treated as an image block, making the information in each column independent. This avoids information mixing between columns to some extent and better preserves the underlying information of each column.

[0073] 3. Spatial Location Encoding: Image patch units also allow for location encoding to help the model understand the relative position of the image patch within the overall image. This is particularly useful when dealing with column-level subsurface structures, as columns at different locations may have different geological features.

[0074] To adapt to the input of column-level image features, the specific adjustment method for image block units is as follows:

[0075] 1. Image Patch Size: Adjust the model's input layer to accept image patches as input. For each image patch, its height is consistent with the height of the entire image, while the width is typically set to 1 pixel, since each column of information contains only one data point. In this way, each column of ground information is treated as a separate image patch input into the model.

[0076] 2. Number of Channels: The height of each image patch is used as the number of input channels. Typically, the height of an image patch is the same as the height of the entire image. For example, if the height of a GPR B-scan image is H pixels, then the height of each image patch is also H, meaning each image patch will have H channels.

[0077] 3. Positional Encoding: To help the model understand the positional relationships of image patches within the overall image, positional encoding information can be combined with the representation of each image patch. This can be achieved through encoding methods such as embedding fixed positional information or using sine / cosine function encoding. Positional encoding helps the model consider the position of each subsurface information in the image when understanding the subsurface information.

[0078] The main difference between the adjusted model's input layer and the original model is that the adjusted model can process individual image patches, meaning that the subsurface information of each column is treated as an independent input. This allows the model to focus more intently on analyzing the features of each column while maintaining the independence of column-level information. This adjustment makes the model more adaptable to the characteristics of GPR B-scan images and better able to handle column-level feature analysis.

[0079] (2) Position Encoding Unit: Image column-level features require position encoding to help the model understand the spatial location of the input. That is, position encoding is performed on each column of image blocks. In the Segmenter-based GPR B-scan image column-level feature analysis method, each column of image blocks is treated as an independent input. To help the model understand the relative positions of these column of image blocks in the overall image, position encoding can be introduced.

[0080] (3) Multi-head self-attention mechanism unit: ensures that the model can perform self-attention feature extraction on each image column-level feature, capturing column-level information and inter-column correlation.

[0081] For example, suppose we have a GPR B-scan image where each column represents subsurface structural information at different depths. Each column may contain different geological features, such as rock layers, cavities, etc. By inputting each column of information as an image patch into the model and utilizing a self-attention mechanism, the model can achieve the following:

[0082] 1. Capturing Column-Level Information: For each image patch (each column), the model can analyze the features of subsurface information in each column through self-attention computation. This helps the model identify the characteristics of subsurface structures in each column, such as changes in geological features at different depths.

[0083] 2. Capturing Inter-column Relationships: Through a self-attention mechanism, the model can focus on the relationships between image patches in different columns. For example, the model can discover that certain features of underground structures are similar across different columns, or that certain underground features are correlated in adjacent columns. This helps the model better understand the spatial distribution of the entire underground structure.

[0084] Specifically, suppose our GPR B-scan image has three columns, representing subsurface information at different depths. In the model's self-attention mechanism, the model can calculate attention weights between each image patch to capture the correlation between columns. If the model finds that the subsurface information between different columns is similar within a specific depth range, it can focus more attention on the image patches in those columns, thereby capturing the features of the subsurface structure.

[0085] Understandably, in each self-attention layer of the column-level feature Segmenter model, the model is ensured to perform self-attention computation on each image column-level feature. This application allows the model to capture the relationships between image column-level features (i.e., the relationships between image patches in different columns) while maintaining the independence of information in each column. This approach enables the model to focus on both the relationships between columns and the information within each column simultaneously.

[0086] Step S204: Analyze the column-level features to identify the underground structural change information corresponding to the B-scan image to be analyzed.

[0087] Among them, the information on changes in underground structures includes changes and characteristics of underground structures.

[0088] As one implementation, step S204 includes: acquiring multiple representative target column-level features in the process of all the column-level features; inputting the multiple target column-level features into a classifier to obtain multiple column-level classification results; wherein, each column-level classification result is the classification result of each pixel in the image; and identifying the underground structure change information corresponding to the B-scan image to be analyzed based on the column-level classification results.

[0089] It is understandable that by obtaining multiple representative target column-level features through the above implementation method, subsequent analysis can be carried out based on these representative target column-level features. This can reduce the amount of data analysis, accelerate analysis efficiency, and improve analysis accuracy.

[0090] Optionally, obtaining multiple representative target column-level features from all the column-level features includes: averaging all the column-level features to obtain multiple representative target column-level features.

[0091] Optionally, obtaining multiple representative target column-level features from all the column-level features includes: performing global pooling on all the column-level features to obtain multiple representative target column-level features.

[0092] It is understood that in this application, averaging or global pooling operations can be used to obtain representative features of each column.

[0093] Of course, in practical use, other methods can be used to obtain the representative characteristics of each column, which will not be specifically limited here.

[0094] Optionally, the step of identifying the subsurface structure change information corresponding to the B-scan image to be analyzed based on the column-level classification results includes: stitching the column-level classification results to obtain a column-processed image; and identifying the column-processed image to obtain the subsurface structure change information corresponding to the B-scan image to be analyzed.

[0095] like Figure 4 As shown, each color in the image represents a category. The classification result obtained in this method is to classify each pixel in the image (e.g., 0 represents concrete, 1 represents rebar, 2 represents defects, etc.). By concatenating all the column-level classification results, a result similar to the one above is formed. Finally, image analysis is performed to identify where the rebar is and where the defects are.

