Shallow sea sediment classification method and system based on multi-scale and multi-direction feature fusion

Abstract

The invention relates to the technical field of marine geological survey, in particular to a shallow sea sediment classification method and system based on multi-scale and multi-direction feature fusion, and the method comprises the steps: obtaining the multi-source acoustic data of a shallow sea sediment, and carrying out the preprocessing of the obtained multi-source acoustic data, constructing a spatial correlation model based on the preprocessed multi-source acoustic data, performing submarine topographic feature extraction according to a water depth value of MBES data, performing multi-scale and multi-direction texture feature extraction by using an SSS image output by the spatial correlation model, performing feature optimization on the obtained topographic features and texture features, and taking a core feature set as an input to perform feature extraction on the obtained topographic features and texture features; according to the method, the echo intensity image is taken as a core, the problem of poor space consistency in traditional multi-source data registration is solved, classification errors caused by feature space dislocation are avoided, and the accuracy of the multi-source data registration is improved. And classification reliability is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of marine geological exploration, in particular to a shallow seabed classification method and system based on multi-scale and multi-direction feature fusion. BACKGROUND

[0002] The shallow island reef area is a key transitional zone between land and marine ecosystems, with high ecological sensitivity and complex topographic structure. The type, composition and spatial distribution information of the seabed are the core basic data for marine resource development, ecological environment protection, marine engineering design and other work. The current mainstream technology mainly relies on two types of acoustic equipment, multi-beam echo sounding system (MBES) and side scan sonar (SSS). The technical complementarity of the two provides the possibility for accurate classification of seabed.

[0003] However, the existing seabed classification technology based on MBES and SSS data still has obvious limitations and is difficult to adapt to the complex environment of shallow island reef areas. On the one hand, MBES is affected by the large beam footprint in shallow water areas, resulting in reduced spatial resolution and inability to provide detailed seabed features. SSS has high texture resolution, but the data positioning accuracy is easily affected by factors such as wind, waves, ship speed and towfish height changes, and the spatial consistency is difficult to guarantee. On the other hand, existing multi-source data fusion methods are mostly limited to simple spatial registration level and do not solve the fundamental differences between MBES and SSS in spatial resolution and acoustic directivity, resulting in insufficient accuracy and stability of the classification results. In recent years, some researchers have tried to explore the optimization path of multi-source acoustic data fusion, such as combining airborne LiDAR and multi-spectral image to assist MBES data classification, or fusing field video data to enhance the texture recognition ability of SSS. However, the former is limited in applicability in turbid shallow island reef areas and cannot overcome the interference of water quality on optical data. The latter fails to address the core issues of scale and directional feature differences between multi-source data, and relies on additional field sampling data, increasing the cost and operational complexity. At present, there is a need for a shallow seabed classification method and system based on multi-scale and multi-direction feature fusion. SUMMARY

[0004] In order to solve the problems of low classification accuracy and insufficient reliability in the classification of seabed in island reefs, the present application provides a shallow seabed classification method and system based on multi-scale and multi-direction feature fusion.

[0005] In the first aspect, the present application provides a shallow seabed classification method based on multi-scale and multi-direction feature fusion, which adopts the following technical scheme: The shallow seabed classification method based on multi-scale and multi-direction feature fusion comprises: Obtaining multi-source acoustic data of the shallow seabed and preprocessing the obtained multi-source acoustic data, including obtaining MBES data and SSS data; The spatial correlation model is constructed based on the pretreated multi-source acoustic data, including determining a key region by using an echo intensity image in the MBES data, and performing local matching and feature point matching optimization on the SSS data according to the key region; The seafloor topographic features are extracted according to the water depth values of the MBES data, including calculating three types of topographic features of slope, slope direction and roughness based on the MBES data to reflect the seafloor bottom characteristics; The texture features are extracted by using the SSS image output by the spatial correlation model, including extracting texture features and constructing a texture feature description vector according to multi-scale and multi-direction parameters by using a gray level co-occurrence matrix; The topographic features and the texture features obtained are optimized to obtain a core feature set, including calculating the weights of different features, sorting and filtering, and removing redundancies by using an improved Relief-F algorithm; The core feature set is taken as input to train a multi-class classifier and construct a multi-model collaborative decision mechanism, and a bottom classification result is output and the classification accuracy is evaluated by a quantitative index.

[0006] Further, the pretreatment of the obtained multi-source acoustic data includes converting the MBES data and the SSS data to the same projection coordinate system, taking the polygon area actually covered by the MBES data as a reference boundary, performing spatial containment cropping on the image of the SSS data according to the reference boundary, converting the water depth values in all MBES data to a unified vertical reference, and performing geometric correction on the water depth values of the SSS data based on the unified vertical reference, and the geometric correction expression is: ; wherein, is the water depth value of the corrected SSS data, is the original water depth value of the SSS data, represents the height of the towed body, is the slant range of the sonar signal, is the horizontal distance component.

[0007] Further, the key region is determined by using the echo intensity image in the MBES data, including extracting the pixel gradient information of the echo intensity image in the MBES data by using a Canny edge detection operator, setting an edge significant region according to the edge result extracted by the Canny edge detection operator, calculating the gray variance in the local window after extracting the pixel gradient information of the image, setting an empirical threshold and determining a high variance region according to the gray variance, and performing intersection operation on the high variance region and the edge significant region to generate an anchor region mask.

[0008] Further, the local matching of the SSS data according to the key area and the feature point matching optimization comprises defining the MBES key area based on an anchor area mask, taking the center coordinates of the MBES key area as the origin of the SSS data image, obtaining a plurality of candidate matching windows in a preset neighborhood by using a sliding window method, calculating the matching degree of each candidate matching window and the MBES key area template by a normalized cross-correlation coefficient, and determining an initial corresponding area pair according to the cross-correlation coefficient, and the matching degree calculation formula is: ; wherein A is the MBES key area template, is the gray value of the template at the coordinates (x, y), is the gray mean value of the template, is the candidate matching window in the SSS image, is the gray value of the window at the coordinates (x, y), is the gray mean value of the window.

[0009] Further, the local matching of the SSS data according to the key area and the feature point matching optimization further comprises extracting feature points in the initial corresponding area pair by using a SIFT algorithm, performing nearest neighbor matching on the extracted feature points based on the Euclidean distance, eliminating false matching points by a random consistency sampling algorithm to obtain a high-confidence matching point pair set, fitting the coordinate mapping relationship of the MBES data and the SSS data in the x direction and the y direction by a cubic spline function respectively, taking the high-confidence matching point pair set as the control points, solving the fitting coefficients by a least square method, and obtaining a spatial correlation model by combining the fitting coefficients and the cubic spline function.

