Cutting workpiece defect detection method and system based on image feature feedback

Abstract

The invention discloses a cut workpiece defect detection method and system based on image feature feedback, and relates to the technical field of image processing.The method comprises the steps that cut workpiece technological characteristics are obtained, a preset defect type library is constructed, a hardware system is built, and parameters are initialized; synchronously acquiring a multi-view original image, and storing and associating annotation information; de-noising the original image, enhancing the contrast, and extracting a region of interest ROI; extracting texture, shape, edge and gray features from the ROI, and screening through a Relief-F algorithm to obtain an optimal feature subset; inputting into an SVM (Support Vector Machine) model for reasoning, and screening to obtain an effective defect detection result; and calculating an evaluation index and generating a feedback signal, and performing iterative optimization after adjusting parameters. The system comprises an acquisition module, a master control module, a data processing module and a display module. Through the precise design and closed-loop feedback of the whole process, the precision, efficiency and long-term adaptability of defect detection of the complex cutting workpiece are improved, and the industrial quality management and control requirements are met.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and system for detecting defects in a cutting workpiece based on image feature feedback. Background Art

[0002] In modern manufacturing, cutting is one of the core processes for workpiece forming, and its processing quality directly determines the performance, reliability and service life of the product. Especially in high-end fields such as aerospace, automobile manufacturing, and precision instruments, defects on the surface and inside of cut workpieces (such as cracks, burrs, dimensional deviations, etc.) may cause serious safety hazards and economic losses. With the development of intelligent manufacturing, traditional defect detection methods that rely on manual vision can no longer meet the production needs of high efficiency, high precision, and automation. Machine vision-based detection technology has become the mainstream trend due to its non-contact, fast speed, and quantifiable advantages. Among them, the detection method based on image feature feedback can dynamically adapt to complex working conditions by extracting the key features of the workpiece image and forming a closed-loop optimization mechanism, further improving detection stability. Therefore, it has important application value in the quality control of cut workpieces.

[0003] Existing methods and systems for detecting defects in cutting workpieces are not adaptable enough to workpieces with complex shapes (such as deep cavities, multiple curved surfaces, and porous structures). Visual blind spots are prone to occur in image acquisition, and uneven lighting makes it difficult to extract defect features, making it difficult to cover the entire area to be inspected. At the same time, their feature extraction is not very targeted and mostly relies on a single or limited feature dimension (such as only focusing on texture or shape features). The distinction between different types of defects (such as micro cracks and macro burrs) is insufficient, and false detection or missed detection is prone to occur. Therefore, it is necessary to provide a method and system for detecting defects in cutting workpieces based on image feature feedback to solve the above-mentioned problems. Summary of the Invention

[0004] In order to solve the above technical problems, a method and system for detecting defects in cutting workpieces based on image feature feedback are provided. This technical solution solves the problems that the existing method and system for detecting defects in cutting workpieces proposed in the above background technology are not sufficiently adaptable to workpieces with complex shapes (such as deep cavities, multiple curved surfaces, and porous structures). Image acquisition is prone to visual blind spots, and uneven lighting makes it difficult to extract defect features, making it difficult to cover all areas to be inspected. At the same time, its feature extraction is not very targeted, and it mostly relies on a single or limited feature dimension (such as only focusing on texture or shape features). The distinction between different types of defects (such as micro cracks and macro burrs) is insufficient, and it is prone to false detection or missed detection.

