Method for evaluating biological attachment level of underwater structure

Abstract

The invention provides an underwater structure biological attachment grade evaluation method, which comprises the steps of performing sharpening processing on an edge blurred image region, extracting attachment contour features of the edge blurred image region, performing multipath suppression processing on a reflection waveform of a corresponding region according to the attachment contour features, and obtaining attachment texture information and depth contour data; analyzing attachment texture information, extracting surface roughness and texture directivity, distinguishing hardness in combination with depth contour data, and if the roughness is high and the depth contour changes drastically, judging that the hardness is high, and determining an attachment type through clustering analysis; attachment visual texture features and sonar depth features are extracted from attachment types, fusion weights of the texture features and the depth features are obtained through weighted calculation, and a target feature weight ratio is determined according to the fusion weights.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a method for evaluating the biological attachment level of underwater structures. BACKGROUND

[0002] The evaluation of biological attachment level of underwater structures plays a crucial role in the fields of marine engineering maintenance, ship anti-pollution, and marine ecological monitoring, and is directly related to the safe operation and economic benefits of marine infrastructure. Accurate evaluation of the degree of biological attachment is of decisive significance for developing reasonable cleaning and maintenance strategies and prolonging the service life of equipment. Current underwater biological attachment detection methods mainly rely on single sensor technology. Visual systems are easily affected by lighting conditions and water quality changes in complex underwater environments, while sonar systems have good penetration ability but limited detail recognition capability on complex-shaped structure surfaces, making it difficult to provide stable and reliable detection results. The increased turbidity caused by ocean current disturbance in underwater environments leads to blurred edges in visual images, making it difficult to segment the boundaries of attached objects. This boundary blurring phenomenon further exacerbates the multipath effect problem of sonar systems. Multipath effects distort the reflected waveform, which directly affects the accurate recognition of different types of attached objects, especially in distinguishing between viscous algae and calcified shellfish with different texture and hardness characteristics. Distorted waveform signals make traditional feature extraction methods ineffective. More complex is the unevenness of the structure surface and the mutual shielding phenomenon between biological attachments, which leads to the loss of key feature information and seriously affects the accurate reconstruction of the attachment distribution. How to dynamically adjust the fusion weight of different sensor data through multi-modal visual and sonar collaborative perception technology under complex underwater environmental conditions such as ocean current disturbance, effectively cope with turbidity changes, multipath effects, and feature occlusion interference factors, and realize accurate evaluation and distribution reconstruction of the biological attachment level of underwater structure surfaces has become a key problem in the development of current underwater detection technology. SUMMARY

[0003] The present application provides a method for evaluating the biological attachment level of underwater structures, mainly including:

[0004] Collecting images of underwater structures and performing gray scale conversion and filtering processing to obtain edge images of the underwater structures, analyzing the spatial frequency and contrast gradient of the edge images, and identifying edge blurred image regions with a rising amplitude of turbidity change exceeding a preset threshold;

[0005] Performing sharpening processing on the edge blurred image regions, extracting the attachment outline features of the edge blurred image regions, and performing multipath suppression processing on the reflected waveform of the corresponding regions according to the attachment outline features to obtain attachment texture information and depth profile data;

[0006] The texture information of the attached object is analyzed, the surface roughness and the texture directionality are extracted, the hardness is distinguished in combination with the depth profile data, if the roughness is high and the depth profile changes sharply, it is judged that the hardness is higher, and the type of the attached object is determined through cluster analysis;

[0007] The visual texture features and the sonar depth features of the attached object are extracted from the type of the attached object, the fusion weight of the texture features and the depth features is obtained through weighted calculation, and the target feature weight ratio is determined according to the fusion weight;

[0008] The visual texture features and the sonar depth features are weighted and fused according to the weight ratio, a comprehensive feature vector is generated, a spatial distribution model is established according to the type of the attached object and the depth profile data, the comprehensive feature vector and the environmental disturbance level are input, and the attached distribution reconstruction data is output;

[0009] According to the attached distribution reconstruction data, the feature loss position of the concave and convex parts and the biological shielding area of the structure surface is identified, the missing features are identified through a feature matching algorithm, the feature loss position is compensated and reconstructed by using the missing features, and complete attached object distribution information is obtained.

[0010] According to the complete attached object distribution information, the attached object coverage area ratio and the distribution density obtained by the number of attached objects per unit area are calculated in combination with the attached object spatial distribution model, and the hardness distinguishing processing result is obtained, and the biological attached grade evaluation result of the underwater structure is obtained.

[0011] The technical scheme provided by the embodiment of the present application can include the following beneficial effects:

[0012] The present application discloses a kind of underwater structure biological attachment grade evaluation methods, through collecting underwater structure image and carrying out edge extraction and turbidity analysis, edge fuzzy area is identified and sharpening processing is carried out, and attached object profile feature is extracted.Combined with sonar depth data, attached object texture information and hardness are analyzed, and the type of the attached object is determined.By fusing visual texture features and sonar depth features, an attached object spatial distribution model is established, and attached distribution reconstruction data is generated.For feature loss position compensation reconstruction, complete attached object distribution information is obtained.Finally, according to the information such as attached object coverage area, distribution density and hardness, the biological attachment grade of underwater structure is comprehensively evaluated.The present application can accurately identify and evaluate the biological attachment condition of underwater structure surface, and provide important basis for maintenance and management of underwater structure. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 The flow chart of the underwater structure biological attachment grade evaluation method of the present application. DETAILED DESCRIPTION

[0014] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be described in detail below with reference to the drawings and specific embodiments.

[0015] As Figure 1 The underwater structure biological attachment grade evaluation method of the embodiment can specifically include the following steps.

[0016] In step S101, an image of the underwater structure is collected, and a gray scale conversion and a filtering process are performed to obtain an edge image of the underwater structure. The spatial frequency and the contrast gradient of the edge image are analyzed, and an edge blurred image region with a turbidity change rising amplitude exceeding a preset threshold is identified.

[0017] After the image of the underwater structure is collected, a gray scale conversion is performed on the image to obtain a gray scale image. A Gaussian filter is used to filter the gray scale image. The smoothing degree of the filter is determined by setting a standard deviation parameter σ. A weighted average operation is performed on each pixel point and its neighborhood. The weight decays exponentially with the increase of the distance from the center pixel. Image noise is removed, and a smoothed gray scale image is obtained. A Sobel operator is used to perform edge detection on the smoothed gray scale image. The horizontal direction gradient and the vertical direction gradient are calculated respectively. The gradient amplitude is obtained by taking the square root of the sum of the squares of the two direction gradients. The pixel points with a gradient amplitude exceeding a preset threshold are extracted as edge points. The gradient direction information of each edge point is recorded. An edge image containing edge position and direction information is generated. For the edge image, an 8x8 pixel neighborhood window is selected for each edge point. The horizontal direction change rate of the gray scale value in the window is calculated as the row frequency. The vertical direction change rate is calculated as the column frequency. The spatial frequency value is obtained by taking the square root of the sum of the squares of the row frequency and the column frequency. The maximum gray scale value and the minimum gray scale value in the window are obtained. The ratio of the difference between the two values to the sum of the two values is calculated as the contrast gradient. The edge points with both the spatial frequency and the contrast gradient lower than the respective preset thresholds are recorded. The recorded low-frequency low-contrast edge points are subjected to connected component analysis. The continuity of the edge points is determined according to the gradient direction information. Independent connected regions are formed. Regions with a spatial frequency higher than a preset standard are selected from the edge image as clear reference regions. The average spatial frequency value is calculated. The difference percentage between the average spatial frequency of each connected region and the average spatial frequency of the clear reference region is calculated as a turbidity change index. The connected regions with a turbidity change index exceeding a preset threshold are identified as edge blurred image regions.

