A dam block stone piling and pouring exposure rate monitoring and evaluation system and method based on AI visual recognition

Abstract

This invention discloses a monitoring and evaluation system and method for the exposure rate of dam boulders during construction and pouring, based on AI visual recognition. It relates to the field of dam monitoring and includes: a correction module for real-time acquisition of continuous image sequences of the dam construction area, synchronous detection of dust concentration and light intensity parameters in the images, and frame-by-frame pixel-level adaptive correction of the original images based on the detection results; and a completion module for extracting the edge geometric and texture features of the boulders from the corrected images, and performing segment-by-segment completion and closure processing of the occluded or blurred boulder outlines based on the inherent geometric morphology of the boulders. This invention effectively eliminates image interference caused by dust and light fluctuations in the construction scene through real-time acquisition and pixel-level adaptive correction of images of the dam construction area, fully preserves the edge details of the boulders, accurately identifies the concrete pouring area and filters out areas irrelevant to construction, and accurately calculates the boulder exposure rate by combining the calibration relationship between pixels and physical dimensions.

Description

Technical Field

[0001] This invention relates to the field of dam monitoring technology, specifically to a monitoring and evaluation system and method for the exposed rate of dam riprap stacking and pouring based on AI visual recognition. Background Technology

[0002] Rockfill concrete dams are an important type of dam in water conservancy projects, combining economy and stability. The arrangement of the riprap and the exposed riprap ratio after casting are core indicators for measuring the dam's structural density, seepage resistance, and long-term operational safety, and are also key aspects of dam construction quality control. With the development of water conservancy projects towards larger scale and greater precision, higher demands are placed on the efficiency and accuracy of dam construction quality monitoring. Machine vision technology, with its advantages of non-contact, high efficiency, and continuous monitoring, has gradually become a core research direction and application hotspot in the field of dam construction quality monitoring.

[0003] The invention patent application with application number 202610409509.4 discloses an intelligent detection method and system for riprap gradation based on AI vision. This application aims to solve the problems of "long detection cycle, high dependence on manual labor, and limited sample representativeness of traditional sieving methods, as well as the problems of incomplete imaging, severe occlusion, poor anti-interference ability, and difficulty in achieving full particle size coverage and direct engineering application of existing image recognition technology in riprap scenarios with a wide particle size range and large sample size".

[0004] However, existing machine vision monitoring technology cannot effectively solve the problems of contour segmentation distortion caused by dust, sudden changes in lighting, and overlapping and obstruction of boulders during dam construction. As a result, it is difficult to accurately quantify the exposed boulder rate in real time, and thus it is impossible to effectively control the construction process.

[0005] To this end, we propose a monitoring and evaluation system and method for the exposed rate of dam riprap and concrete pouring based on AI visual recognition. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a monitoring and evaluation system and method for the exposure rate of dam riprap stacking and pouring based on AI visual recognition, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a monitoring and evaluation system for the exposed rate of dam riprap construction and pouring based on AI visual recognition, comprising: The correction module is used to acquire continuous image sequences of the dam construction area in real time, simultaneously detect dust concentration and light intensity parameters in the images, and perform frame-by-frame pixel-level adaptive correction processing on the original images based on the detection results. The completion module is used to extract the edge geometric and texture features of the boulders from the corrected images, and perform segment-by-segment completion and closure processing on the outlines of occluded or blurred boulders based on the inherent geometric morphology of the boulders. The differentiation module is used to define the boundaries of overlapping boulders in the image based on the completed boulder outline features, assign a unique region label to each independent boulder and record its boundary coordinate information; recognition The module is used to perform semantic feature analysis on areas other than boulders in the image, identify concrete pouring areas and extract their boundaries, and filter out non-construction areas unrelated to dam construction; the calculation module is used to calculate the total area of ​​all boulder areas and the total area of ​​the pouring area based on the pre-calibrated correspondence between image pixels and actual physical dimensions, in order to further calculate the boulder exposure rate of the current construction area; the output module is used to compare the real-time calculated boulder exposure rate with the preset construction standard threshold, generate construction quality evaluation results, and output all monitoring data and evaluation results to the preset construction management and control platform in a standardized format. The correction module is interconnected with a completion module via a local area network. The completion module is interconnected with a differentiation module via a local area network. The differentiation module is interconnected with a recognition module via a local area network. The recognition module is interconnected with a calculation module via a local area network. The calculation module is interconnected with an output module via a local area network.

[0008] Furthermore, the frame-by-frame pixel-level adaptive correction processing in the correction module follows the following rules: For each pixel in the original image, the mean value of the edge gradient magnitude and the variance of brightness in its 3×3 neighborhood are calculated. Combined with the normalized value of the local dust concentration and the measured value of the local illumination intensity corresponding to the pixel, an edge-preserving joint correction factor is constructed. Based on this factor, the original pixel value is nonlinearly corrected for brightness to obtain the corrected image. The formula for calculating the edge-preserving joint correction factor is as follows: ; In the formula: For the first in the image Line number Joint correction factor for column pixels; This represents the inherent attenuation coefficient of pixel brightness due to dust. This is the normalized value of the local dust concentration corresponding to this pixel. This is a preset standard light intensity value; This is the measured value of the local illumination intensity corresponding to this pixel. This is the basic nonlinear adjustment coefficient for illumination compensation; Preserve the weighting coefficients at the edges; This is the average edge gradient magnitude within a 3×3 neighborhood of the pixel. The luminance variance within the 3×3 neighborhood of this pixel; The global maximum brightness variance of the image; The corrected pixel values ​​satisfy ,in For the first image in the original image Line number The original pixel values ​​of the column pixels.

[0009] Furthermore, the segment-by-segment completion and closure processing in the completion module follows the following rules: Extract the edge geometry features of the boulders from the corrected image to generate continuous boulder edge lines with a single pixel width; The curvature value and gradient direction angle of each edge point are calculated sequentially along the edge line of the block. When the curvature value of a certain edge point exceeds a preset curvature threshold and the sum of the changes in the gradient direction angle of that point and the adjacent edge points exceeds a preset angle threshold, that point is determined as a characteristic inflection point of the block edge. Using all feature inflection points as dividing points, the continuous stone edge line is divided into several independent feature segments, each of which is a continuous edge line segment between two adjacent feature inflection points. Based on the inherent geometric shape of the convex polygon of the block, for each missing edge segment, several cubic Bézier candidate completion curves that satisfy the convexity constraint are generated with the adjacent feature inflection points at both ends as the endpoints. Calculate the comprehensive score for each candidate completion curve. The comprehensive score is obtained by weighted summation of the second derivative continuity score and the convex hull fit score. Select the candidate completion curve with the highest comprehensive score as the final completion curve, and repeat the above process until all the stone outlines form closed curves.