[0005] In order to achieve the above objects, the technical solution adopted by the present invention is: A method for detecting defects in a cutting workpiece based on image feature feedback, comprising: Acquire the process characteristics of the cutting workpiece, simultaneously build a preset defect type library, and build a hardware system including an acquisition module, mechanical structure, and computing unit. Based on the performance parameters of the hardware system and the characteristics of the defect type, initialize the image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model thresholds; Based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system, the original images of the cut workpiece at multiple viewing angles are synchronously acquired. The original images are stored in the database and associated with the annotation information according to the acquisition time and workpiece ID of the original images. Based on the original image stored in the database, the adaptive filtering algorithm is used to denoise the original image to obtain a denoised image. According to the grayscale distribution characteristics of the denoised image, the image contrast is enhanced by the CLAHE algorithm to obtain an enhanced image. Then, the edge segmentation algorithm is used to extract the region of interest (ROI) of the workpiece. Based on the extracted region of interest (ROI), texture features, shape features, edge features, and grayscale features are extracted respectively. Based on the extracted multidimensional features, feature screening is performed using the Relief-F algorithm to obtain the optimal feature subset. The optimal feature subset is input into the detection model for inference, and the preliminary detection results of the defects are output, including the location coordinates, type and confidence level of the defects. The defect confidence threshold is preset, and the preliminary detection results are screened to obtain effective defect detection results. Determine the evaluation indicators, compare the evaluation indicators with the defect confidence threshold, locate the problem links in the detection process, and generate corresponding feedback signals; Based on the generated feedback signal, the image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model parameters are adjusted respectively. According to the adjusted parameters, the process from image acquisition to defect detection and classification is repeated, the final detection results are output, and iterative optimization of the detection model and feature extraction method is achieved.

[0006] In an optional embodiment, the process of obtaining the process characteristics of the cutting workpiece, simultaneously constructing a preset defect type library, and building a hardware system including an acquisition module, a mechanical structure, and a computing unit, and initializing image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model thresholds based on the performance parameters of the hardware system and the defect type characteristics, specifically includes: Obtain the process characteristics of the cutting workpiece and call the specific definitions and characteristic parameters of surface defects, size defects and morphological defects from the preset defect type library; Determine the performance requirements of the acquisition module and build an image acquisition subsystem using an industrial camera, a lens adapted to the workpiece size, and a coaxial light source; Based on the size and shape complexity of the workpiece, a workpiece positioning table with rotation / translation axes and a trigger device linked to the production line were designed to construct a mechanical structure subsystem. The angular accuracy of the rotation axis met α≤0.1°, and the displacement accuracy of the translation axis met β≤0.01mm. Based on the real-time requirements of image processing and model inference, GPU servers are configured as computing units to ensure that their floating-point computing capability is ≥10 TFLOPS. Based on the performance parameters of the hardware system, the image acquisition parameters are initialized, including: exposure time t0, which is calculated based on the workpiece surface reflectivity as t0 = k × (1 / ρ), where k is the illumination coefficient and ρ is the surface reflectivity, with a value range of 0.2 ≤ ρ ≤ 0.8. The gain g0 is set to an initial value of 1.0; Initialize the filter window size s0 in the preprocessing parameters according to the scale range of the defect characteristics, satisfying s0=2×floor (d / 2)+1, where d is the estimated diameter of the minimum defect and floor (・) is the floor rounding function; Based on the feature dimension of the defect, initialize the feature extraction dimension m0, which includes the number of texture features m1, the number of shape features m2, the number of edge features m3, and the number of grayscale features m4, satisfying m0=m1+m2+m3+m4, and m1≥5, m2≥3, m3≥2, and m4≥2; The initial confidence threshold of the detection model is set to θ0 = 0.8 for preliminary screening of defect detection results.

[0007] In an optional embodiment, the method of synchronously capturing original images of the cut workpiece from multiple perspectives based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system, storing the original images in a database and associating them with annotation information based on the acquisition time and workpiece ID of the original images, specifically includes: Based on the production line's takt time T, the image acquisition interval Δt=T is set by the trigger device to ensure synchronization with the workpiece transfer rhythm; When the workpiece arrives at the inspection station, the trigger device sends a trigger signal to the acquisition module to start the multi-view image acquisition process; Control the rotating shaft in the mechanical structure subsystem to drive the workpiece to rotate at an angle θ i =i×(360° / n), where i=1, 2, ..., n, and n is the number of viewing angles, satisfying 3≤n≤6; At each viewing angle, the industrial camera collects raw images according to the initialized exposure time t0 and gain g0. The image format is set to BMP or TIFF to preserve the original pixel information. During the collection process, , real-time monitoring of image clarity evaluation index Q, where For the image The gray value at Q<Q min When Q is ≥ Q, the exposure time t=t0+Δt is automatically adjusted until Q≥Q min , Q min is the preset clarity threshold; The collected multi-view original images are named according to the naming rule of workpiece ID - view serial number - acquisition time; Based on the database storage specifications, the original image is stored in the distributed database, and an associated index between the workpiece ID and the image file path is established; Annotation information is added to each original image, including the material of the workpiece, processing technology type, tool model, cutting parameters, and the ambient temperature and humidity during acquisition, to form complete image metadata.