[0096] In summary, the image column-level feature analysis method provided in this embodiment treats each column of information as an independent image block and performs feature analysis under the guidance of self-attention, enabling a more comprehensive understanding of the changes and characteristics of underground structures. Compared with traditional methods, this invention provides more accurate, real-time, and adaptive underground radar image analysis results, bringing a new technological breakthrough to the field of underground detection. Furthermore, the column-level feature analysis in this application adopts a VIT (VisionTransformer) architecture, realizing independent feature analysis for each column of information, which is more adaptable to the characteristics of GPR B-scan images. Secondly, this invention can more accurately capture changes and feature information of underground structures, improving the accuracy of image analysis; finally, through independent feature extraction and a self-attention mechanism, this invention can improve the robustness and adaptability of the model, making it suitable for different underground environments.

[0097] Third embodiment:

[0098] See Figure 5 The image column-level feature analysis device 500 shown includes: an acquisition module 510, an image processing module 520, a feature extraction module 530, and an analysis module 540. Specifically, the functions of each module are as follows:

[0099] The acquisition module 510 is used to acquire the B-scan image to be analyzed;

[0100] Image processing module 520 is used to divide the B-scan image to be analyzed into multiple image blocks according to a preset width;

[0101] The feature extraction module 530 is used to extract column-level features of each image block based on a pre-trained Segmenter model;

[0102] The analysis module 540 is used to analyze the column-level features and identify the underground structural change information corresponding to the B-scan image to be analyzed.

[0103] It should be noted that the specific functions of the image column-level feature analysis device 500 are described in the method embodiment, and will not be repeated here.

[0104] Furthermore, this embodiment also provides a computer-readable storage medium storing a computer program, which, when run by a processing device, executes the steps of any of the image column-level feature analysis methods provided in Embodiment 2 above.

[0105] The computer program product of the image column-level feature analysis method provided in this application includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0106] It should be noted that the above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0107] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0108] In this application, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0109] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0110] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0111] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0112] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0113] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0115] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

Claims

1. A method for image column-level feature analysis, characterized in that, The method includes: Obtain the B-scan image to be analyzed; The B-scan image to be analyzed is divided into multiple image blocks according to a preset width. The division of the B-scan image to be analyzed into multiple image blocks according to the preset width includes: vertically dividing the B-scan image to be analyzed into multiple image blocks with a width of 1 pixel, wherein the height of each image block is equal to the height of the B-scan image to be analyzed. The pre-trained Segmenter model extracts column-level features for each image patch. The pre-trained Segmenter model includes: an image patch unit, a position encoding unit, and a multi-head self-attention mechanism unit, which are sequentially connected. The image patch unit is used to input at least one column of image patches. The position encoding unit is used to encode the position of the image patches input to the image patch unit. The multi-head self-attention mechanism unit is used to extract column-level features from the position-encoded image patches to capture column-level information and inter-column relationships within the image patches. The column-level features are analyzed to identify the subsurface structural change information corresponding to the B-scan image to be analyzed. This analysis includes: acquiring multiple representative target column-level features from all the column-level features; inputting the multiple target column-level features into a classifier to obtain multiple column-level classification results; wherein each column-level classification result is the classification result for each pixel in the image; concatenating the column-level classification results to obtain a column-processed image; and identifying the column-processed image to obtain the subsurface structural change information corresponding to the B-scan image to be analyzed.

2. The method according to claim 1, characterized in that, The step of obtaining multiple representative target column-level features from all the column-level features includes: The average value of all the column-level features is taken to obtain multiple representative target column-level features.

3. The method according to claim 1, characterized in that, The step of obtaining multiple representative target column-level features from all the column-level features includes: Global pooling is performed on all the column-level features to obtain multiple representative target column-level features.

4. An image column-level feature analysis device, characterized in that, The device includes: The acquisition module is used to acquire the B-scan image to be analyzed. The image processing module is used to divide the B-scan image to be analyzed into multiple columns of image blocks according to a preset width. Specifically, the image processing module is used to: vertically divide the B-scan image to be analyzed into multiple columns of image blocks according to a width of 1 pixel, wherein the height of each column of image blocks is equal to the height of the B-scan image to be analyzed. The feature extraction module is used to extract column-level features of each image patch based on a pre-trained Segmenter model. The pre-trained Segmenter model includes: an image patch unit, a position encoding unit, and a multi-head self-attention mechanism unit, which are sequentially connected. The image patch unit is used to input at least one column of image patches. The position encoding unit is used to encode the position of the image patches input by the image patch unit. The multi-head self-attention mechanism unit is used to extract column-level features from the position-encoded image patches to capture column-level information and inter-column relationships of the image patches. The analysis module is used to analyze the column-level features and identify the underground structure change information corresponding to the B-scan image to be analyzed; Specifically, the analysis module is used to: acquire multiple representative target column-level features from all the column-level features; input the multiple target column-level features into a classifier to obtain multiple column-level classification results; wherein each column-level classification result is the classification result of each pixel in the image; stitch the column-level classification results together to obtain a column-processed image; and identify the column-processed image to obtain the underground structure change information corresponding to the B-scan image to be analyzed.

5. An electronic device, characterized in that, include: Memory, used to store executable instructions; A processor, when executing executable instructions stored in the memory, implements the image column-level feature analysis method as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processing device, performs the steps of the image column-level feature analysis method as described in any one of claims 1-3.

Images