Stone surface flaw automatic identification method and system based on intelligent algorithm

Abstract

The invention discloses a stone surface flaw automatic identification method and system based on an intelligent algorithm, and belongs to the technical field of machine vision and stone processing. The invention provides an automatic detection scheme for solving the problems that in the prior art, manual stone slab defect detection is low in efficiency and high in subjectivity, and a traditional machine vision method is insufficient in detection precision under the conditions of complex stone slab textures and mixed noise. The method comprises the following steps: constructing an image acquisition system and carrying out camera calibration and image correction; the method comprises the following steps: preprocessing a stone plate image by adopting a denoising method combining median filtering and non-local mean (NLM) filtering, and extracting a stone plate contour by combining an improved Canny algorithm; and then, solving the maximum inscribed rectangle in the contour through a histogram area method, performing histogram equalization enhancement on the rectangular region, and finally, segmenting and identifying defects such as color spots and color lines by adopting a region splitting and merging algorithm combined with morphology.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and materials processing, and in particular to an automatic identification method and system for surface defects of natural stone based on intelligent algorithms. It is applied to the stone processing industry for intelligent and high-precision identification of surface defects (such as color difference, abnormal texture, micro-cracks, etc.) of various natural stones (such as marble, granite, sandstone, etc.). Background Technology

[0002] Stone, with its unique beauty and excellent durability, is widely used in building decoration. However, due to differences in geological conditions and the mechanical stress during mining and processing, stone surfaces often have various defects. These defects not only damage the aesthetics and reduce value, but some structural problems even threaten safety and service life. Currently, defect detection mainly relies on manual visual inspection, which is inefficient, has inconsistent standards, and is easily affected by lighting and the subjectivity of the inspector. Given the increasing automation of stone processing, developing accurate and efficient intelligent defect identification technology has become an urgent need.

[0003] While current computer vision-based stone defect detection technology has made some progress, it still faces the following technical challenges: First, the surface texture of natural stone is complex and varied, with significant differences in color and texture characteristics between different varieties, making it difficult to construct a universal detection model. Second, there are numerous types of defects on the stone surface (such as color differences, micro-cracks, local unevenness, etc.), and some defects are similar to natural texture features, making it difficult for traditional image processing algorithms to effectively distinguish them. Third, the lighting conditions in actual production environments are unstable, and the reflective properties of stone surfaces also vary, resulting in large fluctuations in image acquisition quality. Fourth, the irregular shape of stone edges increases the difficulty of accurately extracting effective detection areas. Common traditional methods such as threshold segmentation and edge detection are often limited by these factors when dealing with stone defect identification problems, making it difficult to achieve ideal results.

[0004] Developing an intelligent method and system that can adapt to the surface characteristics of various types of stone, has strong environmental adaptability, and can accurately identify multiple types of defects is of great significance for improving the quality control level of stone products and the degree of automation in the industry. Summary of the Invention

[0005] The purpose of this invention is to overcome the aforementioned shortcomings of existing stone slab defect detection technologies and to provide an automatic identification method and system for stone surface defects based on intelligent algorithms, thereby improving the accuracy, robustness, and automation of stone slab defect detection. To achieve the above objective, the technical solution adopted by this invention is as follows: An automatic identification method for stone surface defects based on intelligent algorithms, comprising the following steps: S1: Image acquisition and illumination compensation: Construct a multi-region adjustable light source system and design an automatic exposure compensation algorithm to ensure that high-resolution images of stone surfaces with uniform illumination are acquired. S2: Image Enhancement and Preprocessing: S2.1: Dual Gradient Adaptive Filtering for Noise Reduction: Gradient-guided bidirectional filtering and edge-preserving filtering are applied to reduce image noise while preserving texture edges.

[0006] S2.2: Multi-scale boundary extraction: Construct a multi-scale pyramid, fuse edge features, and extract complete stone boundaries through adaptive threshold segmentation.

[0007] S3: Defect Identification and Location: S3.1: Determination of the optimal analysis area: The maximum rectangle covering the effective area of ​​the stone is determined by applying the maximum rectangle covering algorithm.

[0008] S3.2: Feature Enhancement: Apply local contrast adaptive adjustment and texture enhancement technology to enhance the differences in defect areas.

[0009] S3.3: Defect Identification and Localization: A hybrid segmentation strategy combining deep learning and region clustering is used to identify and locate defects, and generate a quantitative assessment report.

[0010] Preferably, the camera calibration in S1 adopts the Zhang Zhengyou calibration method.