[0008] In an optional embodiment, the method of denoising the original image stored in the database using an adaptive filtering algorithm to obtain a denoised image, enhancing the image contrast using a CLAHE algorithm based on the grayscale distribution characteristics of the denoised image to obtain an enhanced image, and then extracting the region of interest (ROI) of the workpiece using an edge segmentation algorithm specifically includes: Read the original image I from the database and calculate the noise type judgment index N of the image; Among them, the calculation formula of noise type judgment index N is: , median (・) is the median calculation function. When N<N1, it is judged as slight noise, and when N≥N1, it is judged as impulse noise. N1 is the preset noise threshold; If the noise is slight, use Gaussian filtering algorithm to remove the noise: the filter kernel function is , where σ is the Gaussian standard deviation, the value is σ=0.5×s0, s0 is the size of the initialized filter window, and the denoised image I1=I*G is obtained, which represents the convolution operation; If it is impulse noise, use the median filter algorithm to denoise: Denoised image I1(x,y)=median(I(xs:x+s,ys:y+s)), where s=floor(s0 / 2), is obtained. Calculate the grayscale histogram H (k) of the denoised image I1, k = 0, 1, ..., 255, where H (k) is the number of pixels with grayscale value k; If the image grayscale distribution is concentrated, the CLAHE algorithm is used to enhance the contrast and calculate the cumulative distribution function. ,in is the total number of pixels in the image; The enhanced gray value k'=floor (255×CDF (k)) is obtained to obtain the enhanced image I2; If the image has insufficient local contrast, the CLAHE algorithm is used to enhance the contrast. The image is divided into 8×8 sub-blocks. The histogram of each sub-block is calculated and truncation is performed. The truncation threshold is T=α×mean(H(k)), where α is the adjustment coefficient and ranges from 1.0≤α≤3.0, and mean(·) is the mean function. The sub-blocks are spliced ​​together by bilinear interpolation to obtain the enhanced image I2. The OTSU threshold segmentation algorithm is used to separate the background of the enhanced image I2 and calculate the inter-class variance: ,in 、 are the ratio of foreground and background pixels, 、 The grayscale mean of the foreground and background are respectively, and the threshold t corresponding to the maximum inter-class variance is found to segment the image into foreground and background; Combined with the edge segmentation algorithm, the Canny operator is used to extract the edge contour of the foreground area; Among them, the high threshold T of edge detection H =0.3×max (G), low threshold T L =0.15×max (G), G is the image gradient amplitude, and the region of interest (ROI) of the workpiece is determined by the edge contour. The coordinate range of the ROI is (xmin, xmax, ymin, ymax).

[0009] In an optional embodiment, the texture features, shape features, edge features, and grayscale features are extracted based on the extracted region of interest (ROI). Feature screening is performed using the Relief-F algorithm based on the extracted multidimensional features to obtain an optimal feature subset, specifically including: In the region of interest (ROI), texture features are extracted and texture parameters are calculated using the gray-level co-occurrence matrix (GLCM). For distance d = 1, angle The GLCM matrix , calculate the energy characteristics , entropy characteristics , contrast feature , correlation characteristics ,in 、 is the row and column mean, 、 is the row and column standard deviation; The LBP algorithm is used to extract local binary pattern features, and the LBP histogram is statistically analyzed to obtain the texture feature vector F1=[E,Ent,Con,Cor,...]; Extract shape features, extract contours of defect candidate areas within ROI, and , calculate the area of ​​the contour, where R is the defect area; pass , determine the contour perimeter, where x(i) and y(i) are the coordinates of the contour points and m is the number of contour points; The circularity of the contour is determined by C=4πA / L², and the aspect ratio of the contour is determined by R=W / H, where W and H are the width and height of the defect's circumscribed rectangle, respectively. The shape feature vector F2=[A,L,C,R,...] is obtained; Extract edge features and apply Sobel operator to ROI to calculate horizontal and vertical gradients 、 , gradient amplitude , gradient direction , obtain edge direction features by statistical gradient direction histogram, and calculate edge continuity features ,in is the length of the continuous edge segment, is the number of edge segments, and the edge feature vector is obtained , is the mean gradient amplitude, is the gradient direction mode; Extract grayscale features by , calculate the grayscale mean within the ROI, is the number of pixels within the ROI; pass , determine the grayscale variance; by , determine the grayscale entropy, where p (k) is the probability that the grayscale value in the ROI is k, and obtain the grayscale feature vector ; Splice the texture feature F1, shape feature F2, edge feature F3 and grayscale feature F4 into a multidimensional feature matrix , the dimension is m0×1, m0 is the initial feature extraction dimension; The Relief-F algorithm is used to calculate the weight w(j) of each feature and update it iteratively: ;in, , is the nearest neighbor sample of the same type, is the nearest neighbor sample of different classes, n is the number of samples, p (c) is the prior probability of class c, and the weight w (j) ≥ w min The features of min is the weight threshold.