[0010] Furthermore, when the differentiation module defines the boundaries of overlapping stones, it extracts the curvature distribution curves of all the stone contours after completion and locates the contour abrupt change points where the curvature value exceeds the preset curvature threshold. For each pair of overlapping stones, match the corresponding abrupt change points on their contours to generate several candidate segmentation boundaries connecting the corresponding abrupt change points; Calculate the comprehensive energy value of each candidate segmentation boundary, and select the candidate segmentation boundary with the smallest comprehensive energy value that is less than the preset energy threshold as the final segmentation boundary to complete the boundary delineation of a single block of stone. The formula for calculating the comprehensive energy of the candidate segmentation boundary is: ; In the formula: The comprehensive energy value of the candidate segmentation boundary; These are the weighting coefficients for each energy term; The curvature difference between the candidate segmentation boundary and the contours of the two stones at the connection point; The length of the candidate segmentation boundary; Using arc length as the candidate segmentation boundary The second-order curvature derivative with respect to the parameter; The area of ​​the overlapping region of the outlines of the two stones; These represent the total area of ​​the outlines of the two stones; in, All units are pixels 2 .

[0011] Furthermore, in the semantic feature parsing and non-construction area screening stages, the recognition module first converts the image from RGB color space to CIELAB color space for image areas other than the boulders, extracts the brightness and color components of each pixel, and combines the spatial coordinate information of the pixels to iteratively cluster them through preset clustering step size and clustering criteria to generate several superpixel regions of uniform size with boundaries that fit the image content. For each superpixel region, extract four types of texture features: energy, contrast, correlation, and entropy of the gray-level co-occurrence matrix, as well as the region's average gray value and edge density features; The six extracted features are compared with the corresponding features in the preset concrete pouring area feature template by cosine similarity calculation to obtain the individual matching degree of each feature. The comprehensive matching degree of the superpixel area is calculated based on the preset feature weight coefficients. When the comprehensive matching degree exceeds the preset comprehensive matching threshold, the superpixel area is initially identified as a concrete pouring area. Connectivity analysis is performed on all superpixels initially identified as concrete pouring areas. Adjacent connected components of the same category are merged, and isolated connected components with an area smaller than a preset area threshold are excluded. At the same time, dynamic motion features in the images are detected by the inter-frame difference method of consecutive multi-frame images, and regions containing dynamic motion features are screened out from the initially identified concrete pouring areas.

[0012] Furthermore, when the calculation module calculates the exposed stone rate, it calculates the visible area of ​​each individual stone based on the pre-calibrated correspondence between image pixels and actual physical size; for stones with partial occlusion, it estimates their complete three-dimensional morphological parameters based on the geometric features of their visible outline to calculate their actual surface area. The total area of ​​the stone block region is calculated by summing the actual surface areas of all the stones. The formula for calculating the stone block exposure rate is: ; In the formula: This represents the percentage of exposed boulders in the current construction area. This represents the total number of boulders within the current construction area; For the first The actual surface area of ​​each stone block; For the first The angle between the surface normal vector of the stone block and the direction of the camera's optical axis; This refers to the actual area of ​​the concrete pouring area.

[0013] Furthermore, during the operation of the calculation module, the calculation results of the continuous image sequence are simultaneously verified: Calculate the difference in the exposed stone rate of the corresponding area between the current frame image and the previous frame image, and the consistency of the exposure rate change trend between the current frame image and the adjacent multiple frames image; when the difference exceeds the preset time series threshold and the change trend is abnormal, trigger the image re-acquisition and recalculation process of the corresponding area, and mark the abnormal calculation result as invalid data.

[0014] Furthermore, during the operation phase of the output module, the real-time calculated exposed stone rate value is compared with the preset construction standard threshold range. When the exposed rate value is within the preset qualified range, a qualified evaluation result is generated; when the exposed rate value exceeds the preset qualified range, a unqualified evaluation result is generated and the deviation direction and deviation value are marked. Simultaneously, the original image, corrected image, stone outline marked image, and concrete pouring area marked image at the corresponding time are attached and output to the construction management platform along with the monitoring data and evaluation results in a preset standardized format.

[0015] On the other hand, a method for monitoring and evaluating the exposed rate of dam riprap construction and pouring based on AI visual recognition includes: Real-time acquisition of continuous images of the dam construction area, simultaneous detection of dust concentration and light intensity, and frame-by-frame correction using an edge-preserving joint correction factor to eliminate interference and retain the edge details of the boulders; extraction of boulder edge features from the corrected images, location of feature inflection points using curvature and gradient direction angle, and completion of missing edges based on the convex polygon shape of the boulders using optimal cubic Bézier curves to form closed contours; extraction of curvature distribution of the boulder contours to locate abrupt change points, matching corresponding abrupt change points for overlapping boulders and generating three types of candidate segmentation boundaries, selecting the boundary with the lowest comprehensive energy value to complete the segmentation, and assigning a unique region label to each boulder; image conversion to CIEL. Superpixels are generated in the AB color space. The pouring area is initially determined by feature similarity matching. Isolated small areas and dynamic interference areas are screened out by connected component analysis and inter-frame difference method to obtain accurate pouring boundaries. Based on the calibration relationship between pixels and physical dimensions, the complete three-dimensional surface area of ​​the stone blocks is estimated. The stone exposure rate is calculated by combining the concrete pouring area. The results are verified by inter-frame difference and change trend. Re-acquisition and recalculation are triggered when anomalies occur. The real-time exposure rate is compared with the preset construction standard to generate quality evaluation results. Deviation information is marked when the results are unqualified. All monitoring data, processed images and evaluation results are output to the construction management platform in a standardized format.

[0016] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention effectively eliminates image interference caused by dust and light fluctuations in the construction scene by real-time acquisition and pixel-level adaptive correction of images of the dam construction area. It fully preserves the edge details of the boulders, providing a clear and stable image foundation for subsequent monitoring and analysis. Based on the inherent geometric shape of the boulders, it completes the precise completion and closure of occlusion and blurred contours, clearly defines the independent boundaries of overlapping boulders, accurately identifies the concrete pouring area and filters out areas irrelevant to construction, and accurately calculates the exposed rate of boulders by combining the calibration relationship between pixels and physical dimensions. Simultaneously, it conducts time-series data verification to ensure the authenticity and reliability of the calculation results. Finally, it standardizes the output of monitoring data and quality evaluation results to the construction management platform, realizing real-time, accurate, and automated monitoring and evaluation of the exposed rate of boulders in the dam construction. This improves the efficiency and accuracy of dam construction quality control, reduces the error and workload of manual monitoring, and is suitable for the complex construction environment of the dam site. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1This is a schematic diagram of a monitoring and evaluation system for the exposed rate of dam riprap and concrete pouring based on AI visual recognition. Figure 2 This is a flowchart illustrating a method for monitoring and evaluating the exposed rate of dam riprap stacking and pouring based on AI visual recognition. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] The present invention will be further described below with reference to embodiments.