[0011] Preferably, the image denoising process in S2.1 employs a combination of median filtering and nonlocal mean filtering.

[0012] Preferably, the stone slab contour extraction in S2.2 employs an improved Canny edge detection algorithm. This improved Canny edge detection algorithm includes: using an image processed by median filtering and non-local mean filtering as input; automatically determining high and low thresholds using the maximum inter-class variance method; and optimizing edges by combining morphological processing. Preferably, the maximum inscribed rectangle in S3.1 is solved using the histogram area method.

[0013] Preferably, the image enhancement process in S3.2 employs a histogram equalization method.

[0014] Preferably, the defect segmentation and recognition in S3.3 adopts a region splitting and merging algorithm that integrates morphology, and the splitting and merging criteria are the mean and standard deviation of pixels within the region.

[0015] An automatic stone surface defect identification system employing the aforementioned intelligent algorithm-based automatic stone surface defect identification method includes an image acquisition and calibration module for acquiring stone slab images, executing a camera calibration program, and correcting the acquired stone slab images; an image preprocessing module connected to the image acquisition and calibration module for performing image denoising and stone slab contour extraction operations on the corrected stone slab images to obtain the stone slab contour; and a stone slab defect detection module connected to the image preprocessing module for solving the maximum inscribed rectangle based on the stone slab contour, and performing image enhancement processing and defect segmentation and identification operations on the image within the maximum inscribed rectangle area.

[0016] Preferably, the image denoising process in the image preprocessing module employs a combination of median filtering and nonlocal mean filtering.

[0017] Preferably, the slab contour extraction operation in the image preprocessing module adopts an improved Canny edge detection algorithm, which integrates the maximum inter-class variance method to automatically determine the threshold and morphological edge optimization processing.

[0018] This invention offers the following advantages: High-precision defect recognition capability: By combining dual-gradient adaptive filtering with deep learning feature extraction, accurate recognition of minute defects is achieved even against complex texture backgrounds, improving detection accuracy by 27.8% compared to traditional methods. Strong environmental adaptability: The combination of a multi-region adjustable light source system and automatic exposure compensation technology adapts to different stone surface reflectivity and ambient light interference, ensuring stable image acquisition quality; the improved maximum rectangle coverage algorithm automatically adapts to irregular stone boundaries to determine the optimal analysis area. High-efficiency processing speed: Through GPU acceleration and algorithm optimization, the system's processing speed is 6 times faster than traditional methods, and the complete detection process for a single stone (from image acquisition to defect recognition) takes only 3-5 seconds, meeting the needs of industrial production. Good scalability: The hybrid segmentation strategy can continuously improve the recognition model through incremental sample learning, supporting rapid adaptation to new defect types. Simultaneously, the system architecture supports independent configuration of characteristic parameters for various stone varieties, exhibiting good versatility and scalability. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0020] Figure 1This is a flowchart illustrating the automatic identification method for surface defects of natural stone in an embodiment of the present invention. Figure 2 This is a structural block diagram of the automatic identification system for surface defects of natural stone in an embodiment of the present invention. Detailed Implementation

[0021] 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 a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present invention.

[0022] Example The following are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the following embodiments. All technical solutions that fall within the scope of the present invention are within the scope of protection of the present invention.

[0023] In this embodiment, the present invention constructs a complete automatic identification system for surface defects in natural stone. In terms of hardware configuration, a high-resolution industrial camera (2448×2048 pixels) is used, coupled with a 50mm macro lens, to ensure clear imaging of minute defects on the stone surface. The light source system consists of a central main light source and four independently controllable ring-shaped auxiliary light sources. The angle between the light source and the stone surface can be adjusted within the range of 30°-60° to adapt to the reflective characteristics of different stone surfaces. The system control section uses an industrial control computer equipped with an NVIDIA RTX 3060 GPU accelerator card for accelerating deep learning model inference and image processing.

[0024] Image Acquisition and Illumination Compensation. After system startup, the system first initializes the light sources, setting each light source area to default brightness (70% for the main light source and 50% for the auxiliary light source). When the stone is placed in the detection area, the system acquires a preview image and divides it into a 5×5 grid area, calculating the average brightness and standard deviation of each grid. If the brightness difference between grids exceeds 15% or the standard deviation within any grid exceeds 25, an automatic illumination compensation mechanism is triggered: the intensity of the light source corresponding to areas with excessive brightness is reduced, and the intensity of the light source corresponding to areas with insufficient brightness is increased, thus optimizing illumination uniformity. After 2-3 iterations of adjustment, when the brightness difference of all grids drops below the threshold, formal image acquisition is performed, obtaining a high-quality image of the stone surface for subsequent processing.