[0010] In an optional embodiment, the optimal feature subset is input into the detection model for inference, and a preliminary defect detection result is output, including the location coordinates, type, and confidence level of the defect. A defect confidence threshold is preset, and the preliminary detection results are screened to obtain a valid defect detection result, specifically including: Based on the optimal feature subset X' obtained by screening, a detection sample set is constructed, where the positive sample is a feature vector containing defects, and the negative sample is a feature vector without defects. The sample number ratio satisfies 1:1≤positive sample:negative sample≤3:1; Select the SVM model for training and construct the kernel function: K (x, y) = exp (-γ||xy||²); where γ is the kernel parameter. The optimal parameters C and γ are determined through cross-validation, where C is the penalty coefficient, with a value of 1≤C≤100, and γ is 0.01≤γ≤10. Train the SVM model to obtain the classification hyperplane: ;in, is the Lagrange multiplier, is the sample label, is the bias term, is the number of support vectors; Input the optimal feature subset X' into the trained SVM model and output the predicted category of the defect And the decision value d, confidence c=1 / (1+exp (-d)), get the preliminary detection result; Extract the confidence level c from the preliminary inspection results and compare it with the preset defect confidence level threshold θ0; When c≥θ0, it is determined to be a valid defect, and the location coordinates, type and confidence information of the defect are retained; When c<θ0, it is determined to be an invalid defect and is eliminated; The information of all valid defects is summarized to form a list of valid defect detection results, where each defect entry contains a unique identifier ID, location coordinates (x, y), type label t, size parameters (s1, s2) and confidence c.

[0011] Furthermore, a cutting workpiece defect detection system based on image feature feedback is proposed, which is used to implement any of the above detection methods, including: An acquisition module, which is used to obtain the process characteristics of the cut workpiece and simultaneously build a preset defect type library, and is used to synchronously acquire original images of the cut workpiece from multiple perspectives based on initialized image acquisition parameters and the acquisition module and trigger device in the hardware system; A main control module, which is used to initialize image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model thresholds based on the performance parameters of the hardware system and defect type characteristics, input the optimal feature subset into the detection model for inference, output preliminary defect detection results, including the location coordinates, type and confidence of the defects, preset defect confidence thresholds, screen the preliminary detection results, obtain effective defect detection results, determine evaluation indicators, compare the evaluation indicators with the defect confidence thresholds, locate problem links in the detection process, and generate corresponding feedback signals. It is also used to adjust the image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model parameters based on the generated feedback signals, and repeatedly execute the process from image acquisition to defect detection and classification according to the adjusted parameters, output the final detection results and realize iterative optimization of the detection model and feature extraction method; a data processing module, wherein the data processing module is configured to store the original image in a database and associate the annotation information with the original image based on the acquisition time of the original image and the workpiece ID; to perform denoising on the original image based on the original image stored in the database using an adaptive filtering algorithm to obtain a denoised image; and to enhance the image contrast using a CLAHE algorithm based on the grayscale distribution characteristics of the denoised image to obtain an enhanced image; and then to extract the region of interest (ROI) of the workpiece using an edge segmentation algorithm; and based on the extracted region of interest (ROI), respectively extract texture features, shape features, edge features, and grayscale features; and to perform feature screening using a Relief-F algorithm based on the extracted multidimensional features to obtain an optimal feature subset; A display module is used to present the process and results of the cutting workpiece defect detection to the user.