[0021] Example 1:

[0022] This embodiment presents an AI-based visual recognition-based monitoring and evaluation system for the exposed rate of dam riprap construction and pouring, such as... Figure 1 As shown, it includes: The correction module is used to acquire continuous image sequences of the dam construction area in real time, simultaneously detect dust concentration and light intensity parameters in the images, and perform frame-by-frame pixel-level adaptive correction processing on the original images based on the detection results. The frame-by-frame pixel-level adaptive correction processing in the correction module follows the following rules: For each pixel in the original image, the mean value of the edge gradient magnitude and the variance of brightness in its 3×3 neighborhood are calculated. Combined with the normalized value of the local dust concentration and the measured value of the local illumination intensity corresponding to the pixel, an edge-preserving joint correction factor is constructed. Based on this factor, the original pixel value is nonlinearly luminance corrected to obtain a corrected image that eliminates dust and illumination interference while preserving the edge details of the stones. The formula for calculating the edge-preserving joint correction factor is as follows: ; In the formula: For the first in the image Line number Joint correction factor for column pixels; The inherent attenuation coefficient of pixel brightness due to dust is determined by standard dust test. This is the normalized value of the local dust concentration corresponding to this pixel (dimensionless, value range [0,1]). This is a preset standard light intensity value; This is the measured value of the local illumination intensity corresponding to this pixel. This is the basic nonlinear adjustment coefficient for illumination compensation; Preserve the weighting coefficients at the edges; This is the average edge gradient magnitude within a 3×3 neighborhood of the pixel. The luminance variance within the 3×3 neighborhood of this pixel; The global maximum brightness variance of the image; The above formula combines the mean of the edge gradient magnitude and the brightness variance of the pixel neighborhood, and simultaneously incorporates the measured parameters of local dust concentration and light intensity to construct a nonlinear joint correction factor to perform brightness correction on the original pixel. This can not only eliminate image interference caused by dust and uneven lighting at the construction site, but also completely preserve the key details of the rock edge, thus meeting the pixel-level image optimization needs of the complex environment of dam construction. The corrected pixel values ​​satisfy ,in For the first image in the original image Line number The original pixel values ​​of the column pixels; Among them, the normalized value of local dust concentration The measured values ​​of local illumination intensity were obtained through calculations using a dark channel prior and an atmospheric scattering model. The data is obtained by acquiring the light intensity data of the corresponding pixel area through an area array illumination sensor that is triggered synchronously with the image acquisition device. Specifically, the calculation steps for the normalized value of local dust concentration are as follows: First, perform dark channel calculation on the original image to obtain the dark channel value of each pixel. The dark channel value is defined as the minimum value of the pixel in the red, green, and blue channels. Then, based on the atmospheric scattering model, which describes that the pixel value of the original image is equal to the pixel value of the clear image without dust multiplied by the transmittance, plus the global atmospheric light value multiplied by one minus the transmittance. The global atmospheric light value is estimated using the quadtree search method, and the pixel value with the highest brightness among the top 0.1% of the dark channel values ​​is selected as the global atmospheric light value. Then, the transmittance is calculated, which is equal to one minus a constant of 0.95 multiplied by the dark channel value and then divided by the global atmospheric light value. Finally, the transmittance is normalized to the range of zero to one, resulting in the local dust concentration normalized value being equal to one minus the transmittance. The area array illumination sensor and the image acquisition device adopt a hardware synchronous triggering method. Image acquisition and illumination data acquisition are started simultaneously through the same external trigger signal. The frequency of the trigger signal is consistent with the frame rate of the image acquisition device, and the time synchronization accuracy is controlled within one millisecond. The pixel resolution of the area array illumination sensor is the same as that of the image acquisition device, and their field of view completely overlaps. A one-to-one correspondence between the pixels of the illumination sensor and the image pixels is achieved through a pre-calibrated pixel mapping table. The pixel mapping table is calibrated before system deployment using a checkerboard calibration method. ∈[0.5, 2], its value is directly proportional to the amplitude of light fluctuation in the construction area and inversely proportional to the dynamic range of the image acquisition device. The larger the amplitude of light fluctuation and the smaller the dynamic range of the device, the larger the value. ∈[0.1, 1], its value is directly proportional to the required edge retention strength of the stone block and inversely proportional to the average particle size of the stone block. The higher the edge retention requirement and the smaller the average particle size of the stone block, the larger the value. The calibration steps for the standard dust experiment of the inherent attenuation coefficient of pixel brightness caused by dust are as follows: Set up standard dust environments of different concentrations in a closed experimental chamber, collect comparative images with and without dust under the same illumination conditions, calculate the brightness attenuation rate of each pixel, and obtain the inherent attenuation coefficient of pixel brightness caused by dust through linear regression fitting; the calibration method for the basic nonlinear adjustment coefficient of illumination compensation is as follows: collect standard stone images under different illumination intensities, use the manually annotated stone edge clarity as the evaluation index, determine the optimal value through a grid search method, and set the initial value of this coefficient to 0.0; the calibration method for the edge preservation weight coefficient is as follows: collect images of stones with different average particle sizes, use the edge preservation rate as the evaluation index, determine the optimal value through a grid search method, and set the initial value of this coefficient to 0.5. The completion module is used to extract the edge geometric features and texture features of the stones from the corrected image. Based on the inherent geometric shape of the stones, it performs segment-by-segment completion and closure processing on the outline of the occluded or blurred stones. In the completion module, segment-by-segment completion and closure processing follow the following rules: Extract the edge geometry features of the boulders from the corrected image to generate continuous boulder edge lines with a single pixel width; The curvature value and gradient direction angle of each edge point are calculated sequentially along the edge line of the stone. When the curvature value of a certain edge point exceeds the preset curvature threshold and the sum of the changes in the gradient direction angle of that point and the adjacent edge points exceeds the preset angle threshold, the point is determined to be the feature inflection point of the stone edge. The preset curvature threshold and preset angle threshold are obtained by calibration using a standard block sample set. The standard block sample set contains one thousand images of blocks with different particle sizes and shapes, and each image is marked with the actual feature inflection point position. By calculating the curvature value and gradient direction angle change of all marked feature inflection points, the 95th percentile of the curvature value is taken as the preset curvature threshold, and the 90th percentile of the sum of gradient direction angle changes is taken as the preset angle threshold. The initial value of the preset curvature threshold is 0.5 per pixel, and the initial value of the preset angle threshold is 30 degrees. Using all feature inflection points as dividing points, the continuous stone edge line is divided into several independent feature segments. Each independent feature segment is a continuous edge line segment between two adjacent feature inflection points. Its length is determined by the position of the corresponding