[0025] Dual-gradient adaptive filtering for noise reduction. The acquired image is first converted to grayscale, and then the local gradient distribution characteristics are calculated. Specifically, for each pixel (x, y), its gradient values ​​in eight directions are calculated, resulting in a gradient matrix G(x, y). A bidirectional filter is constructed based on the gradient matrix. The spatial kernel and range kernel parameters of the filter are dynamically adjusted according to the local gradient characteristics: in regions with large gradients (possibly stone texture edges), the range kernel parameter is decreased to preserve edge details; in regions with small gradients (possibly smooth stone areas), the spatial kernel parameter is increased to enhance the smoothing effect. After the first stage of filtering, an edge-preserving filter is applied for secondary processing. This filter uses a nonlinear diffusion equation, adjusting the diffusion coefficient to adopt different processing strategies for regions with different gradients, further removing potential artifacts while enhancing the continuity of the original texture features.

[0026] Multi-scale boundary extraction. A four-level pyramid structure (original resolution and three downsampling levels) is constructed for the filtered image. At each scale, a feature-enhanced edge detection operator is used to extract edge features. This operator combines the gradient calculation capability of the traditional Sobel operator with the directional selectivity of the Log-Gabor filter, enabling it to better capture the directional features of the stone boundary. For the edge features extracted at each scale, a mechanism is introduced to ensure consistency in direction: if the edge features at the same location at different scales have high directional consistency (directional difference less than π / 6), the weight of that edge feature is increased; otherwise, the weight is decreased. The weighted multi-scale edge features are fused to obtain a preliminary edge image. Finally, adaptive thresholding (using the Otsu method to automatically calculate the threshold) and connectivity analysis (removing regions with an area less than 1% of the total image area) are applied to extract the complete stone boundary and eliminate background interference.

[0027] The optimal analysis region is determined. A maximum rectangle coverage algorithm is applied to the extracted stone boundaries. This algorithm first converts the boundary image into a binary image, where the stone area is represented by 1 and the background by 0. Then, the image is divided into multiple sub-regions (typically a 4×4 grid), and the maximum rectangle within each sub-region is calculated in parallel. The calculation process employs a dynamic programming strategy: each row is scanned from left to right, maintaining a height array and a left boundary array. When a 1 is encountered, these arrays are updated, and the area of ​​the maximum rectangle at the current position is calculated. After calculating all sub-regions, local results are merged while boundary conflicts are cleared, ultimately determining the global maximum rectangle covering the effective area of ​​the stone. To handle potential irregularities at the stone edges, the algorithm allows a small number of background pixels (not exceeding 5% of the rectangle's perimeter) at the edge of the maximum rectangle. This tolerance mechanism significantly increases the effective analysis area. Details of the maximum rectangle coverage algorithm are as follows: 1. Input preprocessing: The binary image matrix obtained by boundary extraction is denoted as Mm×n Mm × n ,in: M(i,j)={1 (effective pixels of stone) 0 (background pixels)}; M ( i , j = {10 (effective pixels for stone) (background pixels)}} 2. Dynamic programming matrix construction, defining three auxiliary matrices: Height matrix H(i,j), H ( i , j ): Records the maximum height of consecutive 1s above the current position; Left boundary matrix L(i,j), L ( i , j ): Records the leftmost extendable boundary at the current height; Right boundary matrix R(i,j), R ( i , j ): Records the rightmost extendable boundary at the current height; 3. Recursive formula: H(i,j)={H(i-1,j)+1ifM(i,j)}=10otherwise H ( i , j )={ H ( i -1, j +10if M ( i , j )}=1otherwise L(i,j) = max{L(i-1,j)}, the starting position of consecutive 1s to the left of the current row. L ( i , j )=max{ L ( i -1, j The starting position of consecutive 1s to the left of the current line. R(i,j)=min{R(i-1,j)} is the position where the consecutive 1s to the right of the current row terminate. R ( i , j )=min{ R ( i -1, j The position where consecutive 1s to the right of the current line terminate. 4. Maximum area calculation: Calculate the area of ​​the current rectangle while iterating through each pixel. Area(i,j)=H(i,j)×[R(i,j)-L(i,j)+1]Area( i , j )= H ( i , j )×[ R ( i , j )- L ( i , j )+1] 5. Global maximum rectangle area: MaxArea=max1≤i≤m,1≤j≤n(Area(i,j))MaxArea=1≤ i ≤ m ,1≤ j ≤ n max(Area( i , j )) 6. Boundary conflict resolution: When overlapping boundaries occur during sub-region merging, a weight adjustment strategy is employed. w = Area of ​​overlapping region (area of ​​sub-rectangle) × confidence level (0.9noverlap) w =Area of ​​overlapping region of sub-rectangles × Confidence level (0.9) n overlap) 7. The final rectangle coordinates are determined by a weighted average: xfinal=∑wkxk∑wk x final=∑ week ∑ wkxk Feature Enhancement. Feature enhancement processing is performed on the determined optimal analysis region. First, local statistical characteristics within the region are calculated, including mean, standard deviation, and skewness, to construct a local feature map. Based on the local feature map, the contrast parameter is adaptively adjusted: a higher contrast enhancement coefficient is used for regions with drastic feature changes (potentially defective areas); a lower enhancement coefficient is used for regions with gradual feature changes (potentially normal texture areas). This adaptive strategy effectively enhances the difference between defective and normal areas while suppressing interference from natural textures. The enhancement process employs a non-linear mapping function, which dynamically adjusts parameters according to the grayscale distribution characteristics of the local region to ensure optimal enhancement results.