[0018] Specifically, in the process of underwater structure image acquisition, the acquired image often has edge blur phenomenon due to the influence of water turbidity, light scattering, suspended particles and other factors. Gray scale conversion is the process of converting color image into single channel image, which combines the pixel values of red, green and blue three channels into a single gray value through weighted average method. This conversion not only reduces the data processing amount, but more importantly, eliminates the interference of color information on edge detection, making the subsequent edge feature extraction more accurate.

[0019] In a possible implementation, the smoothing degree of the Gaussian filter is controlled by the standard deviation parameter σ. When the σ value is small, the filter mainly removes high-frequency noise and retains more image details; when the σ value is large, the filtering effect is stronger, but it may cause loss of edge information. For underwater images, a moderate σ value is usually selected, which can effectively suppress the random noise caused by particles in the water body and maintain the clarity of the structure edge. The core of the weighted average operation is that the closer the neighborhood pixel is to the center pixel, the greater the weight, and this weight distribution conforms to the continuity feature of natural images.

[0020] It should be noted that the Sobel operator detects the gray scale changes in the horizontal and vertical directions through two 3x3 convolution kernels. The horizontal direction gradient reflects the brightness change in the horizontal direction of the image, while the vertical direction gradient reflects the vertical change. The calculation of the gradient amplitude integrates the change intensity in the two directions. When the gradient amplitude of a certain pixel point exceeds the preset threshold, it indicates that there is a significant gray scale jump at this point, which is likely to be an object edge. The recording of the gradient direction information provides an important basis for subsequent edge continuity judgment, because the real object edge usually has consistent direction.

[0021] Specifically, the calculation of spatial frequency is based on an 8x8 pixel neighborhood window. The row frequency is obtained by calculating the difference between the gray values of adjacent pixels in each row in the window, and the root mean square of these differences reflects the intensity of the gray scale change in the horizontal direction. Similarly, the column frequency reflects the vertical change. When the edge region is affected by turbidity, the originally clear gray scale jump becomes flat, resulting in a decrease in spatial frequency. The contrast gradient evaluates the edge quality from another angle, which quantifies the contrast of the local region through the relative difference between the maximum and minimum gray values. The combination of low spatial frequency and low contrast gradient effectively identifies the blurred edge affected by turbidity.

[0022] In one embodiment, connected domain analysis utilizes the gradient direction information of edge points. If adjacent edge points have similar gradient directions, it means that they belong to the same edge line. The connected region formed in this way represents a complete edge segment. The selection of clear reference areas is based on the spatial frequency distribution, and usually areas with frequency values ​​in the upper quartile are selected, which are least affected by turbidity. The percentage calculation method of the turbidity change index intuitively reflects the degree of blur. When the index exceeds the preset threshold, it indicates that the edge quality of the area has been severely degraded and requires special processing or re-acquisition. This hierarchical recognition method provides accurate area positioning for subsequent image enhancement or quality assessment.

[0023] Step S102 : sharpening the edge-blurred image area, extracting the contour features of the attachments in the edge-blurred image area, and performing multipath suppression on the reflected waveform of the corresponding area according to the contour features of the attachments to obtain attachment texture information and depth profile data.

[0024] The Laplace sharpening operator is applied to image regions with blurred edges. By calculating the second-order derivative of each pixel and its eight neighboring pixels, the high-frequency components in the image are enhanced, blurring the blurred edges and obtaining a sharpened image region. The gradient enhancement value of each pixel after sharpening is also recorded. The sharpened image region is binarized and the attachment contour is extracted using an adaptive threshold segmentation method. A sequence of contour boundary points is obtained by traversing the boundary pixels clockwise. The contour perimeter is calculated by summing the distances between adjacent boundary points. The number of pixels enclosed by the contour is calculated using the Green formula to obtain the contour area. The shape factor is determined based on the ratio of the square of the perimeter to the area. The area ratio is also calculated by calculating the ratio of the number of attachment pixels to the total number of pixels in the region. Based on the contour boundary point sequence, shape factor, and area ratio, the geometric characteristic parameters of the attachment are determined. The pre-collected ultrasonic reflection signal of the corresponding region is obtained. Multiple reflection peaks caused by the irregular surface of the attachment are identified in the signal. The filter parameters are adjusted based on the shape factor to suppress secondary reflection peaks while retaining the primary reflection peak signal and its corresponding time delay information. The retained main reflection peak signal is subjected to amplitude analysis. Based on the correspondence between the signal intensity distribution and the gradient enhancement value, attachment texture information reflecting the roughness of the attachment surface is generated. The depth value of each point on the attachment surface is calculated based on the time delay information of the main reflection peak and the sound wave propagation speed. Combined with the position coordinates of the contour boundary point sequence, depth profile data including lateral position and longitudinal depth are constructed.

[0025] Specifically, the core of Laplacian sharpening operator is to detect the second derivative change of gray value in the image. In the underwater environment, due to the influence of light scattering and water turbidity, the edge of the attached object often presents the characteristics of gradual and fuzzy. By calculating the gray difference between the center pixel and the surrounding eight neighborhood pixels, and assigning different weight coefficients, the gray jump at the edge can be highlighted. The gradient enhancement value records the enhancement degree of each pixel point after sharpening processing, which directly reflects the severity of the original blur at this position, and provides an important reference for subsequent texture analysis.

[0026] It should be noted that the adaptive threshold segmentation method dynamically adjusts the segmentation threshold according to the gray distribution of the local region of the image. For underwater attached object images, due to uneven illumination and differences in attached object materials, there are great differences in their gray characteristics. The traditional global threshold method is difficult to accurately segment the complete attached object contour. The process of traversing the boundary pixels clockwise is actually tracking along the boundary between the target and the background in the binary image, and recording the coordinate position of each boundary point to form an ordered contour point sequence.

[0027] Specifically, Green's formula has a unique advantage in calculating the area of irregular contours. By adding the areas of the triangles formed by the adjacent two points in the contour boundary point sequence and the origin, the area of any shape closed contour can be accurately calculated. The shape factor quantifies the complexity of the contour by the ratio of the square of the perimeter to the area. The shape factor of a circular contour is close to 12.56, while the irregular attached object contour usually has a larger shape factor value. The area ratio reflects the coverage degree of the attached object in the detection area. These two parameters together describe the geometric characteristics of the attached object.