[0010] Further, the seabed topographic feature extraction according to the water depth value of the MBES data comprises calculating the elevation change rate of a target pixel in the x-axis and y-axis directions by using a difference algorithm based on the water depth value of the MBES data, converting the elevation change rate into an angle value by an arctangent function to obtain a slope, calculating an initial slope value by the arctangent function according to the elevation change rate, and adjusting the initial slope value by the positive and negative of the x-axis direction elevation change rate and the positive and negative of the y-axis direction elevation change rate, finally fitting the water depth data of the seabed area into a three-dimensional surface according to the seabed area covered by the MBES data, calculating the ratio of the actual area of the three-dimensional surface to the projection area of the seabed area, and obtaining the roughness.

[0011] Furthermore, the extraction of texture features from the SSS image output by the spatial association model includes presetting multiple sliding windows of different sizes and multiple main directions of different angles as multi-scale parameters and multi-directional parameters, respectively. Gray-level compression processing is performed on the position-calibrated SSS image. For the combination of multi-scale and multi-directional parameters, the occurrence frequency of different gray-level combinations is counted within the corresponding sliding window according to the preset pixel spacing, and the occurrence frequency is normalized to obtain a probability value. A gray-level co-occurrence matrix is ​​constructed based on the probability value. Multiple texture features are extracted based on each gray-level co-occurrence matrix, and the multiple texture features are sorted to obtain a texture feature description vector.

[0012] Furthermore, the feature optimization of the acquired terrain and texture features includes introducing class prior probability and distance normalization mechanisms, calculating feature weights for terrain and texture features, calculating the nearest neighbors of all features of the same class and the nearest neighbors of different classes using Euclidean distance, calculating the feature difference between each feature based on the nearest neighbors of the same class and the nearest neighbors of different classes, iteratively updating the feature weights according to the feature differences, retaining high-contribution features by combining the weight threshold determined by cross-validation, and removing redundant features within the high-contribution features using the Pearson correlation coefficient. The feature difference formula is: ; in, For the feature samples to be processed, For category The prior probability, For category In and sample The j-th nearest sample, For the sample The true substrate category, The maximum value among all samples. The minimum value among all samples. For the sample Prior probability of substrate type.

[0013] Furthermore, the training of multi-class classifiers and the construction of a multi-model collaborative decision-making mechanism include selecting random forest, K-nearest neighbors, support vector machine, random undersampling enhancement algorithm, and wide neural network as multi-class classifiers, training each classifier separately using a 5-fold cross-validation mechanism, calculating the weight of each classifier using a normalization method based on the overall classification accuracy of each classifier on the validation fold of the 5-fold cross-validation, performing a weighted summation based on the weights of each classifier, and taking the class with the highest score after weighted summation as the final background classification result for the sample. The classification result expression is: ; in, For the number of classifiers, weights of the first classifier, weights of the second classifier, predicted probabilities of the classes.

[0014] In a second aspect, a shallow seabed classification system based on multi-scale and multi-directional feature fusion comprises: A data acquisition module configured to acquire multi-source acoustic data of a shallow seabed and pre-process the acquired multi-source acoustic data, including acquiring MBES data and SSS data; A correlation model module configured to construct a spatial correlation model based on the pre-processed multi-source acoustic data, including determining a key area using an echo intensity image in the MBES data, and performing local matching and feature point matching optimization on the SSS data according to the key area; A topographic feature module configured to extract seabed topographic features according to the water depth values of the MBES data, including calculating three types of topographic features, i.e., slope, slope direction and roughness, based on the MBES data to reflect seabed bottom characteristics; A texture feature module configured to extract texture features using the SSS image output by the spatial correlation model, including using a gray level co-occurrence matrix to extract texture features and construct a texture feature description vector according to multi-scale and multi-directional parameters; An optimization module configured to optimize the extracted topographic features and texture features to obtain a core feature set, including using an improved Relief-F algorithm to calculate weights, sort and filter, and remove redundancies for different features; An output module configured to take the core feature set as input, train a multi-class classifier, and construct a multi-model collaborative decision mechanism to output a bottom classification result and evaluate the classification accuracy through quantitative indicators.

[0015] In summary, the present application has the following beneficial technical effects: 1. The present application effectively eliminates spatial misplacement and elevation deviation of multi-beam echo sounding system (MBES) and SSS data caused by differences in acquisition principles through coordinate system, regional boundary and elevation reference unified processing of multi-source acoustic data, combined with exclusive geometric correction of side scan sonar (SSS) data, ensuring high consistency of the two types of data in spatial position, coverage range and elevation reference, avoiding subsequent feature mis-extraction caused by data deviation, providing high-quality and highly consistent basic data for subsequent classification links, and ensuring classification accuracy from the data source 2、The present application takes MBES echo intensity image as the core, generates an anchor area with stable features through edge enhancement, local variance screening and regional intersection, and then combines local matching, feature point extraction and false matching elimination to finally construct a nonlinear spatial correlation model, realize accurate spatial mapping of MBES and SSS data, solve the problem of poor spatial consistency in traditional multi-source data registration, ensure accurate correspondence of terrain features and texture features at the pixel level, provide reliable geometric guarantee for multi-source feature fusion, avoid classification errors caused by feature spatial misplacement, and improve classification reliability.

[0016] 3、The present application extracts key features reflecting the characteristics of seabed topography based on MBES data, and extracts multi-scale and multi-directional texture features based on SSS data, constructs a double-dimensional feature system of terrain and texture, comprehensively covers the terrain fluctuation characteristics and texture detail features of the bottom, effectively makes up for the short board of traditional single feature in distinguishing similar terrain but different texture or similar texture but different terrain, significantly improves the separability of different bottom classes, and provides sufficient feature support for high-precision classification.

[0017] 4、The present application adopts an improved feature optimization algorithm, introduces class prior probability and distance normalization mechanism, accurately calculates feature contribution weight, screens high-value features and eliminates redundant information, reduces the interference of redundant features with no distinguishing power on the classification model, reduces the risk of model overfitting, ensures that the classification model focuses on the core features most critical to distinguishing the bottom, improves the classification stability and generalization ability of the model, and further ensures the reliability of the classification result.

[0018] 5、The present application constructs a multi-classifier collaborative decision mechanism, combines the advantages of different classifiers in high-dimensional features, unbalanced samples and other scenes, and fuses the output results of each classifier through weighted voting, which avoids the performance limitations of traditional single classifier in complex shallow reef environment, effectively deals with the problems of mixed bottom and unbalanced samples, improves the recognition ability of rare bottom types, and ensures that high-precision and high-reliability bottom classification results can be output in different sea environments. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 It is the overall flowchart of the shallow seabed classification method based on multi-scale and multi-directional feature fusion of the embodiment of the present application.

[0020] Figure 2 It is the effect picture of each scale in the shallow seabed classification method based on multi-scale and multi-directional feature fusion of the embodiment of the present application.