[0028] Mathematical description of the feature enhancement steps: 1. For the optimal analysis region IROI IROI Calculate the statistics for the local window (5×5 pixels): μ(x,y)=125∑i=-22∑j=-22I(x+i,y+j) m ( x , y )=251 i =-2∑2 j =-2∑2 I ( x + i , y + j ) σ(x,y)=125∑i=-22∑j=-22(I(x+i,y+j)-μ(x,y))2 s ( x , y )=251 i =-2∑2 j =-2∑2( I ( x + i , y + j )- m ( x , y ))2 2. Nonlinear contrast enhancement function: Ienhanced(x,y)={μ(x,y)+k1⋅σ(x,y)ifI(x,y)>μ(x,y)μ(x,y)-k2⋅σ(x,y)otherwiseIenhanced( x , y ) ={ m ( x , y )+k 1⋅ s ( x , y ) m ( x , y )- k 2⋅ s ( x , y if I ( x , y )> m ( x , y otherwise Where k1 = 2.5 k 1=2.5, k2=1.8 k 2=1.8 is an empirical coefficient, obtained by looking up the table based on the type of stone.

[0029] Defect Identification and Localization. A hybrid segmentation strategy combining deep learning and region clustering is employed. A pre-trained convolutional neural network (using an improved ResNet-18 architecture) is used to extract high-dimensional feature representations of the stone surface. The network, pre-trained on over 1000 stone samples, learns the texture features and common defect features of different stone types, enabling it to distinguish between normal and abnormal regions in the feature space. Dimensionality reduction is performed on the extracted feature maps (using the t-SNE algorithm), and density-based spatial clustering analysis (DBSCAN algorithm) is applied to classify regions with similar features. The clustering parameters ε (neighborhood radius) and MinPts (minimum number of points) are adaptively set according to the stone type. Post-processing of the clustering results includes small region filtering (regions smaller than 50 pixels are considered noise) and region morphological optimization (using opening and closing operations to smooth region boundaries). Through feature fusion (combining original image features, texture features, and depth features) and a multi-threshold decision mechanism, regions are classified into normal regions and different types of defect regions (such as color difference, texture anomalies, microcracks, etc.). The recognition results are overlaid on the original image as colored outlines, with different colors representing different types of defects. A quantitative assessment report containing information such as defect type, location, and area is generated. The specific mathematical supplement for defect recognition is as follows: 1. DBSCAN clustering parameters are adaptive, including the neighborhood radius ϵ. ϵ Automatically determined based on feature space density: ϵ=1n∑i=1ndist(xi,xNNk) ϵ = n 1 i =1∑ n dist( xi , x NN k ) Where NNkNN k For the k-th nearest neighbor, n, n This represents the number of samples.

[0030] 2. Multi-threshold decision-making mechanism, defining defect confidence score: Score=0.4⋅fcolor+0.3⋅ftexture+0.3⋅fdepthScore=0.4⋅ f color + 0.3 ⋅ f texture +0.3⋅ f depth When Score>τScore> t When it is judged as a defect, the threshold τ is... t Adjustments will be made dynamically according to the type of stone.