[0028] In one possible implementation, the ultrasonic wave generates multiple reflection signals when encountering the surface of the adhering object. The main reflection peak is from the specular reflection of the surface of the adhering object, and the secondary reflection peak is caused by the surface roughness, the inclination angle, and the internal structure. The shape factor plays a key role here: the more irregular the shape of the adhering object, the more complex the multipath reflection generated by the surface thereof. By adjusting the bandwidth and the center frequency of the filter through the shape factor, the main reflection signal can be retained and the interference components can be suppressed. The time delay information records the time interval from the emission to the reception of the ultrasonic wave, which is the basic data for calculating the depth. There is a corresponding relationship between the gradient enhancement value and the reflection signal intensity. The area with a high degree of blur in the original image has a large gradient enhancement value, and the surface of the corresponding adhering object is often rougher, and the intensity distribution of the reflection signal is also more dispersed. By analyzing this corresponding relationship, the micro-texture characteristics of the surface of the adhering object can be inferred. The depth calculation is based on the propagation speed of the sound wave in water, which is usually about 1500 meters per second, and the accurate depth value can be calculated in combination with the round-trip time delay. The depth information is combined with the plane coordinates of the contour boundary points to form complete three-dimensional depth profile data, realizing accurate reconstruction from two-dimensional images to three-dimensional morphology.

[0029] In step S103, the adhering object texture information is analyzed, the surface roughness and the texture directionality are extracted, the hardness is distinguished in combination with the depth profile data, if the surface roughness is high and the depth profile changes dramatically, it is judged that the hardness is high, and the type of the adhering object is determined through cluster analysis.

[0030] The adhering object texture information is analyzed, the adjacent pixels with a distance of 1 in the horizontal, vertical, and diagonal directions of each pixel point are selected, the frequency of the occurrence of different gray level combinations is counted, a gray level co-occurrence matrix is constructed, a contrast value is obtained by calculating the square sum of the difference between each element in the matrix and the average value of the row and column thereof, the contrast value is taken as a surface roughness index, the contrast values in the horizontal, vertical, main diagonal, and secondary diagonal directions are calculated respectively, and the difference between the maximum value and the minimum value is taken as a texture directionality parameter. According to the depth profile data, the height difference between each sampling point and its eight adjacent sampling points is calculated, the root mean square value of all height difference values is taken as the local variation degree of the point, the average value of the local variation degrees of all sampling points is taken as an index of the profile change degree, the surface roughness index and the profile change degree index are compared with respective preset threshold values respectively, if both of them exceed the threshold values, a high hardness identifier is generated, otherwise, a low hardness identifier is generated. The surface roughness index, the texture directionality parameter, the profile change degree index, and the hardness identifier are combined into a four-dimensional feature vector, the K-means clustering algorithm is used to cluster the feature vectors of multiple adhering object regions, the number of clusters is set to be the preset number of adhering object types, the Euclidean distance between each cluster center and the standard adhering object feature value is calculated, and the standard type with the smallest distance is selected as the adhering object type corresponding to the cluster.

[0031] Specifically, the gray level co-occurrence matrix is an important tool for texture analysis, whose core idea is to count the frequency of pixel pairs with specific spatial relationship in an image. In the underwater fouling image, the adjacent pixels with distance of 1 are chosen because the texture features on the surface of the fouling are usually obvious in micro scale. When the gray values of two pixels are similar, it indicates that the texture in this region is smooth; when the gray values are quite different, it indicates that there is texture fluctuation. By counting the occurrence times of different gray level combinations, a two-dimensional matrix is formed, and each element of the matrix represents the probability of occurrence of a specific gray pair.

[0032] It should be noted that the calculation of the contrast value reflects the clarity of the texture. The elements far from the main diagonal in the matrix represent pixel pairs with large gray difference, and the larger the value of these elements, the rougher the texture. The sum of the square of the mean difference in row and column is an effective method to quantify this distribution feature. In the underwater environment, the contrast value of the smooth metal fouling surface is low, while the rough biological fouling such as barnacles, shellfish, etc. has a higher contrast value. The selection of the four directions covers the main texture direction, and the horizontal and vertical directions reflect the regular texture, and the diagonal direction captures the oblique texture feature.

[0033] Specifically, the calculation of the local variation degree makes full use of the spatial information of the depth profile data. Each sampling point and its eight surrounding neighborhood points form a 3x3 local window, and the height difference value reflects the fluctuation of the point relative to the surrounding. The use of the root mean square value avoids the mutual offset of positive and negative differences, and truly reflects the severity of the height variation. Although the surface of hard fouling such as shell, cement block, etc. may have fine texture, its overall profile is relatively flat, and the local variation degree is small. Soft fouling such as seaweed, silt, etc. has a significantly larger local variation degree due to its softness and the formation of fluctuating surface under the action of water flow.

[0034] In a possible implementation, the combination of surface roughness and profile variation intensity embodies the idea of multi-dimensional feature fusion. High roughness usually corresponds to hard mineral deposition or biological calcification layer, while intense profile variation may be caused by deformation of soft material. When both indicators exceed the threshold, it means that the attachment has both rough microtexture and undulating macroform, and this feature combination is most likely to correspond to hard biological attachment. The generation of hardness identification provides an important basis for subsequent classification. The K-means clustering algorithm processes four-dimensional feature vectors by iterative optimization to make the features of the same type of attachment as similar as possible. The preset number of clusters is based on common types of underwater attachments, usually including barnacles, shellfish, seaweed, and silt. The calculation of Euclidean distance considers the differences in all feature dimensions, and the smaller the distance, the higher the similarity. The standard attachment feature values are obtained by a large number of sample statistics, representing the typical feature distribution of each type of attachment. This feature-based classification method considers both the micro characteristics of the texture and the macro performance of the form, achieving accurate identification of the type of attachment.

[0035] Check the texture features of the attachment surface, evaluate the roughness or smoothness of the attachment surface, combine the texture directionality, judge whether the attachment is a natural deposition layered structure or a disordered trace of biological activity, and combine the depth profile data to distinguish the hardness difference of the attachment through three-dimensional topographic analysis, judge whether the attachment belongs to mineral deposition or organic residue according to the hardness difference, and determine the type of attachment by using a clustering algorithm.