[0012] In an optional embodiment, the acquisition module includes: A first acquisition unit, the first acquisition unit is used to obtain the process characteristics of the cutting workpiece and simultaneously build a preset defect type library; The second acquisition unit is used to synchronously acquire original images of the cutting workpiece under multiple viewing angles based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system.

[0013] In an optional embodiment, the main control module includes: An initialization management unit, configured to initialize image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model thresholds based on hardware system performance parameters and defect type characteristics; A detection unit, which is used to input the optimal feature subset into the detection model for inference, output preliminary defect detection results, including the location coordinates, type and confidence of the defect, preset defect confidence threshold, screen the preliminary detection results, and obtain effective defect detection results; A feedback unit is used to determine an evaluation index, compare the evaluation index with a defect confidence threshold, locate the problem link in the detection process, and generate a corresponding feedback signal; An optimization unit is used to adjust image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model parameters based on the generated feedback signal, and repeatedly execute the process from image acquisition to defect detection and classification according to the adjusted parameters, output the final detection results, and realize iterative optimization of the detection model and feature extraction method.

[0014] In an optional embodiment, the data processing module includes: A data storage unit, the data storage unit being used to store the original image in a database and associate the annotation information with the original image according to the acquisition time of the original image and the workpiece ID; A preprocessing unit, configured to perform denoising on the original image using an adaptive filtering algorithm based on the original image stored in the database to obtain a denoised image; The feature processing unit is used to enhance the image contrast by using the CLAHE algorithm according to the grayscale distribution characteristics of the denoised image to obtain an enhanced image, and then use the edge segmentation algorithm to extract the region of interest (ROI) of the workpiece. Based on the extracted region of interest (ROI), texture features, shape features, edge features and grayscale features are extracted respectively. Based on the extracted multidimensional features, feature screening is performed by using the Relief-F algorithm to obtain the optimal feature subset.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This proposal proposes a method for detecting defects in cut workpieces based on image feature feedback. By acquiring the process characteristics of the cut workpiece and building an adaptive hardware system, and initializing key parameters such as image acquisition and preprocessing, it achieves a precise match between the detection system and the workpiece characteristics, ensuring that everything from hardware to parameter settings meets the defect detection requirements of the cut workpiece, providing a stable and reliable foundation for subsequent full-process inspection. By synchronously capturing multi-view original images based on the production line beat, dynamically adjusting exposure time based on clarity, and standardizing image naming and storage, it achieves efficient and complete image acquisition, covers complex workpiece surfaces, and retains original pixel information, providing high-quality data support for defect feature extraction. This proposal proposes a method for detecting defects in cutting workpieces based on image feature feedback. By distinguishing noise types and employing a targeted filtering algorithm, combined with the CLAHE algorithm for contrast enhancement, the method then uses OTSU segmentation and the Canny operator to extract ROIs. This method effectively eliminates image noise and significantly enhances defect features, precisely focusing on the workpiece inspection area and reducing the impact of background interference on subsequent analysis. By extracting four types of features—texture, shape, edge, and grayscale—and selecting the optimal subset using the Relief-F algorithm, the method achieves comprehensive capture and efficient simplification of defect features. This method not only encompasses the multiple attributes of defects but also eliminates redundant information, improving the ability of features to distinguish defects. This proposal proposes a method for detecting defects in cutting workpieces based on image feature feedback. By constructing a balanced sample set to train an SVM model and combining it with a confidence threshold to screen the detection results, it achieves accurate identification and effective judgment of defects, reduces misjudgments caused by sample imbalance, and retains multi-dimensional information of defects, providing complete data for subsequent analysis. By calculating indicators such as accuracy, false detection rate, and missed detection rate and generating targeted feedback signals, it achieves scientific evaluation of detection results and precise location of problems, provides a clear direction for system optimization, and promotes continuous improvement of the detection process. By analyzing the feedback signals, it dynamically adjusts parameters such as image acquisition, feature extraction, and model training. After cyclic verification, it ensures that the indicators meet the standards, realizes closed-loop iterative optimization of the detection system, and enables the system to adapt to complex working conditions in the long term and maintain stable detection performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a cutting workpiece defect detection method based on image feature feedback proposed by the present invention; Figure 2 This is a data initialization flow chart of the present invention; Figure 3 This is a flowchart of image acquisition in the present invention; Figure 4 This is a system framework diagram of a cutting workpiece defect detection system based on image feature feedback proposed in the present invention. DETAILED DESCRIPTION