[0021] Figure 3 It is the RF classifier effect picture in the shallow seabed classification method based on multi-scale and multi-directional feature fusion of the embodiment of the present application.

[0022] Figure 4 is a RUSBoost effect diagram in the shallow seabed classification method based on multi-scale and multi-direction feature fusion of an embodiment of the present application. DETAILED DESCRIPTION

[0023] The present application will be further described in detail below with reference to the accompanying drawings.

[0024] Embodiment 1 Referring to Figure 1 The shallow seabed classification method based on multi-scale and multi-direction feature fusion of the present embodiment comprises: Obtaining multi-source acoustic data of the shallow seabed and pre-processing the obtained multi-source acoustic data, including obtaining MBES data and SSS data; Constructing a spatial correlation model based on the pre-processed multi-source acoustic data, including determining a key area using an echo intensity image in the MBES data, and performing local matching and feature point matching optimization on the SSS data according to the key area; Extracting seabed topographic features according to the water depth value of the MBES data, including calculating three types of topographic features, i.e., slope, slope direction and roughness, reflecting seabed bottom characteristics based on the MBES data; Extracting texture features using the SSS image output by the spatial correlation model, including extracting texture features and constructing a texture feature description vector according to multi-scale and multi-direction parameters using a gray level co-occurrence matrix; Optimizing the obtained topographic features and texture features to obtain a core feature set, including performing weight calculation, sorting and filtering, and redundancy elimination on different features using an improved Relief-F algorithm; Taking the core feature set as input, training a multi-class classifier and constructing a multi-model collaborative decision mechanism, outputting a bottom classification result and evaluating the classification accuracy through quantitative indicators.

[0025] Specifically, the shallow seabed classification method based on multi-scale and multi-direction feature fusion comprises the following steps: As shown in Figure 1 , Figure 2 S1, obtaining multi-source acoustic data of the shallow seabed and pre-processing the obtained multi-source acoustic data, including obtaining MBES data and SSS data; In the target shallow island reef survey area, a collaborative operation mode is adopted to lay out the multi-beam echo sounder (MBES) and side scan sonar (SSS) equipment, and synchronous data collection is carried out. The MBES data collection transmits multi-beam sonar signals to the sea bottom through the MBES equipment, and after receiving the sea bottom reflection signals, two types of core data are recorded synchronously. One is high-precision depth data, that is, the initial vertical distance from each sampling point on the sea bottom to the sea level. The other is echo intensity image data, that is, the gray scale image formed by quantizing the reflection intensity of different seabed materials to sonar signals, with a gray value range of 0-255, reflecting the difference in reflection ability of the seabed. In the collection process, the GPS and attitude sensor carried by the MBES synchronously record the original spatial coordinates and equipment attitude parameters corresponding to each set of water depth data and echo intensity data. The SSS data collection moves along the preset survey line by dragging the SSS body from the ship body, the SSS body transmits high-frequency sonar signals to the sea bottom on both sides, and after receiving the reflection signals, two types of core data are recorded. One is high-resolution sonar intensity image data, that is, the continuous gray scale image formed by the reflection signal intensity of the seabed. The other is the original water depth data, that is, the initial vertical distance of the seabed sampling point calculated by the round trip time of the sonar signal. In the collection process, the pressure sensor carried by the body records the height of the body , combined with the propagation speed of the sonar signal and the round trip time to calculate the slant range of the sonar signal , and the original spatial coordinates and track information are recorded synchronously.

[0026] All collected MBES data and SSS data need to be converted to WGS84-UTM projection coordinate system to ensure the consistency of the plane position reference standard. If the original data uses local coordinate system or other projection method, the conversion needs to be completed through a seven-parameter coordinate conversion model. The formula of the seven-parameter coordinate conversion model is: ; Among them, ( , , ) are the source coordinates of the original data, ( , , ) are the target coordinates in the unified coordinate system after conversion, is the scale factor for correcting the scale difference in different coordinate systems, is the rotation matrix to realize the alignment of coordinates in space angle, , , is the translation vector, which is the position offset of the whole coordinate in the plane and vertical direction. Through this model, the spatial misalignment problem caused by different coordinate references of different data sources can be eliminated.

[0027] Based on the coordinate conversion of the MBES water depth data sampling point coordinates, a polygon fitting algorithm (such as Delaunay triangulation) is used to generate the polygon area actually covered by the MBES data , the vertices of the polygon are the outermost coordinates of the MBES data sampling points, which ensure complete coverage of all valid MBES data, and then the polygon area actually covered by the MBES measurement is used as the reference boundary to perform spatial inclusion cropping on the SSS image. A spatial inclusion judgment algorithm is used, and in this embodiment, a ray method is used to judge the coordinate-converted SSS sonar intensity image pixel by pixel, only the overlapping area of the two kinds of data is retained, and invalid pixels outside the survey area are removed, to ensure that the data ranges are strictly consistent. The cropping formula is: ; , wherein is the original side-scan sonar image, is the cropped side-scan sonar image consistent with the MBES data range, is the reference boundary. Then all MBES water depth data is converted to a unified vertical reference, to ensure that the reference standards of the water depth values are consistent. Specifically, it includes two parts of MBES water depth value conversion and SSS water depth value geometric correction. Among them, the MBES water depth value vertical reference conversion calculates the instantaneous tide value at the time of MBES original water depth data collection according to the tidal harmonic constant provided by the tide observation station in the survey area, and subtracts the instantaneous tide value from the MBES original water depth value based on the instantaneous sea level, to obtain the MBES unified water depth value based on the theoretical lowest tide surface, to ensure that all MBES water depth data is consistent in the vertical reference. The SSS water depth value geometric correction is to eliminate the significant deviation of the SSS original water depth data due to the influence of the tow body height , i.e. the distance from the bottom of the tow body to the sea level, and the slant range of the sonar signal, to obtain the corrected SSS water depth value based on the theoretical lowest tide surface. For SSS data, the tow body height and the slant range will introduce geometric deviation, which needs to be corrected. The correction formula is: , wherein is the water depth value of the corrected SSS data, is the original water depth value of the SSS data, represents the tow body height, is the slant range of the sonar signal, is the horizontal distance component. Through this correction, the influence of the tow body height and the slant range on the accuracy of the SSS water depth data can be eliminated, so that the MBES and SSS data remain consistent in the elevation dimension.