[0036] The texture feature of the attached surface is checked, the absolute value of the gray level difference of each pixel point in the texture image and its four adjacent pixels is calculated, the average value of all the difference values is taken as the roughness evaluation value, if the value exceeds the preset roughness threshold, it is determined as a rough surface, otherwise it is determined as a smooth surface, the number of pixel pairs with the gray value change less than the preset change threshold in the horizontal direction and the vertical direction is counted respectively, the ratio of the two directions is calculated to obtain the texture directionality index. According to the texture directionality index and the roughness evaluation value, if the texture directionality index is greater than the preset stratification threshold, it is judged as a natural deposition stratified structure, if the texture directionality index is close to 1 and the roughness evaluation value exceeds the preset roughness threshold, it is judged as a biological activity disordered trace, and the structure type identifier and the surface feature parameter are generated. The height value sequence in the depth profile data is used to calculate the angle change formed by each sampling point and its adjacent two points, and the average value of the angle change is taken as the morphological relief degree. The surface feature parameter and the morphological relief degree are combined, and the comprehensive hardness index is obtained by weighted summation. According to the comparison result of the comprehensive hardness index and the preset hardness threshold, it is determined that the attached object belongs to mineral deposition or organic residue. The roughness evaluation value, the structure type identifier, the comprehensive hardness index and the attached object attribute determination result are combined to form a feature vector, the hierarchical clustering algorithm is used to calculate the distance between the feature vectors of different attached object samples, the samples with a distance less than a preset clustering threshold are classified into the same class, and the attached object type corresponding to each class is determined according to the common features of the samples in the class.

[0037] Specifically, the roughness evaluation of the texture feature is realized by accurate quantization through four-neighbor pixel difference calculation. In the underwater environment, the attached surface is affected by factors such as water flow scouring and biological erosion, forming rough textures of different degrees. The gray value difference between adjacent pixels on the smooth metal oxide layer surface is very small, usually within 5 gray levels; while the rough shell attached surface, due to the unevenness of calcium deposition, the difference between adjacent pixels can reach more than 30 gray levels. The average value of such difference directly reflects the degree of surface micro relief.

[0038] It should be noted that the calculation of the texture directionality index is based on the statistical principle of gray value continuity. The number of pixel pairs with a gray value change less than the threshold in the horizontal direction reflects the extension characteristics of the texture in the horizontal direction; the vertical direction statistics reveal the longitudinal texture distribution. The stratified structure formed by natural deposition, such as silt deposition layer, presents obvious horizontal stratification characteristics under the action of gravity, and the horizontal continuity is much higher than the vertical direction, and the ratio is usually more than 3. On the contrary, the growth of biological attached objects such as barnacles is not limited by a specific direction, and the continuity in each direction is similar, and the ratio is close to 1.

[0039] Specifically, the generation of structure type identification comprehensively considers the two dimensions of directionality and roughness. Although the layered structure has obvious directionality, its surface is relatively smooth, and the roughness evaluation value is low. The disordered traces formed by biological activities show a combination of high roughness and low directionality. This dual judgment mechanism avoids misjudgment caused by a single feature. Surface feature parameters as intermediate results provide important input for subsequent hardness analysis.

[0040] In one possible implementation, the calculation of morphological relief degree utilizes the spatial variation information of depth profile data. The included angle formed by three consecutive sampling points reflects the bending degree of the local surface. The included angle of the smooth mineral deposition surface changes little, usually within 5 degrees; while the soft organic residues form a wavy relief under the action of water flow, and the included angle changes by more than 20 degrees. The comprehensive hardness index combines the surface roughness and morphological relief degree by weighted summation, and the weight coefficients are determined in advance according to the characteristic distribution of different types of attachments. The hierarchical clustering algorithm shows unique advantages in processing attachment classification. The algorithm starts from each sample as an independent category, gradually merges the categories with the closest distance, and forms a tree-like clustering structure. The distance between feature vectors is measured by Euclidean distance, which comprehensively considers the differences in multiple dimensions such as roughness, structure type, hardness, etc. When the distance between two samples is less than the preset threshold, it means that they have similar physical and chemical properties, and are likely to belong to the same type of attachment. By analyzing the common features of each cluster, such as the combination of high roughness, layered structure, and high hardness usually corresponds to calcareous deposits, while low roughness, disordered structure, and low hardness point to soft algal attachments, thus achieving accurate identification of attachment types.

[0041] Step S104, extracting the visual texture features and sonar depth features of the attachments from the attachment types, calculating the fusion weight of the texture features and the depth features by weighting, and determining the target feature weight ratio according to the fusion weight.

[0042] The visual texture features are extracted from the determined attachment types, the number of texture edge points in a unit area is calculated to obtain a texture density value, the proportion of the main direction in the total edge point number is taken as a direction consistency index by using a gradient direction histogram, and a gray scale contrast value is calculated by calculating the difference between the maximum gray scale value and the minimum gray scale value in a local window. The sonar detection signal corresponding to the attachment area is obtained, the depth value is calculated according to the time difference between transmission and reception, the depth gradient is obtained by calculating the ratio of the depth difference value of adjacent sampling points to the horizontal distance, the curvature change value is obtained by selecting three continuous sampling points to fit a quadratic curve by the least square method and calculating the second derivative thereof, and the reflection intensity is obtained according to the ratio of the receiving signal amplitude to the transmitting signal amplitude. The texture density value, the direction consistency index and the gray scale contrast value are normalized by dividing by the maximum value of each, respectively, to form a three-dimensional texture feature vector, the depth gradient, the curvature change value and the reflection intensity are normalized by the same method to form a three-dimensional depth feature vector, and the variance values of the six feature components between different attachment types are calculated. According to the variance values, the initial weight values of the feature components are proportionally distributed, the greater the variance, the higher the weight, the corresponding components of the texture feature vector and the depth feature vector are weighted using the initial weight values, the ratio of the inter-class distance to the intra-class distance of the weighted features between different attachment types is calculated, if the ratio is lower than a preset discrimination threshold, the weight of the feature with larger variance is increased and the weight of the feature with smaller variance is decreased, the calculation is repeated until the ratio meets the requirements, and the final six feature component weight values are output as the target feature weight ratio.

[0043] Specifically, the calculation of texture density is based on the edge detection result, and the number of edge points in a unit area is counted to quantify the fineness of the texture. In the underwater environment, the surface of barnacle attachment is covered with small calcareous protrusions, and there are hundreds of edge points per square centimeter; while the smooth metal corrosion layer has only a dozen edge points. This density difference directly reflects the surface structure characteristics of the attachment, providing an important distinguishing dimension for subsequent feature fusion.

[0044] It should be noted that the construction process of the gradient direction histogram involves the statistics of the gradient direction of each edge point. The direction range of 0 to 360 degrees is evenly divided into 36 intervals, each interval representing a direction range of 10 degrees. The number of edge points falling into each interval is counted, and the interval with the most number is taken as the main direction. The direction consistency index is obtained by calculating the proportion of the main direction edge points in the total edge points. The edges of regular artificial structure surfaces, such as concrete blocks, are mainly distributed along the horizontal or vertical directions, and the consistency index can reach more than 0.8; while the growth direction of biological attachment is random, and the consistency index is usually less than 0.3.