two feature inflection points, without equal length constraints. Based on the inherent geometric shape of the convex polygon of the block, for each missing edge segment, several cubic Bézier candidate completion curves that satisfy the convexity constraint are generated with the adjacent feature inflection points at both ends as the endpoints. Calculate the overall score for each candidate completion curve. The overall score is obtained by weighted summation of the second derivative continuity score and the convex hull fit score. Specifically, the second derivative continuity score is calculated as follows: divide by one and add the absolute value of the difference between the second derivative of the candidate completion curve at the starting point and the second derivative of the previous adjacent feature segment at the ending point, and add the absolute value of the difference between the second derivative of the candidate completion curve at the ending point and the second derivative of the next adjacent feature segment at the starting point. The convex hull fit score is calculated as follows: the overlap length between the candidate completion curve and the minimum convex hull of the corresponding block is divided by the total length of the candidate completion curve; the comprehensive score is equal to 0.6 times the second derivative continuity score plus 0.4 times the convex hull fit score. Among them, the second derivative continuity score is calculated by normalizing the inverse of the difference between the second derivatives of the candidate completion curve and the adjacent feature segment at the connection point, and the convex hull fit score is calculated by the ratio of the overlap length of the candidate completion curve and the minimum convex hull of the corresponding block to the total length of the candidate completion curve. Select the candidate completion curve with the highest comprehensive score as the final completion curve, and repeat the above process until all the stone outlines form closed curves. The differentiation module is used to define the boundaries of overlapping blocks in the image based on the completed block outline features, assign a unique region label to each independent block and record its boundary coordinate information. When the differentiation module defines the boundaries of overlapping stones, it extracts the curvature distribution curves of all the stone contours after completion and locates the contour abrupt change points where the curvature value exceeds the preset curvature threshold. For each pair of overlapping stones, match the corresponding abrupt change points on their contours to generate several candidate segmentation boundaries connecting the corresponding abrupt change points; The method for determining whether two stones overlap is as follows: calculate the intersection-union ratio (IUR) of the minimum bounding rectangles of the two stone outlines. When the IUR is greater than 0.1, the two stones are considered to overlap. The matching method for corresponding abrupt change points is as follows: for two stones that overlap, extract all abrupt change points on their outlines respectively, calculate the Euclidean distance from each abrupt change point in the first stone to all abrupt change points in the second stone, and select the point pair with the smallest distance that is less than a preset distance threshold as the corresponding abrupt change point. The initial value of the preset distance threshold is 10 pixels. If there are multiple point pairs that meet the conditions, select the point pair with the smallest curvature difference as the final corresponding abrupt change point. Calculate the comprehensive energy value of each candidate segmentation boundary, and select the candidate segmentation boundary with the smallest comprehensive energy value that is less than the preset energy threshold as the final segmentation boundary to complete the boundary delineation of a single block of stone. The formula for calculating the comprehensive energy of the candidate split boundary is: ; In the formula: The comprehensive energy value of the candidate segmentation boundary; These are the weighting coefficients for each energy term; The curvature difference between the candidate segmentation boundary and the contours of the two stones at the connection point; The length of the candidate segmentation boundary; Using arc length as the candidate segmentation boundary The second-order curvature derivative with respect to the parameter; The area of ​​the overlapping region of the outlines of the two stones; These represent the total area of ​​the outlines of the two stones; The above formula integrates three indicators—curvature difference, second-order curvature derivative integral, and percentage of overlapping area of ​​stones—to calculate the comprehensive energy, thereby selecting the optimal segmentation boundary. The weighting coefficients can also be adaptively adjusted according to the size of the stones and the overlap rate, which can accurately define the boundary of overlapping stones and adapt to the stone segmentation needs under different construction conditions. in, All units are pixels 2 ; When generating several candidate segmentation boundaries connecting corresponding mutation points, for each pair of matching corresponding mutation points, the minimum convex hull of the overlapping region of the two stone contours is used as the constraint range to generate three types of candidate segmentation boundaries: the first type is a straight line segment connecting the two points; the second type is a quadratic Bézier curve with the two points as endpoints and the control points taken from the midpoints of the edges of the two stone contours in the overlapping region; the third type is a cubic Bézier curve with the two points as endpoints and the two control points taken from the edge points on the two stone contours in the overlapping region within a preset pixel range from the two points; all candidate segmentation boundaries must satisfy the constraint condition of being completely contained within the overlapping region of the two stone contours. Among them, the first type of candidate segmentation boundary is a straight line segment connecting two corresponding mutation points; The second type of candidate segmentation boundary is a quadratic Bézier curve with two corresponding mutation points as endpoints and the control points are the midpoints of the two stone contour edges in the overlapping area. The midpoints of the two stone contour edges are defined as the average coordinates of the edge points of the two stone contours in the overlapping area. The third type of candidate segmentation boundary is defined by taking two corresponding mutation points as endpoints, and taking two control points from the edge points of the two stone contours within the overlapping area, which are within a range of five to fifteen pixels from the corresponding mutation points. The angle between the line connecting the control point and the corresponding mutation point and the tangent direction of the stone contour at that point is less than 45 degrees. The values ​​were determined through a pre-set calibration experiment: standard dam construction sample images containing different degrees of overlap and different block sizes were collected. Using manually annotated real block segmentation boundaries as a benchmark, the initial optimal values ​​of each weight coefficient were obtained by minimizing the Pearson correlation coefficient between the comprehensive energy value and the segmentation error. All are positive numbers, and their sum is 1. In the actual monitoring process, the system adaptively fine-tunes the weight coefficient within the preset coefficient adjustment range based on the average size of the boulders in the current construction area and the real-time overlap rate. The recognition module is used to perform semantic feature analysis on areas other than the boulders in the image, identify the concrete pouring area and extract its boundary, and screen out non-construction areas that are not related to the dam construction. In the semantic feature parsing and non-construction area screening stages, the recognition module first converts the image from RGB color space to CIELAB color space for image areas other than the boulders, extracts the brightness and color components of each pixel, and combines the spatial coordinate information of the pixels to iteratively cluster them through preset clustering step size and clustering criteria to generate several superpixel regions of uniform size with boundaries that fit the image content. Specifically, superpixel generation employs a simple linear iterative clustering algorithm with a preset clustering step size of fifty pixels, meaning the initial distance between cluster centers is fifty pixels. The clustering criterion is to minimize the weighted sum of the color distance and spatial distance from each pixel to its cluster center, where the weight of the color distance is 0.6 and the