[0031] System Validation and Performance Evaluation. The system performance was evaluated using a test set containing 500 stones of different types (marble, granite, sandstone, etc.). Each stone included manually labeled defect location and type information as a ground truth reference. Experimental results show that the system achieves a defect detection accuracy of 92.7%, a 27.8% improvement over the traditional threshold-based segmentation method (64.9% accuracy). In terms of processing speed, the complete detection process for a single stone (from image acquisition to defect identification) takes an average of 4.2 seconds, six times faster than traditional manual detection (average 25 seconds per stone). Adaptability tests to different lighting conditions show that the detection accuracy fluctuates by no more than 3% within a range of ±30% variation in ambient light intensity, demonstrating good environmental adaptability. Furthermore, the system exhibits high consistency in test results on 10 types of stones with different texture complexities, validating its versatility across various stone varieties.

[0032] The method and system proposed in this invention are not only applicable to the stone industry, but can also be extended to other material fields requiring surface quality inspection, such as wood, ceramics, and metal sheets. By adjusting relevant parameters and feature extraction models, the system can adapt to the surface characteristics and defect types of different materials, and has broad application prospects.

[0033] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for automatic identification of surface defects in stone based on intelligent algorithms, characterized in that, Includes the following steps: S1: Image acquisition and illumination compensation: Construct a multi-region adjustable light source system and design an automatic exposure compensation algorithm to ensure that high-resolution images of stone surfaces with uniform illumination are acquired. S2: Image Enhancement and Preprocessing: S2.1: Dual Gradient Adaptive Filtering Denoising: Apply gradient-guided bidirectional filtering and edge-preserving filtering to denoise the image while preserving texture edges; S2.2: Multi-scale boundary extraction: Construct a multi-scale pyramid, fuse edge features, and extract complete stone boundaries through adaptive threshold segmentation; S3: Defect Identification and Location: S3.1: Determination of the optimal analysis area: The maximum rectangle covering the effective area of ​​the stone is determined by applying the maximum rectangle covering algorithm; S3.2: Feature Enhancement: Apply local contrast adaptive adjustment and texture enhancement technology to enhance the differences in defect areas; S3.3: Defect Identification and Localization: A hybrid segmentation strategy combining deep learning and region clustering is used to identify and locate defects, and generate a quantitative assessment report.

2. The method for automatic identification of stone surface defects based on intelligent algorithms according to claim 1, characterized in that, The camera calibration in S1 adopts the Zhang Zhengyou calibration method.

3. The method for automatic identification of stone surface defects based on intelligent algorithms according to claim 1, characterized in that, The image denoising process in S2.1 employs a combination of median filtering and nonlocal mean filtering.

4. The method for automatic identification of stone surface defects based on intelligent algorithms according to claim 1 or 3, characterized in that, The stone slab contour extraction in S2.2 uses an improved Canny edge detection algorithm, which includes: using the image after median filtering and nonlocal mean filtering as input, automatically determining the high and low thresholds using the maximum inter-class variance method, and optimizing the edges by combining morphological processing.

5. The method for automatic identification of stone surface defects based on intelligent algorithms according to claim 1, characterized in that, The maximum inscribed rectangle in S3.1 is solved using the histogram area method.

6. The method for automatic identification of stone surface defects based on intelligent algorithms according to claim 1, characterized in that, The image enhancement process in S3.2 employs a histogram equalization method.

7. The method for automatic identification of stone surface defects based on intelligent algorithms according to claim 1, characterized in that, The defect segmentation and recognition in S3.3 adopts a region splitting and merging algorithm that integrates morphology, and the splitting and merging criteria are the mean and standard deviation of pixels within the region.

8. An automatic stone surface defect identification system employing the automatic stone surface defect identification method based on intelligent algorithms as described in any one of claims 1-7, characterized in that, It includes an image acquisition and calibration module for acquiring stone slab images, executing camera calibration procedures, and correcting the acquired stone slab images; and an image preprocessing module connected to the image acquisition and calibration module for performing image denoising and stone slab contour extraction operations on the corrected stone slab images to obtain the stone slab contours. The stone slab defect detection module is connected to the image preprocessing module. It is used to solve the maximum inscribed rectangle of the stone slab based on its outline, and to perform image enhancement processing and defect segmentation and recognition operations on the image within the maximum inscribed rectangle area.

9. The automatic stone surface defect identification system according to claim 8, characterized in that, The image denoising process in the image preprocessing module employs a combination of median filtering and nonlocal mean filtering.

10. The automatic stone surface defect identification system according to claim 8 or 9, characterized in that, The slab contour extraction operation in the image preprocessing module adopts an improved Canny edge detection algorithm, which integrates the maximum inter-class variance method to automatically determine the threshold and morphological edge optimization processing.

Images