[0045] Specifically, the depth calculation of the sonar detection signal utilizes the propagation characteristics of sound waves in water. The propagation speed of sound waves in seawater is about 1500 meters per second, and by accurately measuring the time difference between the transmitted pulse and the received echo, the depth of the target point can be calculated. The depth gradient reflects the degree of inclination of the surface of the attached object, and the depth gradient of the steep shell edge can reach 0.5, while the depth gradient of the flat silt layer is only 0.05. The curvature change value is obtained by least squares fitting, which can effectively smooth the measurement noise and accurately reflect the bending characteristics of the surface.

[0046] In one possible implementation, the normalization processing ensures the comparability of different dimensional features. The texture density value can vary from 10 to 200, while the reflection intensity is usually between 0.1 and 0.9, and direct comparison will lead to the dominance of features with large numerical values in the fusion result. By dividing by the respective maximum value, all features are mapped to the range of 0 to 1. Variance calculation reveals the discrimination ability of each feature, and a large variance indicates that the feature is significantly different between different types of attached objects and should be given a higher weight. The calculation of the ratio of inter-class distance to intra-class distance embodies the idea of discriminant analysis. The inter-class distance measures the separation degree of the feature centers of different types of attached objects, and the intra-class distance reflects the aggregation degree of the features of the same type of attached objects. The ideal feature weight allocation should separate the features of different types of attached objects as much as possible while keeping the features of the same type closely aggregated. By iteratively adjusting the weights, when the ratio exceeds a preset threshold such as 3.0, it indicates that the weighted features have good discriminant ability. The output of the six weight values, such as texture density 0.2, direction consistency 0.15, gray contrast 0.15, depth gradient 0.2, curvature change 0.15, and reflection intensity 0.15, constitutes an optimized feature fusion scheme, realizing the effective combination of visual texture features and sonar depth features.

[0047] Step S105, according to the weight allocation, the visual texture features and the sonar depth features are weighted and fused to generate a comprehensive feature vector, a spatial distribution model is established based on the attached object type and depth profile data, the comprehensive feature vector and the environmental disturbance level are input, and the attached distribution reconstruction data is output.

[0048] According to the determined weight ratio, each component of the visual texture feature is multiplied by the corresponding weight value, and each component of the sonar depth feature is multiplied by the corresponding weight value. All the weighted feature components are arranged in order to generate a comprehensive feature vector containing six weighted feature components. Using the attached object type identifier and the depth profile data, the detection area is divided into grid cells at a predetermined interval. The center coordinates, the contained attached object type, and the average depth value of each grid cell are recorded. At the same time, the comprehensive feature vector is assigned to the corresponding grid cell to construct a spatial distribution data set containing position, type, depth, and feature information. Environmental parameters including water flow velocity, temperature, and suspended matter concentration are obtained from the sensor. According to the comparison results of each parameter with the preset threshold, the environmental disturbance level is determined. The comprehensive feature vector of each grid in the spatial distribution data set is associated with the environmental disturbance level for calculation. The inverse distance weighted interpolation method is used to calculate the attached object existence probability of each grid according to the feature similarity and distance of adjacent grids. According to the attached object existence probability, the number of grids with probability exceeding the preset existence threshold in a unit area is counted to obtain the distribution density value. Adjacent grids with probability exceeding the threshold are identified and their boundaries are connected to obtain the coverage area contour. The difference between the depth value of each high-probability grid and the reference depth is calculated as the thickness of the point. The average thickness and maximum thickness are calculated by counting all the thickness values. The distribution density value, coverage area contour coordinate sequence, average thickness, and maximum thickness are combined and output as attached distribution reconstruction data.

[0049] Specifically, the generation process of the comprehensive feature vector embodies the core idea of multi-modal data fusion. The six weighted feature components are arranged in a fixed order to form a unified numerical sequence. The texture density component is weighted by 0.2, reflecting the richness of the attached object surface details. The direction consistency component is multiplied by a weight of 0.15, representing the regularity of the texture. The gray contrast component retains the visual light and dark difference information. The depth gradient, curvature change, and reflection intensity in the sonar depth feature carry morphological and material information. This ordered arrangement ensures that the correspondence of the features remains consistent in subsequent processing.

[0050] It should be noted that the grid division discretizes the continuous underwater space using equal intervals. The selection of the preset interval needs to balance accuracy and computational efficiency, and is usually set to half of the minimum size of the attached object. Each grid cell serves as an independent analysis unit, with its center coordinates determining the spatial position, the attached object type identifier recording the main attached object species at that position, and the average depth value obtained by averaging all depth sampling points within the grid. Assigning the comprehensive feature vector to the grid actually establishes a correspondence between the feature and the spatial position, so that each spatial point carries complete multi-dimensional feature information.

[0051] Specifically, the determination of the environmental disturbance level is based on a comprehensive assessment of multiple environmental parameters. Water velocity directly influences the distribution of biofouling, with biofouling typically taking on a streamlined pattern in areas with high-speed water flow. Temperature fluctuations affect the growth rate of biofouling, with areas with large temperature fluctuations exhibiting a greater diversity of biofouling species. Suspended matter concentration influences the rate of biofouling deposition. When water velocity exceeds 2 meters per second, the daily temperature fluctuation exceeds 5 degrees Celsius, and the suspended matter concentration exceeds 100 milligrams per liter, the environmental disturbance level is classified as high. This grading mechanism provides an environmental correction factor for subsequent probability calculations.

[0052] In one possible implementation, the inverse distance weighted interpolation method takes into account the principle of spatial correlation. This method assumes that closer points have more similar features, with weights inversely proportional to distance. For each grid point to be calculated, the probability of the presence of an attachment is calculated by taking the weighted average of the probabilities of surrounding known points. Weight calculation not only considers spatial distance but also incorporates feature similarity, quantifying the degree of similarity by calculating the Euclidean distance between the combined feature vectors. The environmental disturbance level is incorporated as a correction factor, increasing the interpolation range in high-disturbance environments. The generation of attachment distribution reconstruction data converts a discrete probability distribution into a practical engineering parameter. The distribution density is calculated by counting the number of high-probability grid cells and dividing it by the total area, providing a direct reflection of the concentration of attachments. The coverage area outline is extracted using a boundary tracing algorithm. Starting from a high-probability grid, the boundary points of adjacent high-probability grid cells are sequentially connected to form a closed outline. The reference depth for thickness calculation is typically the average depth of the unattached area, and the thickness of each point is the difference between the depth at that point and the reference depth. The statistics of average thickness and maximum thickness provide important reference for subsequent cleaning operations, realizing the complete transformation from feature analysis to practical application.

[0053] Step S106: Identify the feature loss locations of the concave and convex parts of the structure surface and the biological occlusion area based on the attachment distribution reconstruction data, identify the missing features through a feature matching algorithm, and use the missing features to compensate and reconstruct the feature loss locations to obtain complete attachment distribution information.