[0028] S2, constructing a spatial correlation model based on the pre-processed multi-source acoustic data, including determining a key area by using an echo intensity image in the MBES data, and performing local matching and feature point matching optimization on the SSS data according to the key area; Taking the pre-processed MBES echo intensity image as the core, the seabed terrain boundary and the transition zone of different seabed materials in the MBES echo intensity image will present a gray value mutation, and the gradient information of the pixels needs to be extracted by a Canny edge detection operator. First, Gaussian filtering is performed on the MBES echo intensity image, and a 5*5 Gaussian kernel is used to smooth the image noise, such as the gray fluctuation caused by the bubble interference in shallow water. The expression of the Gaussian kernel function is: ; wherein, is the standard deviation of the Gaussian kernel, which is used to balance the denoising effect and edge preservation, is the relative coordinates of the pixels in the filter window, and the function is used to convolve the image window by window to eliminate the interference of high-frequency noise on edge extraction. For the filtered image, the Sobel operator is used to calculate the gray change rate in the x-axis horizontal direction and the y-axis vertical direction, respectively, and then the gradient strength of each pixel is calculated by the gradient amplitude formula , the expression of which is: ; wherein, is the gray change rate in the x-axis direction, is the gray change rate in the y-axis direction, and the greater the gradient strength, the more likely it is that the pixel is a terrain or seabed boundary, i.e., an edge region. Non-maximum suppression is performed on the obtained gradient strength image. First, the gradient direction of each pixel is determined. The gradient direction is the direction in which the gray change is most severe at the pixel, which is calculated by the gray change rates of the pixel in the x-axis and y-axis directions. Along the gradient direction of each pixel, two adjacent pixels are selected as comparison objects. If the gradient direction is 0°, the adjacent pixels to the left and right of the current pixel are compared. If the gradient direction is 45°, the adjacent pixels to the upper left and lower right of the current pixel are compared. The gradient strength of the current pixel is compared with the gradient strengths of the two comparison pixels. If the gradient strength of the current pixel is greater than the gradient strengths of the two comparison pixels, it is determined that the pixel is a local maximum value pixel in the gradient direction, and the gradient strength value is retained. If the gradient strength of the current pixel is less than or equal to the gradient strength of any one of the comparison pixels, it is determined that the pixel is a redundant pixel that is not in the edge core, and the gradient strength value is set to 0 to be removed.

[0029] Through statistical analysis of the actual sample gray distribution of the MBES echo intensity image in the survey area, the low threshold value is adaptively determined, and the high threshold value is determined according to , the ratio can ensure that the high threshold value can accurately lock the edge core, and the low threshold value can cover the extended part of the edge. Based on the double threshold value, the gradient intensity image after non-maximum suppression is classified pixel by pixel. If the gradient intensity of the pixel , it is determined as a strong edge pixel. Such pixels correspond to the core area of the seabed terrain or bottom boundary, the gray scale changes sharply and stably, and are directly retained. If the gradient intensity of the pixel , it is determined as a non-edge pixel. Such pixels are mostly in the internal gray scale flat area or noise interference area of the bottom, and the gradient intensity value is set to 0 to be removed. If the gradient intensity of the pixel satisfies , it is determined as a weak edge pixel. For all weak edge pixels, 8-neighbor connectivity judgment is used, that is, whether there is a strong edge pixel in the 8-neighbor pixels around the weak edge pixel. If there is at least one strong edge pixel, the weak edge pixel is determined as “an extended part of the strong edge”, and its gradient intensity value is retained. If there is no strong edge pixel, the weak edge pixel is determined as “an isolated false edge”, and its gradient intensity value is set to 0 to be removed. The retained strong edge pixels and the connected weak edge pixels jointly constitute the edge significant area, and the edge significant area determination function is: ; wherein, represents whether the pixel belongs to the significant edge area, is an edge significance threshold value. Through the determination mechanism, the anchor area has obvious edge intensity characteristics and sufficient texture change information, thereby enhancing the stability and representativeness of subsequent anchoring.

[0030] A 5x5 sliding window is used to traverse the edge-enhanced image, and the local variance of all pixel gray scale values in each window is calculated. An empirical variance threshold value is set through the statistics of the known bottom sample in the survey area. In this embodiment, 1.5 times the average variance of fine sand bottom is used, and all pixels corresponding to the window are retained to form a high-variance area. The edge significant area and the high-variance area are subjected to spatial intersection operation, only the pixels belonging to both the edge significant area and the high-variance area are retained, and the anchor area mask is generated. The expression of the anchor area mask is: ; wherein, is an edge significant area binary mask, is a high-variance area binary mask, This represents a pixel-wise logical AND operation. The set of pixels with a value of 1 in the mask is the MBES key region. Using the MBES key region as a template, in the preprocessed SSS sonar intensity image (with coordinate unification and region cropping completed), the optimal SSS matching region most similar to the template is found through a sliding window and normalized cross-correlation coefficient (NCC), forming an initial corresponding region pair, and anchoring the region mask. Extract the WGS84-UTM coordinates of all cells with a median value of 1. The spatial center coordinates of the key area were calculated using the mean method. The expression is: Where N is the total number of pixels in the key region of MBES. Since the SSS coordinates are consistent with the MBES coordinate reference after preprocessing, The search center is directly mapped to the SSS image. Combined with the maximum offset caused by the attitude fluctuation of the SSS tow body in shallow sea environment, a square search neighborhood of 30×30 pixels is set so that the search neighborhood can completely cover the possible offset range and ensure that the optimal matching area is not missed.

[0031] Then, the grayscale matrix of the key MBES region is extracted as the matching template A. Within the SSS search neighborhood, a sliding window of the same size as the template is moved pixel by pixel. Each move generates an SSS candidate matching window B. For each candidate window B, its similarity to the MBES template A is calculated using the cross-correlation coefficient (NCC) formula. The matching degree calculation formula is as follows: ; Where A is the MBES critical region template, This represents the grayscale value of the template at the coordinates (x, y). The average grayscale value of the template. For candidate matching windows in the SSS image, This represents the grayscale value of the window at coordinates (x, y). Given the mean grayscale value of the window, iterate through all candidate windows and retain the window with the largest NCC value. The SSS image region corresponding to this window is the optimal matching region of SSS. The MBES key region and the optimal matching region of SSS together constitute the initial corresponding region pair.