weight of the spatial distance is 0.4. The iteration count is ten, and the final number of superpixels generated is approximately the total number of pixels in the image divided by the square of the clustering step size. For each superpixel region, extract four types of texture features: energy, contrast, correlation, and entropy of the gray-level co-occurrence matrix, as well as the region's average gray value and edge density features; The six extracted features are compared with the corresponding features in the preset concrete pouring area feature template by cosine similarity calculation to obtain the individual matching degree of each feature. The comprehensive matching degree of the superpixel area is calculated based on the preset feature weight coefficients. The feature weight coefficients are initially set to be equal and are adjusted by the system user. When the comprehensive matching degree exceeds the preset comprehensive matching threshold, the superpixel area is initially identified as a concrete pouring area. Specifically, the method for constructing the pre-defined feature template for the concrete pouring area is as follows: First, two hundred images of the concrete pouring area were collected under different construction stages and lighting conditions, and the boundaries of the concrete pouring area were marked on each image. Then, four types of texture features—energy, contrast, correlation, and entropy—were extracted from the gray-level co-occurrence matrix of each marked concrete pouring area, as well as the average gray value and edge density features of the area. Finally, the mean and standard deviation of the six types of features of all marked areas were calculated. The mean was used as the feature value of the feature template, and three times the standard deviation was used as the tolerance range for feature matching. The feature template was updated every three months, and newly collected concrete pouring area image data were added during the update. Connectivity analysis is performed on all superpixels initially identified as concrete pouring areas. Adjacent connected components of the same category are merged, and isolated connected components with an area smaller than a preset area threshold are excluded. At the same time, dynamic motion features in the images are detected by the inter-frame difference method of consecutive multi-frame images. Dynamic motion features are the set of pixels whose gray value changes exceed a preset gray value change threshold in consecutive multi-frame images. These correspond to the areas where moving objects such as construction workers and machinery are located. Areas containing the above dynamic motion features are screened out from the initially identified concrete pouring areas, and finally the accurate boundary of the concrete pouring area is obtained. The calculation module is used to calculate the total area of ​​all boulders and the total area of ​​the pouring area based on the pre-calibrated correspondence between image pixels and actual physical dimensions, so as to further calculate the exposed boulders rate of the current construction area. When calculating the exposed stone rate, the calculation module calculates the visible area of ​​each individual stone based on the pre-calibrated correspondence between image pixels and actual physical size; for stones with partial occlusion, it estimates their complete three-dimensional morphological parameters based on the geometric features of their visible outline in order to calculate their actual surface area. The total area of ​​the stone block region is calculated by summing the actual surface areas of all the stones. The formula for calculating the stone block exposure rate is: ; In the formula: This represents the percentage of exposed boulders in the current construction area. This represents the total number of boulders within the current construction area; For the first The actual surface area of ​​each stone block; For the first The angle between the surface normal vector of the stone block and the direction of the camera's optical axis; This refers to the actual area of ​​the concrete pouring area. The above formula combines the actual surface area of ​​the boulders, the angle between the surface normal vector and the camera optical axis, and the concrete pouring area to accurately calculate the exposed boulders rate. The surface area of ​​the boulders is obtained by fitting the triaxial geometric parameters, which can complete the estimation of the complete shape of the obscured boulders, making the exposed boulders rate calculation result fit the actual physical state of the dam construction, and the data is more realistic and has more reference value. in, ; This formula is based on the major and minor semi-axes of the visible outline of the stone block and the fitted third semi-axe. It calculates the actual surface area of ​​the stone block by power weighting and averaging the three-axis geometric parameters. It can accurately restore the three-dimensional shape of the stone block based on the two-dimensional visible outline, effectively solving the problem that the complete surface area cannot be directly measured when the stone block is obscured in the construction scene. The calculation results are consistent with the real geometric characteristics of the stone block, providing data support for the calculation of the exposure rate. In the formula: For the first The actual surface area of ​​each stone block; For the first The long semi-axis of the visible outline of each stone; For the first The short semi-axis of the visible outline of each stone block; For the first The third half-axis estimated by the block stone; Among them, the semi-major axis of the visible contour short half shaft The third semi-axis is determined by half the length and width of the smallest bounding rectangle of the visible outline of the stone. The ratio of the perimeter to the area of ​​the visible contour is determined by a preset geometric mapping relationship, which is trained from a large number of measured data of standard stone samples. The preset geometric mapping relationship of the third half-axis of the stone adopts a quadratic polynomial regression model, which is in the form that the third half-axis is equal to the coefficient a multiplied by the square of the ratio of perimeter to area, plus the coefficient b multiplied by the ratio of perimeter to area, plus the coefficient c, where P is the perimeter of the visible outline of the stone and A is the area of ​​the visible outline of the stone. The regression coefficients were obtained through training with measured data from one thousand standard stones. The size of the standard stones ranged from 0.5 meters to 3 meters. The measured data included the visible perimeter, area, and actual three-dimensional dimensions of each stone. The regression coefficients obtained from the training were 0.02, 0.3, and 0.1, all in meters. During the computation module's execution phase, the computation results for the continuous image sequence are simultaneously verified: Calculate the difference in the exposed stone rate of the corresponding area between the current frame image and the previous frame image, and the consistency of the exposure rate change trend between the current frame image and multiple adjacent frames; when the difference exceeds the preset time threshold and the change trend is abnormal, trigger the image re-acquisition and recalculation process of the corresponding area, and mark the abnormal calculation result as invalid data; The output module is used to compare the real-time calculated exposed stone rate with the preset construction standard threshold, generate construction quality evaluation results, and output all monitoring data and evaluation results to the preset construction management and control platform in a standardized format. During the output module's operation phase, the real-time calculated exposed stone rate value is compared with the preset construction standard threshold range. When the exposed rate value is within the preset qualified range, a qualified evaluation result is generated; when the exposed rate value exceeds the preset qualified range, an unqualified evaluation result is generated and the deviation direction and deviation value are marked. Simultaneously, the original image, corrected image, stone outline mark image, and concrete pouring area mark image at the corresponding time are attached and output to the construction control platform along with the monitoring data and evaluation results in a preset standardized format. The correction module is interconnected with the completion module via a local area network. The completion module is interconnected with the differentiation module via a local area network. The differentiation module is interconnected with the recognition module via a local area network. The recognition module is interconnected with the calculation module via a local area network. The calculation module is interconnected with the output module via a local area network.