[0054] According to the distribution of the attached reconstruction data and the coverage area information, the depth difference of adjacent grids is calculated, and if the difference exceeds the preset concave-convex threshold, it is marked as a concave-convex part. The distribution density values of each grid and its eight adjacent grids are compared, and if the central grid density is lower than the preset proportion of the average density of the surrounding area, it is marked as a biological shelter area. The coordinates of the marked concave-convex parts and biological shelter areas are combined to obtain a set of feature loss positions. For each position in the set of feature loss positions, the texture density, direction consistency and gray contrast values of the non-loss area around the position are extracted from the attached distribution reconstruction data. The rate of change of these values along the space is calculated to obtain the texture change pattern. The points with a value of zero or deviating from the average value of the eight adjacent grids by more than a preset deviation threshold in the depth data are recorded as depth missing points. The gradient operator is used to detect the abrupt change position of the density value and connect it into a shelter area boundary line. According to the texture change pattern, the difference between the texture values on both sides of the missing position and the similarity of the pattern are calculated, and the highest similarity change pattern is selected to estimate the texture density, direction consistency and gray contrast values of the missing position. The effective adjacent depth values of the depth missing points are summed in inverse proportion to the distance to fill in the missing points. According to the tangent direction of each point on the shelter area boundary line, the boundary direction is extended by cubic spline interpolation. The estimated texture density, direction consistency, gray contrast value and filled depth value are updated to the grid unit corresponding to the feature loss position, and the extended boundary coordinates are added to the coverage area contour sequence. The updated distribution density and thickness estimation value are recalculated, and the complete attached distribution information containing the compensated distribution density, complete coverage area contour and continuous thickness distribution is output.

[0055] Specifically, the identification of concave-convex parts is based on the local variation characteristics of depth values. The surface of underwater structures is subjected to long-term water flow erosion and growth of attached organisms, forming a complex micro-topography. The depth difference between adjacent grids directly reflects the degree of surface undulation. When the depth difference exceeds the preset threshold of 5 millimeters, it indicates that there is a significant depression or protrusion at this position. This concave-convex feature often leads to multi-path reflection of sonar signals, making it difficult to extract features in this area. By marking these positions, accurate spatial positioning is provided for subsequent feature compensation.

[0056] It should be noted that the discrimination of biological shelter area uses a relative density comparison method. Marine organisms such as sea anemones and soft corals, which are flexible attachments, have umbrella-shaped or fan-shaped structures that can block other attachments below. If the distribution density value of the central grid is only 30% of the average density of the surrounding eight adjacent grids, it indicates that the position is likely to be blocked by the upper layer of organisms. This density mutation is different from the natural density gradient and has obvious boundary characteristics. The coordinates of the concave-convex parts and the biological shelter areas are combined to form a set of feature loss positions that require special processing.

[0057] Specifically, the extraction process of the texture change pattern involves spatial gradient calculation. The texture density values extracted from the non-lost region are arranged in a sequence along the direction from the lost position to the far position. The difference values of adjacent values are calculated, and the regularity of the difference values reflects the spatial continuity of the texture. Normally, the change of the texture density is gradual, and the difference value sequence presents a smooth transition. The detection of the depth missing point uses a statistical outlier identification method. When the depth value of a point is zero or differs from the average value of the eight surrounding points by more than 50% of the average value, it is determined to be an outlier. The application of the gradient operator on the density image can accurately locate the mutation edge of the density value, and these edge points are connected to form the outline of the occluded region.

[0058] In a possible implementation, the estimation of the texture value utilizes the spatial continuity principle. When the texture values on both sides of the lost position are matched with the change pattern, the similarity is calculated based on the mean square error of the difference sequence. The pattern with the highest similarity represents the most likely texture evolution rule of the region. When filling the depth missing point, the weight is inversely proportional to the square of the distance, ensuring that the influence of the neighboring points is greater. The application of the cubic spline interpolation in the boundary extension ensures the smoothness of the boundary. By constructing a cubic polynomial curve through the position and tangent direction of the known boundary point, the extended part is naturally connected with the original boundary. The integration process of the compensated data realizes the unification of local repair and global update. After updating the estimated and filled values to the corresponding grid, the statistical indicators of the affected area need to be recalculated. The recalculation of the distribution density takes into account the number of newly added valid grids, and the coverage area outline is formed by adding the extended boundary coordinates to form a closed curve. The continuity of the thickness distribution is restored, and the data void caused by the loss of features is eliminated. The complete attachment distribution information contains all the compensated feature data, providing accurate and comprehensive data support for subsequent attachment cleaning and maintenance.

[0059] After obtaining the feature missing position, the boundary of the lost region caused by biological occlusion is divided to obtain the range of the occluded region, the missing feature information matched with the lost region is extracted from the pre-established feature information database, the data content required for compensation is determined, the extracted missing feature information is matched and fused with the lost region, and the feature missing position is compensated and reconstructed according to the determined data content required for compensation, to obtain the reconstructed surface feature distribution.

[0060] After obtaining the position coordinates of the missing feature, the point where the rate of change of the feature value around the position exceeds the preset mutation threshold is identified as an edge point. Starting from any edge point, the nearest unvisited edge point in the clockwise direction is searched and connected, and the process is repeated until the starting point is returned to form a closed curve. The coordinates of all grid points inside the curve are recorded to obtain the range of the region obscured by the biological barrier. The type of the attached object, the corresponding texture density value sequence, the depth distribution value sequence, and the reflection intensity value sequence are read from the pre-established feature information database. The Euclidean distance between the measured feature value of the edge grid of the region and each type of feature value sequence in the database is calculated, and the feature value sequence corresponding to the type with the smallest distance is selected as the matched missing feature information. Based on the texture density value sequence, the depth distribution value sequence, and the reflection intensity value sequence in the missing feature information, and in combination with the measured values of the four corner points of the obscured region, the horizontal and vertical coordinates of each grid point in the region are normalized. The texture density interpolation, depth interpolation, and reflection intensity interpolation of the grid point are calculated through linear interpolation in the horizontal and vertical directions to obtain the compensation data content of each grid point. The texture density interpolation, depth interpolation, and reflection intensity interpolation of each grid point in the compensation data content are updated to the corresponding positions of the original feature distribution data, the data of the non-missing region remains unchanged, all feature data are rearranged in the order of grid coordinates, and the reconstructed surface feature distribution containing the mapping relationship between complete grid coordinates and corresponding feature values is output.

[0061] Specifically, the identification of edge points is based on the mutation detection principle of feature value change rate. In underwater environment, biological obstruction usually causes sharp change of feature value. The texture density of normal region may be 80 feature points per square centimeter, while the region obscured by sea anemone suddenly decreases to less than 10. When the change rate exceeds the preset threshold such as 70%, the position is marked as an edge point. The clockwise search strategy ensures the uniqueness and continuity of the boundary. Starting from the starting edge point, the unvisited point with the smallest angle and the nearest distance is selected each time to avoid boundary crossing and breaking phenomenon.