[0017] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0018] Reference Figure 1 - Figure 4 As shown, a method for detecting defects in cutting workpieces based on image feature feedback includes: Acquire the process characteristics of the cutting workpiece, simultaneously build a preset defect type library, and build a hardware system including an acquisition module, mechanical structure, and computing unit. Based on the performance parameters of the hardware system and the characteristics of the defect type, initialize the image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model thresholds; Based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system, the original images of the cut workpiece at multiple viewing angles are synchronously acquired. The original images are stored in the database and associated with the annotation information according to the acquisition time and workpiece ID of the original images. Based on the original image stored in the database, the adaptive filtering algorithm is used to denoise the original image to obtain a denoised image. According to the grayscale distribution characteristics of the denoised image, the image contrast is enhanced by the CLAHE algorithm to obtain an enhanced image. Then, the edge segmentation algorithm is used to extract the region of interest (ROI) of the workpiece. Based on the extracted region of interest (ROI), texture features, shape features, edge features, and grayscale features are extracted respectively. Based on the extracted multidimensional features, feature screening is performed using the Relief-F algorithm to obtain the optimal feature subset. The optimal feature subset is input into the detection model for inference, and the preliminary detection results of the defects are output, including the location coordinates, type and confidence level of the defects. The defect confidence threshold is preset, and the preliminary detection results are screened to obtain effective defect detection results. Determine the evaluation indicators, compare the evaluation indicators with the defect confidence threshold, locate the problem links in the detection process, and generate corresponding feedback signals; Based on the generated feedback signal, the image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model parameters are adjusted respectively. According to the adjusted parameters, the process from image acquisition to defect detection and classification is repeated, the final detection results are output, and iterative optimization of the detection model and feature extraction method is achieved.

[0019] Furthermore, the process characteristics of the cutting workpiece are obtained, a preset defect type library is built simultaneously, and a hardware system including an acquisition module, mechanical structure and computing unit is built. According to the performance parameters of the hardware system and the characteristics of the defect type, the image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model thresholds are initialized, including: Obtain the process characteristics of the cutting workpiece and call the specific definitions and characteristic parameters of surface defects, size defects and morphological defects from the preset defect type library; Determine the performance requirements of the acquisition module and build an image acquisition subsystem using an industrial camera, a lens adapted to the workpiece size, and a coaxial light source; Based on the size and shape complexity of the workpiece, a workpiece positioning table with rotation / translation axes and a trigger device linked to the production line were designed to construct a mechanical structure subsystem. The angular accuracy of the rotation axis met α≤0.1°, and the displacement accuracy of the translation axis met β≤0.01mm. Based on the real-time requirements of image processing and model inference, GPU servers are configured as computing units to ensure that their floating-point computing capability is ≥10 TFLOPS. Based on the performance parameters of the hardware system, the image acquisition parameters are initialized, including: exposure time t0, which is calculated based on the workpiece surface reflectivity as t0 = k × (1 / ρ), where k is the illumination coefficient and ρ is the surface reflectivity, with a value range of 0.2 ≤ ρ ≤ 0.8. The gain g0 is set to an initial value of 1.0; Initialize the filter window size s0 in the preprocessing parameters according to the scale range of the defect characteristics, satisfying s0=2×floor (d / 2)+1, where d is the estimated diameter of the minimum defect and floor (・) is the floor rounding function; Based on the feature dimension of the defect, initialize the feature extraction dimension m0, which includes the number of texture features m1, the number of shape features m2, the number of edge features m3, and the number of grayscale features m4, satisfying m0=m1+m2+m3+m4, and m1≥5, m2≥3, m3≥2, and m4≥2; The initial confidence threshold of the detection model is set to θ0 = 0.8 for preliminary screening of defect detection results.