[0023] In this embodiment, the correction module acquires a continuous sequence of images of the dam construction area in real time, simultaneously detecting dust concentration and light intensity parameters in the images. Based on the detection results, it performs frame-by-frame pixel-level adaptive correction processing on the original images. Simultaneously, the completion module extracts the edge geometric and texture features of the boulders from the corrected images. Based on the inherent geometric shape of the boulders, it performs segment-by-segment completion and closure processing on the outlines of occluded or blurred boulders. The differentiation module, running subsequently, defines the boundaries of overlapping boulders in the image based on the completed boulder outline features, assigning a unique region marker to each independent boulder and recording its boundary coordinate information. The recognition module then performs semantic feature analysis on the areas outside the boulders in the image, identifies the concrete pouring area and extracts its boundary, and filters out non-construction areas unrelated to dam construction. The calculation module further calculates the total area of ​​all boulder areas and the total area of ​​the pouring area based on the pre-calibrated correspondence between image pixels and actual physical dimensions, in order to further calculate the boulder exposure rate of the current construction area. Finally, the output module compares the real-time calculated boulder exposure rate with the preset construction standard threshold to generate the construction quality evaluation result, and outputs all monitoring data and evaluation results to the preset construction control platform in a standardized format.

[0024] In the above embodiments, when the system is applied to the monitoring of dam construction sites, it can effectively eliminate image acquisition interference caused by dust and uneven lighting, accurately restore the complete outline of the boulders, clearly define the boundaries of overlapping boulders, quickly identify the concrete pouring area and screen out areas that are irrelevant to construction, accurately calculate the exposed boulder rate, complete quality evaluation in real time by comparing with construction standards, and simultaneously upload complete monitoring data. Moreover, the entire monitoring and evaluation process is automated, which effectively improves the quality control efficiency of dam boulder stacking and pouring construction, reduces manual monitoring errors and workload, and is suitable for the complex construction environment of dam sites.

[0025] The system is used in the following ways: The system was applied to the construction phase of a rockfill concrete dam in a medium-sized water conservancy project in a certain river basin. The main body of the dam is a combination of rockfill and concrete pouring. This monitoring selected the upstream rockfill area and the middle pouring section of the dam as the core monitoring area, with an actual construction area of ​​approximately 1200㎡. The on-site conditions in this construction section are complex, with problems such as construction dust dispersion, significant fluctuations in daytime sunlight intensity, natural stacking of boulders obstructing each other, the intermingling of concrete pouring areas and boulder areas, and interference from the dynamic operations of construction personnel and machinery. Traditional manual monitoring methods have drawbacks such as low efficiency, poor accuracy, and data lag. Therefore, this system was adopted to conduct real-time monitoring and construction quality evaluation of the boulder stacking and pouring exposure rate of the dam body.

[0026] The on-site deployment consists of a high-definition industrial image acquisition device, an area array light sensor, and a dust concentration detection sensor, forming a front-end acquisition unit. The device is fixed on a stable support around the construction area and transmits data in real time to the industrial control computer equipped with this system at the back end. The industrial control computer is synchronously connected to the engineering construction management and control platform, forming a complete on-site monitoring-data processing-result output link.

[0027] After the system is started, the front-end acquisition unit acquires a continuous sequence of images of the dam construction area in real time, and simultaneously acquires the dust concentration and light intensity parameters of the corresponding areas of the images. The correction module immediately performs frame-by-frame pixel-level adaptive correction processing on the original images, and completes nonlinear brightness correction through joint correction factors, effectively eliminating the interference of dust fogging and uneven lighting on the images, while fully preserving the edge details of the boulders. After correction, the image clarity meets the requirements of subsequent recognition and analysis, and the edge contours of the boulders are undistorted and unblurred.

[0028] The completion module extracts the features of the stones in the corrected image, accurately acquiring the geometric and textural features of the stone edges. Addressing issues such as missing or broken stone outlines caused by stacking or occlusion by construction equipment, it identifies edge feature inflection points and divides them into independent feature segments based on the inherent geometric shape of the convex polygons of the stones. It then generates optimal completion curves for the missing outline segments, completing the closure of all stone outlines. After completion, all stones within the construction area form complete closed outlines, with the outline shape perfectly matching the actual physical shape of the stones, without any missing or misaligned outlines.

[0029] The differentiation module addresses the issue of numerous overlapping and contiguous stones within the construction area. It extracts the curvature distribution features of the completed stone outlines, locates abrupt change points, matches corresponding abrupt change points on overlapping stones, and generates candidate segmentation boundaries that meet the constraints. After calculating the comprehensive energy value of each candidate segmentation boundary, the optimal segmentation boundary is selected to define the boundary of each individual stone. Simultaneously, a unique region marker is assigned to each independent stone, and the boundary coordinate information is accurately recorded. After boundary definition, overlapping stones are accurately segmented without boundary confusion, merging of multiple stones, or mis-splitting of single stones. A total of 216 independent stones were identified within the monitored area, and the boundary coordinate data perfectly matched the actual locations of the stones on site.

[0030] The recognition module performs semantic feature analysis on areas outside the boulders in the image. First, the image is converted from the RGB color space to the CIELAB color space, generating superpixel regions whose boundaries fit the construction area. Texture, grayscale, and edge density features of each superpixel region are extracted and matched against a preset concrete pouring area feature template. After preliminary identification of the pouring area, connected component analysis is performed to eliminate small, isolated areas. Then, inter-frame difference analysis of consecutive multi-frame images is used to identify and remove dynamically moving areas such as construction workers and machinery. Finally, the boundaries of the concrete pouring area are accurately extracted, and all non-construction areas are eliminated. After recognition processing, the boundaries of the concrete pouring area are complete and clear, with no misclassifications or omissions, and all non-construction interference areas are completely removed.

[0031] The calculation module, based on the pre-defined correspondence between image pixels and actual physical dimensions, first calculates the visible area of ​​each individual stone block. For partially occluded stones, it estimates the complete three-dimensional morphological parameters based on the visible contour geometric features and obtains the actual surface area. The total area of ​​the stone block region is obtained by summing the actual surface areas of all stones. This is then combined with the actual area of ​​the concrete pouring area to calculate the stone exposure rate of the current construction area. The calculated stone exposure rate of the monitored construction area is 78.5%. Simultaneously, the system verifies the calculation results of continuous image sequences. The difference in exposure rate between the current frame and the previous frame, as well as the trend of exposure rate changes across multiple frames, are all within the normal range, with no calculation anomalies. The calculated data is marked as valid data.

[0032] The output module compares the real-time calculated 78.5% exposed boulders rate with the preset threshold range for boulders exposure rate in the dam body. If the value falls within the preset acceptable range, the system generates a construction quality qualification evaluation result. Subsequently, the system outputs all monitoring data, including the original image, corrected image, boulders outline marked image, concrete pouring area marked image, number of individual boulders, total area of ​​boulders area, total area of ​​concrete pouring area, exposed boulders rate, and construction quality evaluation result, to the construction management platform in a preset standardized format. The construction management platform receives and stores all data in real time, allowing on-site construction management personnel to directly view the monitoring results and quality evaluation, and to guide precise adjustments to on-site boulders stacking and concrete pouring operations based on the data.