[0062] It should be noted that the construction of the feature information database is based on the accumulation of a large amount of historical detection data. Each type of attached object in the database corresponds to a set of standard feature sequences. The texture density value sequence of barnacle type attached object presents high density distribution, usually fluctuating between 60-100; while the texture density value sequence of soft seaweed changes in the range of 20-40. The depth distribution value sequence reflects the three-dimensional morphological characteristics of the attached object, and the reflection intensity value sequence is related to the material property. These sequence data are stored in order of spatial position to form a feature template for matching.

[0063] Specifically, the calculation of the Euclidean distance provides a quantitative indicator of the similarity of multi-dimensional features. For the measured feature value vector of the edge grid and the standard feature value vector in the database, the square sum of the difference of the corresponding dimensions is calculated and then the square root is taken. The smaller the distance, the higher the similarity. When the texture density of a certain edge grid is 45, the depth value is 12 mm, and the reflection intensity is 0.6, it is closest to the standard features of the shellfish attachments in the database, and it is determined that the shellfish attachments are very likely under the occlusion area.

[0064] In a possible implementation, the bilinear interpolation fully utilizes the spatial continuity assumption. The normalization process maps the irregular occlusion area to a standard unit square, facilitating interpolation calculation. For any point in the area, first, linear interpolation is performed on the upper and lower boundaries according to the normalized horizontal coordinates of the point, to obtain two intermediate values, and then the two intermediate values are subjected to a second linear interpolation according to the vertical coordinates. This method ensures smooth transition of the interpolation result and avoids sudden changes in the feature values. When the texture densities of the four corner points are 30, 35, 40, and 45 respectively, the interpolation result of the center point is about 37.5, which conforms to the gradual change rule. The reconstructed surface feature distribution realizes the recovery of data integrity. The position index method is adopted in the updating process, ensuring that each compensation value is accurately corresponded to the missing position. The original data of the non-missing area remains unchanged, maintaining the authenticity of the measured data. The rearrangement is performed in the grid order from left to right and from top to bottom, forming a standard data structure. In the output mapping relationship table, each grid coordinate corresponds to a complete feature triplet, including the texture density, the depth value, and the reflection intensity, eliminating the data void caused by the occlusion, and providing a complete and reliable data basis for subsequent attachment analysis and processing.

[0065] In step S107, according to the complete attachment distribution information, the attachment coverage area ratio and the distribution density obtained by combining the attachment spatial distribution model and the attachment number per unit area, and the hardness classification processing result, the underwater structure biological attachment grade evaluation result is obtained.

[0066] According to the coverage area contour data in the complete attachment distribution information, the total coverage area of the attachments is calculated by multiplying the number of grids inside the contour by the area of a single grid, the coverage area ratio is obtained by dividing the total coverage area by the total area of the structure detection region, the number of individual attachments is determined by identifying the number of texture density peak points in the coverage area, and the distribution density value is obtained by dividing the number of individual attachments by the total coverage area. The hardness identification data of each grid is extracted from the complete attachment distribution information, the total area of high-hardness identification grids is calculated, and the high-hardness proportion is obtained by dividing the total coverage area of the attachments. The total area of low-hardness identification grids is calculated, and the low-hardness proportion is obtained by dividing the total coverage area of the attachments. If the high-hardness proportion exceeds the preset hardness proportion threshold, the hardness risk value is recorded as the numerical value of the high-hardness proportion. The coverage area ratio, distribution density value, and hardness risk value are normalized respectively, and according to the preset area weight, density weight, and hardness weight, the biological attachment comprehensive score is calculated by multiplying the coverage area ratio by the area weight, adding the distribution density value multiplied by the density weight, and adding the hardness risk value multiplied by the hardness weight. According to the biological attachment comprehensive score, the attachment level is determined. If the score is lower than the preset mild threshold, the mild attachment level identification and the corresponding score value are output. If the score is higher than the preset severe threshold, the severe attachment level identification and the corresponding score value are output. If the score is between the two thresholds, the moderate attachment level identification and the corresponding score value are output. The underwater structure biological attachment level evaluation result containing the level identification and the score value is generated.

[0067] Specifically, the calculation of the coverage area ratio provides a direct quantitative indicator of the degree of attachment pollution. The number of grids inside the contour is calculated using the seed filling algorithm principle, which starts from any point inside the contour and spreads to the surrounding area to mark all internal grids, avoiding the calculation difficulties caused by complex contour shapes. When the total area of the detection region is 100 square meters and the attachments cover 35 square meters, the coverage area ratio is 0.35, indicating that more than one-third of the structure surface has been occupied by attachments. This ratio calculation provides a key basis for subsequent maintenance decisions.

[0068] It should be noted that the identification of texture density peak points is based on the principle of local maximum value detection. In the texture density distribution map, each attachment individual usually corresponds to a local high-density area, and the density value at the center position is higher than that of the surrounding eight neighborhood points. By traversing all grid points, the points that satisfy the local maximum value condition are found, and the number of attachment individuals is determined. A typical dense area of barnacles may contain 200 peak points within 10 square meters, and the calculated distribution density is 20 per square meter, reflecting the degree of attachment aggregation.

[0069] Specifically, the statistics of the hardness identification data reveal the material composition characteristics of the attachments. High-hardness attachments such as shellfish, barnacles, etc. contain a large amount of calcareous components, which are difficult to clean and have a high risk of damaging the structure surface. When the high-hardness proportion reaches 0.7, it means that 70% of the attachments need to be cleaned with high-strength cleaning methods. Low-hardness attachments such as seaweed, mollusks, etc. are relatively easy to remove. The hardness risk value is directly equal to the high-hardness proportion, which provides an important reference for subsequent cleaning scheme development. This classification statistical method ensures that different types of attachments are treated in a targeted manner.

[0070] In one possible implementation, normalization processing and weight calculation realize the comprehensive evaluation of multi-dimensional indicators. Normalization maps indicators of different dimensions to a unified interval of 0 to 1, eliminating the influence of numerical size. The weight setting reflects the relative importance of each indicator, and the area weight is usually set to 0.4, the density weight is set to 0.3, and the hardness weight is set to 0.3. When the normalized coverage area proportion of a certain structure is 0.6, the distribution density value is 0.8, and the hardness risk value is 0.5, the comprehensive score is calculated as 0.6x0.4+0.8x0.3+0.5x0.3=0.63. The generation of the level evaluation result adopts a segmented judgment method. The mild attachment threshold is usually set to 0.3, and the heavy attachment threshold is set to 0.7. The score below 0.3 indicates that the attachment has a small impact, and regular maintenance can be used; the score between 0.3 and 0.7 belongs to moderate attachment, which needs to be regularly monitored and moderately cleaned; and the score exceeding 0.7 indicates that the attachment is serious and professional cleaning measures need to be taken immediately. The output result includes a clear level identifier and a specific score value, such as "heavy attachment level, score 0.82", which provides a clear decision basis for the management department, realizing the change from qualitative judgment to quantitative evaluation.

[0071] The above is only a specific embodiment of the present specification, and those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, module and unit can refer to the corresponding process in the foregoing method embodiment, which will not be described here. It should be understood that the protection scope of the present specification is not limited to this. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed in the present specification, and these modifications or replacements should be covered within the protection scope of the present specification.