[0032] The next step is to perform feature point matching and optimization on the MBES key region and the SSS optimal matching region in the initial region pairing. First, feature points in the initial corresponding region pair are extracted using the SIFT algorithm. Gaussian difference pyramids are constructed for the MBES key region and the SSS optimal matching region respectively. Multi-scale images are generated by convolution with Gaussian kernels of different standard deviations. Then, the difference between adjacent scale images is calculated, and local extrema in the pyramid are detected as potential feature points to ensure that the feature points have scale invariance. For potential feature points, their sub-pixel positions are accurately determined by fitting a quadratic function. Thresholds for contrast and edge response are set to remove unstable points with low contrast and edge response. In this embodiment, the thresholds for contrast and edge response are 0.03 and 10, respectively. Then, based on the gradient orientation histogram in the neighborhood of the feature point, the direction corresponding to the peak of the histogram is taken as the main direction of the feature point. Taking the feature point as the center, a 16×16 neighborhood is taken and divided into 4×4 sub-blocks. Gradient histograms of 8 directions are calculated for each sub-block to generate a 128-dimensional feature descriptor. The similarity is measured based on the Euclidean distance of the feature descriptors. For each feature point descriptor in the MBES key region... All descriptors in the optimal matching region of SSS In the process, the descriptor with the smallest distance is selected as the matching candidate. The Euclidean distance formula is: ; in, , Represented as descriptors and The smaller the distance between the k-th component, the higher the descriptor similarity. The Random Consensus Sampling (RANSAC) algorithm is used to remove mismatched points. Four pairs of initial matching points are randomly selected, and a fundamental matrix is ​​fitted to describe the geometric relationship between the two views. An error threshold is set. The error threshold is set to 1-2 pixels, determined based on the image resolution, and all matching points are counted to satisfy the fundamental matrix constraints. For the interior points, repeat the above process, and retain the set of interior points corresponding to the basis matrix with the largest number of interior points. This set is the high-confidence matching point pair set. .

[0033] Matching point pairs with high confidence Using control points, a cubic spline function is used to fit the nonlinear mapping relationship from SSS pixel coordinates to MBES pixel coordinates, generating a spatial correlation model. The cubic spline function is selected as the nonlinear spatial transformation model, fitting the coordinate mappings in the x and y directions respectively. The model expression is: ; ; in, SSS pixel coordinates, (m, n) are mapped MBES pixel coordinates, t is a normalized spatial position parameter, mapping SSS pixel coordinates to the interval [0, 1] to eliminate the influence of coordinate range difference, 、 、 and indicate the fitting coefficients different in the x direction, 、 、 and indicate the fitting coefficients different in the y direction.

[0034] The high-confidence matching point pair set SSS pixel coordinates and corresponding MBES pixel coordinates , wherein, , M is the number of matching point pairs, and the above model is substituted to construct an overdetermined equation group; the least squares method is used to minimize the sum of squares of errors between actual mapping coordinates and model predicted coordinates, and the fitting coefficients in the x and y directions are solved to calculate the root mean square error (RMSE) of all matching point pairs, and the model mapping accuracy is verified, and the RMSE expression is: ; wherein, is the SSS pixel coordinates obtained by inversely mapping the MBES pixel coordinates Finally, the cubic spline functions satisfying the accuracy requirement 、 jointly constitute the spatial correlation model, which can accurately map any SSS pixel coordinates to MBES pixel coordinates, and realize pixel-level spatial correlation of the two types of data.

[0035] S3, seabed terrain feature extraction is performed according to the water depth value of the MBES data, including three types of terrain features of slope, slope direction and roughness reflecting seabed bottom characteristics based on MBES data calculation; Based on the preprocessed MBES high-precision water depth data, the feature extraction is performed, the original MBES water depth data is in the form of discrete sampling points, which is first converted into regular grid data and the abnormal values are eliminated, the inverse distance weighted interpolation algorithm is used to interpolate the discrete MBES water depth sampling points into regular grid with resolution of 1m×1m, and the gridded water depth data is obtained, wherein, is the row and column number of the grid cell, is the water depth value corresponding to the cell, the 3σ criterion is used to screen and eliminate the abnormal values in the gridded water depth data, and then the three types of terrain features of slope, slope direction and roughness are extracted.

[0036] The slope refers to the included angle between the seabed slope and the horizontal plane, reflecting the steepness of the terrain, and the calculation process is based on the x-axis and y-axis elevation rate of the target pixel, which is converted into an angle value by the arctangent function, and a 3x3 sliding window center difference method is used to calculate the elevation rate of each target pixel (x, y) in the x-axis and y-axis of the gridded bathymetric data , and the slope is calculated according to the elevation rate, and the slope calculation formula is: ; Wherein, ∂H / ∂x and ∂H / ∂y are the elevation rates in the horizontal and vertical directions, is the arctangent function, and the calculation result is in radians. In addition, the aspect is the azimuth corresponding to the direction of maximum slope (i.e. the direction of most steep terrain), reflecting the slope direction, such as 0° indicating north, 90° indicating east, and the range being 0°-360°. The calculation process is based on the positive and negative and ratio of the x-axis and y-axis elevation rates, and the initial aspect is calculated by the arctangent function and then adjusted. The initial aspect is calculated by the ratio of the x-axis and y-axis elevation rates, and the formula is as follows: ; According to the positive and negative of the x-axis and y-axis elevation rates, the initial aspect value is adjusted to the standard azimuth angle of 0°-360°, and the adjustment rules are as follows: 1. If , it indicates that the water depth increases along the positive direction of the y-axis, i.e. the slope faces south, and the final aspect is ; 2. If , it indicates that the water depth decreases along the positive direction of the y-axis, i.e. the slope faces north, and the final aspect is ; 3. If , it indicates that there is no elevation change along the y-axis direction, and when , it indicates that the water depth decreases along the positive direction of the x-axis, and the slope faces west ; When , it indicates that the water depth increases along the positive direction of the x-axis, and the slope faces east, ; 4. If , the target pixel is a flat area without inclination direction, and is set to .

[0037] Finally, roughness calculation is carried out, and roughness refers to the complexity of the ups and downs of the seafloor topography. The coarse-grained bottom material has a significantly larger roughness value than the fine-grained bottom material due to the uneven surface. By fitting the three-dimensional surface of the seafloor, the ratio of the actual area of the surface to the horizontal projection area is realized. For the gridded water depth data, the bicubic B-spline interpolation algorithm is used to fit the water depth data of the target seafloor area into a continuous three-dimensional surface , the expression is: ; Among them, is the B-spline fitting coefficient, which is solved by minimizing the error sum of squares of the fitting surface and the actual water depth data by the least square method. The projection of the target seafloor area on the horizontal plane is a regular rectangle, and the area is equal to the area of a single grid multiplied by the number of area grids, and the formula is as follows: ; Among them, K is the total number of grid pixels of the target area, is the area of a single grid, and based on the fitted three-dimensional surface , the actual area of the target area is calculated by the surface integral , and the discrete calculation formula is as follows: ; Among them, is the number of grid rows and columns of the target area, , is the partial derivative of the fitted surface in the x-axis and y-axis directions, and the roughness is defined as the ratio of the actual area of the three-dimensional surface to the horizontal projection area, that is: ; Among them, is the unit surface area in the area, is the projection area of the area on the horizontal plane. The calculated slope, slope direction and roughness are one-to-one corresponding to the pixels of the gridded water depth data, forming three types of terrain feature data.