[0033] Example 2:

[0034] At the implementation level, based on Example 1, this example refers to... Figure 2 A further detailed explanation is provided regarding the AI-based visual recognition-based monitoring and evaluation system for monitoring and evaluating the exposed rate of dam riprap construction and pouring in Example 1: A method for monitoring and evaluating the exposed rate of dam riprap construction and pouring based on AI visual recognition, comprising: Real-time acquisition of continuous images of the dam construction area, simultaneous detection of dust concentration and light intensity, and frame-by-frame correction using edge-preserving joint correction factors to eliminate interference and preserve the edge details of the boulders; Extract the edge features of the boulders in the corrected image, locate the feature inflection points by curvature and gradient direction angle, and fill in the missing edges with the optimal cubic Bézier curve based on the convex polygon shape of the boulders to form a closed contour. Extract the curvature distribution of the stone outline to locate abrupt change points, match the corresponding abrupt change points for overlapping stones and generate three types of candidate segmentation boundaries, select the boundary with the smallest comprehensive energy value to complete the segmentation, and assign a unique region label to each stone. The image is converted to the CIELAB color space to generate superpixels. The pouring area is initially determined by feature similarity matching. Isolated small areas and dynamic interference areas are screened out by connected component analysis and inter-frame difference method to obtain the accurate pouring boundary. Based on the calibration relationship between pixels and physical size, the complete three-dimensional surface area of ​​the stone block is estimated, the exposed stone rate is calculated by combining the concrete pouring area, and the results are verified by inter-frame difference and change trend. In case of anomalies, re-acquisition and recalculation are triggered. The real-time exposure rate is compared with the preset construction standards to generate quality evaluation results. If the results are unqualified, deviation information is marked. All monitoring data, processed images and evaluation results are output to the construction management platform in a standardized format.

[0035] In summary, the system and method described in the above embodiments effectively eliminate image interference caused by dust and light fluctuations in the construction scene by real-time acquisition and pixel-level adaptive correction of images of the dam construction area. This preserves the details of the boulders' edges, providing a clear and stable image foundation for subsequent monitoring and analysis. Based on the inherent geometric shape of the boulders, it accurately completes and closes occlusions and blurred outlines, clearly defines the independent boundaries of overlapping boulders, accurately identifies the concrete pouring area and filters out areas irrelevant to construction, accurately calculates the exposed boulder rate by combining the calibration relationship between pixels and physical dimensions, and simultaneously conducts time-series data verification to ensure the accuracy and reliability of the calculation results. Finally, the monitoring data and quality evaluation results are standardized and output to the construction management platform, achieving real-time, accurate, and automated monitoring and evaluation of the exposed boulders in the dam construction and pouring process. This improves the efficiency and accuracy of dam construction quality control, reduces errors and workload in manual monitoring, and adapts to the complex construction environment of the dam site.

[0036] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A monitoring and evaluation system for the exposed rate of dam riprap stacking and pouring based on AI visual recognition, characterized in that, include: The correction module is used to acquire continuous image sequences of the dam construction area in real time, simultaneously detect dust concentration and light intensity parameters in the images, and perform frame-by-frame pixel-level adaptive correction processing on the original images based on the detection results. The completion module is used to extract the edge geometric features and texture features of the stones from the corrected image. Based on the inherent geometric shape of the stones, it performs segment-by-segment completion and closure processing on the outline of the occluded or blurred stones. The differentiation module is used to define the boundaries of overlapping blocks in the image based on the completed block outline features, assign a unique region label to each independent block and record its boundary coordinate information. The recognition module is used to perform semantic feature analysis on areas other than the boulders in the image, identify the concrete pouring area and extract its boundary, and screen out non-construction areas that are not related to the dam construction. The calculation module is used to calculate the total area of ​​all boulders and the total area of ​​the pouring area based on the pre-calibrated correspondence between image pixels and actual physical dimensions, so as to further calculate the exposed boulders rate of the current construction area. The output module is used to compare the real-time calculated exposed stone rate with the preset construction standard threshold, generate construction quality evaluation results, and output all monitoring data and evaluation results to the preset construction management and control platform in a standardized format.

2. The monitoring and evaluation system for the exposed rate of dam riprap stacking and pouring based on AI visual recognition as described in claim 1, characterized in that, The frame-by-frame pixel-level adaptive correction processing in the correction module follows the following: For each pixel in the original image, the mean value of the edge gradient magnitude and the variance of brightness in its 3×3 neighborhood are calculated. Combined with the normalized value of the local dust concentration and the measured value of the local illumination intensity corresponding to the pixel, an edge-preserving joint correction factor is constructed. Based on this factor, the original pixel value is nonlinearly corrected for brightness to obtain the corrected image. The formula for calculating the edge-preserving joint correction factor is as follows: ； In the formula: For the first in the image Line number Joint correction factor for column pixels; This represents the inherent attenuation coefficient of pixel brightness due to dust. This is the normalized value of the local dust concentration corresponding to this pixel. This is a preset standard light intensity value; This is the measured value of the local illumination intensity corresponding to this pixel. This is the basic nonlinear adjustment coefficient for illumination compensation; Preserve the weighting coefficients at the edges; This is the average edge gradient magnitude within a 3×3 neighborhood of the pixel. The luminance variance within the 3×3 neighborhood of this pixel; The global maximum brightness variance of the image; The corrected pixel values ​​satisfy ,in For the first image in the original image Line number The original pixel values ​​of the column pixels.

3. The monitoring and evaluation system for the exposed rate of dam riprap stacking and pouring based on AI visual recognition as described in claim 1, characterized in that, The segment-by-segment completion and closure processing in the completion module follows the following rules: Extract the edge geometry features of the boulders from the corrected image to generate continuous boulder edge lines with a single pixel width; The curvature value and gradient direction angle of each edge point are calculated sequentially along the edge line of the block. When the curvature value of a certain edge point exceeds a preset curvature threshold and the sum of the changes in the gradient direction angle of that point and the adjacent edge points exceeds a preset angle threshold, that point is determined as a characteristic inflection point of the block edge. Using all feature inflection points as dividing points, the continuous stone edge line is divided into several independent feature segments, each of which is a continuous edge line segment between two adjacent feature inflection points. Based on the inherent geometric shape of the convex polygon of the block, for each missing edge segment, several cubic Bézier candidate completion curves that satisfy the convexity constraint are generated with the adjacent feature inflection points at both ends as the endpoints. Calculate the comprehensive score for each candidate completion curve. The comprehensive score is obtained by weighted summation of the second derivative continuity score and the convex hull fit score. Select the candidate completion curve with the highest comprehensive score as the final completion curve, and repeat the above process until all the stone outlines form closed curves.