Claims

1. A method for evaluating the biofouling level of underwater structures, characterized in that: The method comprises: Acquire underwater structure images and sonar data; perform edge detection on the underwater structure images, extract edge fuzzy areas, analyze the spatial frequency and contrast gradient of the edge fuzzy areas, and determine turbidity change areas; perform sharpening on the edge fuzzy areas, extract attachment contour features, the attachment contour features including contour boundaries and shape factors, and extract depth contour data in combination with the sonar data; analyze the attachment contour features and the depth contour data, extract texture features and depth features, and generate a comprehensive feature vector by weighted fusion of the texture features and the depth features; construct an attachment spatial distribution model based on the comprehensive feature vector and environmental disturbance data, and generate attachment distribution data; identify feature loss locations based on the attachment distribution data, compensate for the feature loss locations, and generate complete attachment distribution information; calculate the coverage area ratio and distribution density based on the complete attachment distribution information and the hardness identifier, and output an attachment level assessment result.

2. The method for evaluating the biofouling level of underwater structures according to claim 1, wherein: The performing edge detection on the underwater structure image, extracting the edge fuzzy area, analyzing the spatial frequency and contrast gradient of the edge fuzzy area, and determining the turbidity change area includes: The method includes performing grayscale conversion on the image of the underwater structure to generate a grayscale image; performing filtering on the grayscale image to generate a smoothed image; applying an edge detection operator to the smoothed image to calculate horizontal and vertical gradients to generate an edge image; selecting a neighborhood window for each edge point in the edge image, calculating the grayscale change rate within the neighborhood window as the spatial frequency, and calculating the ratio of the difference between the maximum grayscale value and the minimum grayscale value to the sum as the contrast gradient; identifying edge points below a preset threshold based on the spatial frequency and the contrast gradient, generating a connected area, calculating the spatial frequency difference between the connected area and a reference area, and determining the turbidity change area.

3. The method for evaluating the biofouling level of underwater structures according to claim 1, wherein: The sharpening process is performed on the edge fuzzy area, the attachment contour features are extracted, and the depth contour data is extracted in combination with the sonar data, including: A sharpening operator is applied to the edge blurred area to enhance high-frequency components and generate a sharpened image; the sharpened image is binarized to extract a sequence of contour boundary points, calculate the contour perimeter and area, and generate a shape factor; based on the sequence of contour boundary points, the reflected signal in the sonar data is obtained, the main reflection peak is identified, the signal delay is calculated, and depth profile data is generated, wherein the depth profile data includes a lateral position and a longitudinal depth.

4. The method for evaluating the biofouling level of underwater structures according to claim 1, wherein: The analyzing the attachment contour features and the depth contour data, extracting texture features and depth features, and generating a comprehensive feature vector by weighted fusion of the texture features and the depth features includes: The texture density and directional consistency of the attachment contour feature are extracted to generate a texture feature vector, where the texture density is the number of edge points per unit area, and the directional consistency is the ratio of edge points in the main direction; the depth gradient and reflection intensity are extracted from the depth contour data, where the depth gradient is the ratio of the depth difference between adjacent points to the distance, and the reflection intensity is the ratio of the received signal amplitude to the transmitted signal amplitude; the texture density, the directional consistency, the depth gradient, and the reflection intensity are normalized to generate a texture feature vector and a depth feature vector; and weights are assigned according to the variances of the texture feature vector and the depth feature vector to generate a comprehensive feature vector.

5. The method for evaluating the biofouling level of underwater structures according to claim 4, wherein: The step of extracting texture density and direction consistency from the attachment contour feature to generate a texture feature vector includes: For the texture image of the attachment contour feature, the grayscale difference between each pixel and the adjacent pixels is calculated, and the average value of the difference is statistically calculated as the texture density; the pixel contrast ratio of grayscale changes in the horizontal and vertical directions is calculated to generate directional consistency; and a texture feature vector is generated based on the texture density and the directional consistency.

6. The method for evaluating the biofouling level of underwater structures according to claim 1, wherein: The step of constructing an attachment spatial distribution model based on the comprehensive feature vector and the environmental disturbance data and generating attachment distribution data includes: The detection area is divided into grid cells, and the center coordinates of the grid cells and the comprehensive feature vectors are recorded; environmental disturbance data is obtained, and the environmental disturbance data includes water flow velocity and suspended matter concentration; the probability of the presence of attachments in the grid cells is calculated based on the environmental disturbance data; based on the probability of the presence of attachments, the distribution density is statistically analyzed, the high-probability grid boundaries are connected to generate a coverage area, the depth difference is calculated to generate a thickness estimate, and the attachment distribution data is output.

7. The method for evaluating the biofouling level of underwater structures according to claim 1, wherein: The step of identifying a feature loss position based on the attachment distribution data, compensating for the feature loss position, and generating complete attachment distribution information includes: Based on the depth values ​​in the attachment distribution data, concave and convex parts with depth differences exceeding a threshold are marked; the distribution density of grid cells is compared, and biological occlusion areas with a density lower than the average density of the neighborhood are marked; the texture change pattern of the concave and convex parts and the biological occlusion areas is extracted, and the texture density and directional consistency are inferred; the depth missing points of the biological occlusion areas are interpolated and filled, the coverage area outline is updated, and complete attachment distribution information is generated.

8. The method for evaluating the biofouling level of underwater structures according to claim 7, wherein: The extracting of texture change patterns from the concave-convex parts and the biological occlusion area, and estimating texture density and directional consistency, includes: For the concave-convex parts and the biological occlusion area, the change rate of the neighborhood texture value is calculated to generate a texture change pattern; based on the texture change pattern, the similarity of the texture values ​​on both sides of the lost position is calculated to infer the texture density; based on the edge point distribution of the texture change pattern, directional consistency is generated.

9. The method for evaluating the biofouling level of underwater structures according to claim 7, wherein: The interpolation filling of the depth missing points in the biological occlusion area to update the coverage area outline includes: For the depth missing points in the biological occlusion area, the neighborhood depth values ​​are extracted and the filling values ​​are calculated by weighting inversely proportional to the distance; according to the boundary points of the biological occlusion area, the tangent direction is generated and the boundary contour is extended; and the coordinate sequence of the coverage area contour is updated.

10. The method for evaluating the biofouling level of underwater structures according to claim 1, wherein: The calculation of the coverage area ratio and distribution density based on the complete attachment distribution information and hardness identification, and output of the adhesion grade evaluation result, includes: The coverage area ratio is calculated based on the coverage area outline of the complete attachment distribution information; the texture density peak points in the coverage area are counted to generate the distribution density; the high hardness grid proportion is counted based on the hardness identification; the coverage area ratio, the distribution density and the high hardness grid proportion are weighted to generate a comprehensive score; and the attachment grade identification and score value are output based on the comprehensive score.

Images