[0038] S4, using the SSS image output by the spatial correlation model to extract texture features, including using a gray level co-occurrence matrix, extracting texture features according to multi-scale and multi-direction parameters, and constructing a texture feature description vector; Based on the position calibrated SSS image output by the spatial correlation model, the pixel-level spatial alignment with the MBES data is realized, and the spatial misalignment caused by device posture and environmental interference is eliminated. Although the position calibrated SSS image has realized spatial alignment, its original gray value range may fluctuate due to the strength of the sonar signal. Therefore, the gray scale is standardized, and the minimum value of the original gray value is obtained by traversing all the pixels of the position calibrated SSS image with the maximum value , the original gray value is mapped to the standard range of 0-255 by linear transformation, eliminating the influence of signal intensity fluctuation in different areas, and the standardization formula is as follows: ; wherein, is the original gray value of the pixel (x, y) in the SSS image after position calibration, is the normalized gray value, ensuring that all pixel gray values are on the same order of magnitude for subsequent processing, is the minimum value of the original gray value, is the maximum value of the original gray value.

[0039] Select 3x3, 5x5, 7x7 three sliding window sizes as multi-scale parameters, each scale corresponds to the response demand of substrate texture, the sliding window is traversed by pixel sliding, ensuring that each pixel can be analyzed under different scales, then select 0°, 45°, 90°, 135° four main directions as multi-direction parameters, fully covering the spatial orientation difference of substrate texture, each direction parameter acts independently on the sliding window, then performs gray level compression on the SSS image after position calibration, reduces the calculation complexity and preserves the key information of texture, combined with the demand of substrate texture classification in shallow island reef area, selects 16 levels as the compression target, divides the range of normalized gray value into 16 intervals according to equal interval, each interval corresponds to a compressed gray level , , the specific division formula is as follows: ; wherein, is the floor function, then construct the gray level co-occurrence matrix GLCM, which is a matrix describing the probability of the combination of the gray levels of two pixels with a certain distance and direction in the image, which can quantify the spatial distribution of texture, combined with the resolution of SSS image and the period of substrate texture, select 1 pixel interval as the statistical interval, traverse the image with the sliding window of the current scale, for each window, according to the set direction and pixel interval d=1, count the number of gray level combinations (i,j) of all target pixels and neighborhood pixels , wherein d represents the pixel interval, indicates the direction parameter, i represents the gray level of the target pixel, and j represents the gray level of the neighborhood pixel, divide the number of occurrences by the total number of all gray level combinations , get the probability of gray level combination ， The normalization formula is as follows: ; wherein, is the number of occurrences of the gray level combination (i,j) of the neighborhood pixels, is the total number of all gray level combinations, based on the probability , a gray level co-occurrence matrix with the dimension of compressed gray level number x compressed gray level number is constructed , the matrix element is Each combination of scale and direction corresponds to 1 GLCM matrix, a total of 3(scale) x 4(direction) = 12 GLCM matrices are generated, eight types of statistical features of mean, variance, homogeneity, contrast, correlation, angular second moment, entropy and dissimilarity are extracted from each GLCM matrix, wherein the homogeneity represents the similarity degree of the local region of the image, and the calculation formula is ; The angular second moment represents the regularity or uniformity of the texture, and the calculation formula is: ; The entropy reflects the information complexity of the texture, the more complex the texture (such as the shell sand mixed area), the greater the entropy value; the simpler the texture (such as pure fine sand), the smaller the entropy value, and the calculation formula is: ; The dissimilarity represents the absolute difference between the gray values, and the calculation formula is: ; The texture features extracted under all scale and direction combinations are integrated in a predetermined order, and for each ordered scale and direction combination, the eight types of statistical feature values of mean, variance, homogeneity, contrast, correlation, angular second moment, entropy and dissimilarity are spliced in turn to form a texture feature description vector .

[0040] S5, optimizing the terrain features and texture features obtained to obtain a core feature set, including using an improved Relief-F algorithm to calculate the weights of different features, sort and filter, and remove redundancies; Based on the extracted terrain feature and texture feature description vectors, the two types of features are spliced in the order of terrain features first and texture features last to form a multi-dimensional initial feature space , wherein the terrain features are represented as Since the terrain features and the texture features have significant dimensional differences, Z-score standardization is used to eliminate dimensional interference to obtain a standardized feature space Then, sampling points of known seabed types were obtained in the target shallow island and reef survey area. Each sampling point corresponds to a set of coordinates and a seabed type label. Based on the sampling point coordinates (X,Y), the data was analyzed in the standardized feature space. In the process, multidimensional feature values ​​of corresponding pixels are extracted to form a labeled sample set: ,in, Let be the multidimensional feature vector of the i-th sample. Let N be the total number of samples, and the percentage of samples in each category be counted as the prior probability of the category. ,in, For category The number of samples, This is used to balance the contribution of different classes of samples to the weight calculation in subsequent improvements to the Relief-F algorithm, and to prevent the feature differences of minority class samples from being masked by the majority class.

[0041] The traditional Relief-F algorithm does not consider imbalanced class distribution and differences in feature dimensions. This step improves the algorithm by introducing prior class probabilities and distance normalization mechanisms to calculate the contribution weight of each feature to sediment classification. The initial weights of all features are set to 0, i.e., a weight vector. For each sample in the sample set... The nearest neighbor samples of the same class and the nearest neighbor samples of different classes are searched using Euclidean distance. With sample Euclidean distance Used to measure the feature similarity between two samples, in relation to Among samples of the same class, find the sample with the smallest Euclidean distance and denote it as the nearest neighbor of the same class. For each category and For samples of different categories, find the sample with the smallest Euclidean distance for each category, and denote it as the nearest neighbor of that category. To eliminate the interference of differences in feature dimensions on the distance calculation, for each feature dimension k, calculate the sample... The difference in normalized features between the sample and its nearest neighbor sample in this dimension. Nearest neighbor of the same kind The normalized difference of the k-th dimension feature is given by the following formula: ; in, These are objects to be classified, among which, This represents the N feature values ​​of the i-th sample. Indicates and Nearest neighbor samples of the same type express and Differences in features, followed by calculation of differences between classes, samples nearest neighbors of different class l The normalized difference of the k-th feature is: ; wherein, represents the nearest neighbor sample of different classes, represents the difference value of the k-th feature between the sample and the nearest neighbor sample of different classes, represents the difference value of the k-th feature between the sample and the nearest neighbor sample of the same class, and the class prior probability , the weight of the k-th feature of each sample is updated according to the following formula, and the weight of the high-contribution feature is iteratively reinforced: ; wherein, represents the weighted sum of the difference of the k-th feature between different class samples, and the greater the weight, represents the difference of the k-th feature between the same class samples, and the smaller the value, the more stable the feature is in the same class samples, After T iterations, the weight of each feature is normalized to map the weight value range to 0-1, and the normalized weight .