4. The monitoring and evaluation system for the exposed rate of dam riprap stacking and pouring based on AI visual recognition as described in claim 1, characterized in that, When the differentiation module defines the boundaries of overlapping stones, it extracts the curvature distribution curves of all the stone contours after completion and locates the contour abrupt change points where the curvature value exceeds the preset curvature threshold. For each pair of overlapping stones, match the corresponding abrupt change points on their contours to generate several candidate segmentation boundaries connecting the corresponding abrupt change points; Calculate the comprehensive energy value of each candidate segmentation boundary, and select the candidate segmentation boundary with the smallest comprehensive energy value that is less than the preset energy threshold as the final segmentation boundary to complete the boundary delineation of a single block of stone. The formula for calculating the comprehensive energy of the candidate segmentation boundary is: ； In the formula: The comprehensive energy value of the candidate segmentation boundary; These are the weighting coefficients for each energy term; The curvature difference between the candidate segmentation boundary and the contours of the two stones at the connection point; The length of the candidate segmentation boundary; Using arc length as the candidate segmentation boundary The second-order curvature derivative with respect to the parameter; The area of ​​the overlapping region of the outlines of the two stones; These represent the total area of ​​the outlines of the two stones; in, All units are pixels 2 .

5. The monitoring and evaluation system for the exposed rate of dam riprap construction and pouring based on AI visual recognition as described in claim 1, characterized in that, In the semantic feature parsing and non-construction area screening stages, the recognition module first converts the image from RGB color space to CIELAB color space for image areas other than the boulders, extracts the brightness and color components of each pixel, and combines the spatial coordinate information of the pixels to iteratively cluster them using a preset clustering step size and clustering criteria to generate several superpixel regions of uniform size with boundaries that fit the image content. For each superpixel region, extract four types of texture features: energy, contrast, correlation, and entropy of the gray-level co-occurrence matrix, as well as the region's average gray value and edge density features; The six extracted features are compared with the corresponding features in the preset concrete pouring area feature template by cosine similarity calculation to obtain the individual matching degree of each feature. The comprehensive matching degree of the superpixel area is calculated based on the preset feature weight coefficients. When the comprehensive matching degree exceeds the preset comprehensive matching threshold, the superpixel area is initially identified as a concrete pouring area. Connectivity analysis is performed on all superpixels initially identified as concrete pouring areas. Adjacent connected components of the same category are merged, and isolated connected components with an area smaller than a preset area threshold are excluded. At the same time, dynamic motion features in the images are detected by the inter-frame difference method of consecutive multi-frame images, and regions containing dynamic motion features are screened out from the initially identified concrete pouring areas.

6. The monitoring and evaluation system for the exposed rate of dam riprap stacking and pouring based on AI visual recognition as described in claim 1, characterized in that, When calculating the exposed stone rate, the calculation module calculates the visible area of ​​each individual stone based on the pre-calibrated correspondence between image pixels and actual physical size; for stones with partial occlusion, it estimates their complete three-dimensional morphological parameters based on the geometric features of their visible outline to calculate their actual surface area. The total area of ​​the stone block region is calculated by summing the actual surface areas of all the stones. The formula for calculating the stone block exposure rate is: ； In the formula: This represents the percentage of exposed boulders in the current construction area. This represents the total number of boulders within the current construction area; For the first The actual surface area of ​​each stone block; For the first The angle between the surface normal vector of the stone block and the direction of the camera's optical axis; This refers to the actual area of ​​the concrete pouring area.

7. The monitoring and evaluation system for the exposed rate of dam riprap construction and pouring based on AI visual recognition as described in claim 1, characterized in that, During the operation of the calculation module, the calculation results of the continuous image sequence are verified synchronously: Calculate the difference in the exposed stone rate of the corresponding region between the current frame image and the previous frame image, and the consistency of the exposed stone rate change trend between the current frame image and the adjacent multiple frames image; When the difference exceeds the preset time-series threshold and the trend of change is abnormal, the image re-acquisition and recalculation process of the corresponding area is triggered, and the abnormal calculation result is marked as invalid data.

8. A monitoring and evaluation system for the exposed rate of dam riprap stacking and pouring based on AI visual recognition, as described in claim 1, is characterized in that... During the operation phase of the output module, the real-time calculated exposed stone rate value is compared with the preset construction standard threshold range. When the exposed rate value is within the preset qualified range, a qualified evaluation result is generated; when the exposed rate value exceeds the preset qualified range, an unqualified evaluation result is generated and the deviation direction and deviation value are marked. Simultaneously, the original image, corrected image, stone outline mark image, and concrete pouring area mark image at the corresponding time are attached and output to the construction management platform along with the monitoring data and evaluation results in a preset standardized format.

9. A monitoring and evaluation system for the exposed rate of dam riprap stacking and pouring based on AI visual recognition, as described in claim 1, is characterized in that... The correction module is interconnected with a completion module via a local area network. The completion module is interconnected with a differentiation module via a local area network. The differentiation module is interconnected with a recognition module via a local area network. The recognition module is interconnected with a calculation module via a local area network. The calculation module is interconnected with an output module via a local area network.

10. A method for monitoring and evaluating the exposed rate of dam riprap construction and pouring based on AI visual recognition, wherein the method is an implementation method of the monitoring and evaluation system for the exposed rate of dam riprap construction and pouring based on AI visual recognition as described in any one of claims 1-9, characterized in that... include: Real-time acquisition of continuous images of the dam construction area, simultaneous detection of dust concentration and light intensity, and frame-by-frame correction using edge-preserving joint correction factors to eliminate interference and preserve the edge details of the boulders; Extract the edge features of the boulders in the corrected image, locate the feature inflection points by curvature and gradient direction angle, and fill in the missing edges with the optimal cubic Bézier curve based on the convex polygon shape of the boulders to form a closed contour. Extract the curvature distribution of the stone outline to locate abrupt change points, match the corresponding abrupt change points for overlapping stones and generate three types of candidate segmentation boundaries, select the boundary with the smallest comprehensive energy value to complete the segmentation, and assign a unique region label to each stone. The image is converted to the CIELAB color space to generate superpixels. The pouring area is initially determined by feature similarity matching. Isolated small areas and dynamic interference areas are screened out by connected component analysis and inter-frame difference method to obtain the accurate pouring boundary. Based on the calibration relationship between pixels and physical size, the complete three-dimensional surface area of ​​the stone block is estimated, the exposed stone rate is calculated by combining the concrete pouring area, and the results are verified by inter-frame difference and change trend. In case of anomalies, re-acquisition and recalculation are triggered. The real-time exposure rate is compared with the preset construction standards to generate quality evaluation results. If the results are unqualified, deviation information is marked. All monitoring data, processed images and evaluation results are output to the construction management platform in a standardized format.

Images