[0042] The multi-dimensional features are sorted according to the normalized weights from large to small to obtain a sorted feature list , wherein, is the feature with the largest weight, is the feature with the smallest weight, and the optimal weight threshold is determined by the classification accuracy through 5-fold cross-validation , specifically: 1. The sample set S is randomly divided into 5 mutually exclusive subsets, and each time 4 subsets are taken as the training set and 1 subset is taken as the validation set; 2. The sorted feature list is selected in turn, and the first m features are selected to form a feature subset, and the random forest (RF) classifier is used to train on the training set and calculate the overall classification accuracy (OA) on the validation set; 3. Steps 1-2 are repeated 5 times, the average OA corresponding to each m is calculated, and the minimum m value that makes the average OA highest is found, which is denoted as ; 4. The normalized weight of the th feature after sorting is taken as the weight threshold , that is, .

[0043] After that, features with normalized weights greater than the weight threshold are retained to obtain a high-contribution feature subset. Then, for any two features in the high-contribution feature subset, the Pearson correlation coefficient of the two features is calculated. A redundancy threshold is set. If the absolute value of the correlation coefficient of the two features is greater than the redundancy threshold, the two features are determined to be highly redundant features. The normalized weight of the two highly redundant features is compared, and the feature with the smaller weight is removed. The comparison operation is repeated until the absolute value of the correlation coefficient of any two features in the high-contribution feature subset is less than the redundancy threshold. After the above weight calculation, sorting, screening and redundancy removal, the final core feature set is obtained.

[0044] S6, taking the core feature set as input, training a multi-class classifier and constructing a multi-model collaborative decision mechanism, outputting a substrate classification result and evaluating the classification accuracy through a quantitative index.

[0045] Taking the core feature set as input, a multi-class classifier is trained, a multi-model collaborative decision mechanism is constructed to output a substrate classification result, and a quantitative index is used to evaluate the classification accuracy. The specific process is as follows: Zero-mean standardization is performed on the core feature set to eliminate the dimensional difference of different dimensional features. The samples containing substrate category labels are divided into a training set and a test set in a ratio of 6:4. The substrate category labels are obtained by grab sampling or diving observation. As shown in Figure 3 , Figure 4 Random forest (RF), K nearest neighbor (KNN), support vector machine (SVM), random undersampling enhancement algorithm (RUSBoost) and width neural network (BLS) are selected as multi-class classifiers. 5-fold cross-validation mechanism is used to train each classifier on the training set. The key parameters of each classifier are optimized through grid search to ensure that a single classifier achieves optimal performance in its advantage application scenario. According to the overall classification accuracy (OA) of each classifier on the validation fold of 5-fold cross-validation, the weight of each classifier is calculated by normalization method. The weight value is positively correlated with the overall classification accuracy of the classifier, and the sum of the weights of all classifiers is 1. The test set is input into the trained classifiers to obtain the class probability vector output by each classifier. Based on the weight of each classifier, the weighted sum of the class probability vectors of each test sample is performed. The class with the highest score after weighted sum is taken as the final substrate classification result of the sample. The classification result expression is: ; Wherein, is the number of classifiers (5 in the present application), is the weight of the i-th classifier, is the weight of the i-th classifier, is the weight of the i-th classifier, is the weight of the i-th classifier, The prediction probability of each sample is calculated, a seabed bottom type classification map of the shallow sea island reef area is output, different bottom types are distinguished by color coding, overall classification accuracy (OA) and Kappa coefficient are calculated as two quantitative indexes for evaluating classification accuracy, the overall classification accuracy is the proportion of the number of correctly classified samples to the total number of samples, the Kappa coefficient is used to eliminate the interference of random classification on the accuracy evaluation, measure the consistency of the classification result and the true bottom type, and finally output the result.

[0046] Embodiment 2 The embodiment is different from embodiment 1 in that the embodiment provides a shallow seabed classification system based on multi-scale and multi-directional feature fusion, comprising: The data acquisition module is configured to acquire multi-source acoustic data of the shallow seabed, and pre-process the acquired multi-source acoustic data, including acquiring MBES data and SSS data; The correlation model module is configured to construct a spatial correlation model based on the pre-processed multi-source acoustic data, including determining a key area using an echo intensity image in the MBES data, and performing local matching and feature point matching optimization on the SSS data according to the key area; The terrain feature module is configured to extract seabed terrain features according to the water depth value of the MBES data, including calculating three types of terrain features reflecting seabed bottom characteristics, including slope, slope direction and roughness based on the MBES data; The texture feature module is configured to extract texture features using the SSS image output by the spatial correlation model, including using a gray level co-occurrence matrix to extract texture features and construct a texture feature description vector according to multi-scale and multi-directional parameters; The optimization module is configured to optimize the acquired terrain features and texture features to obtain a core feature set, including using an improved Relief-F algorithm to calculate the weight of different features, sort and filter, and remove redundancies; The output module is configured to input the core feature set, train a multi-class classifier and construct a multi-model collaborative decision mechanism, output the bottom type classification result and evaluate the classification accuracy through the quantitative index.

[0047] The above are preferred embodiments of the present application, and do not limit the protection scope of the present application, therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A shallow seabed sediment classification method based on multi-scale and multi-directional feature fusion, characterized in that, include: Acquire multi-source acoustic data of shallow seabed sediments and preprocess the acquired multi-source acoustic data, including acquiring MBES data and SSS data; A spatial correlation model is constructed based on preprocessed multi-source acoustic data, including determining key regions using echo intensity images in MBES data, and optimizing SSS data by local matching and feature point matching based on key regions. Seabed topographic features are extracted based on water depth values ​​from MBES data, including three types of topographic features that reflect the characteristics of the seabed sediment: slope, aspect, and roughness. Texture feature extraction is performed on the SSS image output by the spatial association model, including using the gray-level co-occurrence matrix, extracting texture features based on multi-scale and multi-directional parameters, and constructing a texture feature description vector; The acquired terrain and texture features are optimized to obtain a core feature set, including weight calculation, sorting and filtering, and redundancy removal of different features using an improved Relief-F algorithm. Using the core feature set as input, a multi-class classifier is trained and a multi-model collaborative decision-making mechanism is constructed. The bottom sediment classification results are output and the classification accuracy is evaluated by quantitative indicators.

2. The shallow seabed sediment classification method based on multi-scale and multi-directional feature fusion according to claim 1, characterized in that, The preprocessing of the acquired multi-source acoustic data includes converting MBES and SSS data to the same projection coordinate system, using the polygonal region actually covered by the MBES data as the reference boundary, performing spatial inclusion cropping on the SSS data image based on the reference boundary, converting the water depth values ​​in all MBES data to a unified vertical reference, and performing geometric correction on the water depth values ​​of the SSS data based on the unified vertical reference. The geometric correction expression is as follows: ； in, To correct the water depth values ​​in the SSS data, The raw water depth values ​​from the SSS data. Indicated as towing height, The slant range of the sonar signal. This represents the horizontal distance component.

3. The shallow seabed sediment classification method based on multi-scale and multi-directional feature fusion according to claim 1, characterized in that, The method of determining key regions using echo intensity images in MBES data includes using the Canny edge detection operator to extract pixel gradient information of the echo intensity image in MBES data, setting significant edge regions based on the edge results extracted by the Canny edge detection operator, calculating the gray-level variance within a local window for the image after extracting pixel gradient information, setting an empirical threshold and determining high variance regions based on the gray-level variance, and performing intersection operations between the high variance regions and significant edge regions to generate an anchoring region mask.

4. The shallow seabed sediment classification method based on multi-scale and multi-directional feature fusion according to claim 1, characterized in that, The optimization of local matching and feature point matching of SSS data based on key regions includes defining MBES key regions based on anchored region masks, using the center coordinates of the MBES key regions as the origin of the SSS data image, obtaining multiple candidate matching windows in a preset neighborhood using the sliding window method, calculating the matching degree between each candidate matching window and the MBES key region template through normalized cross-correlation coefficients, and determining the initial corresponding region pairs based on the cross-correlation coefficients. The matching degree calculation formula is as follows: ； Where A is the MBES critical region template, This represents the grayscale value of the template at the coordinates (x, y). The average grayscale value of the template. For candidate matching windows in the SSS image, This represents the grayscale value of the window at coordinates (x, y). This represents the average grayscale value of the window.

5. The shallow seabed sediment classification method based on multi-scale and multi-directional feature fusion according to claim 1, characterized in that, The optimization of local matching and feature point matching of SSS data based on key regions also includes using the SIFT algorithm to extract feature points in the initial corresponding region pairs, performing nearest neighbor matching on the extracted feature points based on Euclidean distance, eliminating mismatched points through the random consistency sampling algorithm to obtain a set of high-confidence matching point pairs, using cubic spline functions to fit the coordinate mapping relationship between MBES data and SSS data in the x and y directions respectively, using the set of high-confidence matching point pairs as control points, solving the fitting coefficients through the least squares method, and combining the fitting coefficients and cubic spline functions to obtain the spatial association model.

6. The shallow seabed sediment classification method based on multi-scale and multi-directional feature fusion according to claim 1, characterized in that, The process of extracting seabed topographic features based on the water depth values ​​from MBES data includes: calculating the elevation change rate of target pixels in the x-axis and y-axis directions using a difference algorithm based on the water depth values ​​from MBES data; converting the elevation change rate into an angle value using an arctangent function to obtain the slope; calculating the initial aspect value based on the elevation change rate using the arctangent function; adjusting the initial aspect value using the positive and negative signs of the elevation change rate in the x-axis and y-axis directions; and finally, fitting the water depth data of the seabed area covered by MBES data into a three-dimensional surface, calculating the ratio of the actual area of ​​the three-dimensional surface to the projected area of ​​the seabed area to obtain the roughness.

7. The shallow seabed sediment classification method based on multi-scale and multi-directional feature fusion according to claim 1, characterized in that, The process of extracting texture features from the SSS image output by the spatial association model includes: presetting multiple sliding windows of different sizes and multiple main directions of different angles as multi-scale parameters and multi-directional parameters, respectively; performing gray-level compression processing on the position-calibrated SSS image; for the combination of multi-scale and multi-directional parameters, counting the occurrence frequency of different gray-level combinations within the corresponding sliding window according to the preset pixel spacing; normalizing the occurrence frequency to obtain probability values; constructing a gray-level co-occurrence matrix based on the probability values; extracting multiple texture features based on each gray-level co-occurrence matrix; and sorting the multiple texture features to obtain a texture feature description vector.

8. The shallow seabed sediment classification method based on multi-scale and multi-directional feature fusion according to claim 1, characterized in that, The feature optimization of the acquired terrain and texture features includes introducing class prior probability and distance normalization mechanisms, calculating feature weights for terrain and texture features, calculating the nearest neighbors of all features of the same class and the nearest neighbors of different classes using Euclidean distance, calculating the feature difference between each feature and the nearest neighbors of the same class and the nearest neighbors of different classes, iteratively updating the feature weights according to the feature differences, retaining high-contribution features by combining the weight threshold determined by cross-validation, and removing redundant features within the high-contribution features using the Pearson correlation coefficient. The feature difference formula is as follows: ； in, For the feature samples to be processed, For category The prior probability, For category In and sample The j-th nearest sample, For the sample The true substrate category, The maximum value among all samples. The minimum value among all samples. For the sample Prior probability of substrate type.

9. The shallow seabed sediment classification method based on multi-scale and multi-directional feature fusion according to claim 1, characterized in that, The training of multi-class classifiers and the construction of a multi-model collaborative decision-making mechanism include selecting random forest, K-nearest neighbors, support vector machine, random undersampling enhancement algorithm, and wide neural network as multi-class classifiers; training each classifier separately using a 5-fold cross-validation mechanism; calculating the weight of each classifier using a normalization method based on the overall classification accuracy of each classifier on the validation fold of the 5-fold cross-validation; performing a weighted summation based on the weights of each classifier; and taking the class with the highest score after weighted summation as the final subspecies classification result for the sample. The classification result expression is: ； in, For the number of classifiers, For the first The weights of each classifier For classifier Category The predicted probability.

10. A shallow seabed sediment classification system based on multi-scale, multi-directional feature fusion, executed according to the method of claim 1, characterized in that, include: The data acquisition module is configured to acquire multi-source acoustic data of shallow seabed sediment and preprocess the acquired multi-source acoustic data, including acquiring MBES data and SSS data. The correlation model module is configured to: construct a spatial correlation model based on preprocessed multi-source acoustic data, including using echo intensity images in MBES data to determine key regions, and optimizing local matching and feature point matching of SSS data based on key regions; The terrain feature module is configured to extract seabed terrain features based on the water depth values ​​of MBES data, including three types of terrain features that reflect the characteristics of the seabed sediment: slope, aspect, and roughness, calculated based on MBES data. The texture feature module is configured to extract texture features from the SSS image output by the spatial association model, including using the gray-level co-occurrence matrix, extracting texture features based on multi-scale and multi-directional parameters, and constructing a texture feature description vector. The optimization module is configured to: perform feature optimization on the acquired terrain and texture features to obtain a core feature set, including using the improved Relief-F algorithm to calculate weights, sort and filter different features, and remove redundancy. The output module is configured to: take the core feature set as input, train a multi-class classifier and build a multi-model collaborative decision-making mechanism, output the sediment classification result and evaluate the classification accuracy through quantitative indicators.

Images