[0020] Specifically, the process characteristics of the cutting workpiece are obtained, a preset defect type library is simultaneously constructed, and a hardware system including an acquisition module, a mechanical structure, and a computing unit is built. According to the performance parameters of the hardware system and the defect type characteristics, the image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model thresholds are initialized to achieve a comprehensive construction of the cutting workpiece defect detection system and the initialization of key parameters, laying a solid foundation for the subsequent efficient and accurate defect detection process. In terms of system construction, based on the process characteristics of the cutting workpiece and the preset defect type library, the definitions and characteristic parameters of various defects such as surface, size, and morphology are obtained, and then the required performance of the image acquisition module is clarified, and an adaptive image acquisition subsystem is built; based on the size and shape complexity of the workpiece, a mechanical structure subsystem with high-precision rotation / translation axes is designed, and a trigger device that can be linked to the production line is constructed; at the same time, according to the real-time requirements, a GPU server with powerful floating-point computing capabilities is configured as a computing unit, realizing the full construction of the hardware system from image acquisition, workpiece positioning to data calculation and processing.

[0021] The advantages are: in terms of parameter initialization, it combines hardware performance to scientifically initialize image acquisition parameters, such as accurately calculating exposure time based on the reflectivity of the workpiece surface and reasonably setting gain; initializing the preprocessing filter window size based on the defect feature scale to ensure that defects of different scales can be effectively processed; based on the defect feature dimension, it rationally plans the feature extraction dimension, covering multiple types of features such as texture, shape, edge, and grayscale, and sets the initial confidence threshold of the detection model. The implementation of these functions has significant advantages, enabling the entire detection system to be highly adapted to the characteristics and detection requirements of the cutting workpiece, improving image acquisition quality, enhancing the accuracy and comprehensiveness of defect feature extraction, and optimizing the initial screening effect of the detection model, thereby greatly improving the accuracy and efficiency of defect detection, reducing the probability of false detection and missed detection, and effectively ensuring the quality inspection of cutting workpieces.

[0022] Furthermore, based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system, the original images of the cut workpiece from multiple perspectives are synchronously acquired. The original images are stored in the database and associated with the annotation information according to the acquisition time and workpiece ID of the original images, including: Based on the production line's takt time T, the image acquisition interval Δt=T is set by the trigger device to ensure synchronization with the workpiece transfer rhythm; When the workpiece arrives at the inspection station, the trigger device sends a trigger signal to the acquisition module to start the multi-view image acquisition process; Control the rotating shaft in the mechanical structure subsystem to drive the workpiece to rotate at an angle θ i =i×(360° / n), where i=1, 2, ..., n, and n is the number of viewing angles, satisfying 3≤n≤6; At each viewing angle, the industrial camera collects raw images according to the initialized exposure time t0 and gain g0. The image format is set to BMP or TIFF to preserve the original pixel information. During the collection process, , real-time monitoring of image clarity evaluation index Q, where For the image The gray value at Q<Q min When Q is ≥ Q, the exposure time t=t0+Δt is automatically adjusted until Q≥Q min , Q min is the preset clarity threshold; The collected multi-view original images are named according to the naming rule of workpiece ID - view serial number - acquisition time; Based on the database storage specifications, the original image is stored in the distributed database, and an associated index between the workpiece ID and the image file path is established; Annotation information is added to each original image, including the material of the workpiece, processing technology type, tool model, cutting parameters, and the ambient temperature and humidity during acquisition, to form complete image metadata.

[0023] Specifically, the entire process of cutting workpiece image acquisition is precisely controlled and standardized. In terms of acquisition timing, based on the production line takt time, the image acquisition interval is precisely controlled through a trigger device, so that the acquisition rhythm is synchronized with the workpiece transmission, ensuring that the workpiece is captured as soon as it arrives. In terms of multi-view acquisition, the rotation axis of the mechanical structure subsystem is used to capture the image at specific angle intervals (θ i =i×(360° / n), 3≤n≤6) rotates the workpiece to obtain multi-dimensional perspective images, covering different surface areas of the workpiece; when the acquisition parameters are executed, the industrial camera works according to the initialized exposure time t0 and gain g0, retains the original pixel information in BMP or TIFF format, and passes the clarity evaluation index , I(x,y) is the image grayscale value) is monitored in real time, and the exposure time is dynamically adjusted to ensure image clarity; in the image management stage, multi-view original images are named according to rules, stored in a distributed database, and an associated index between the workpiece ID and the path is established. In addition, the images are supplemented with annotation information such as material, process, cutting parameters, and environmental data to build a complete metadata system.

[0024] The advantages are: focusing on acquisition accuracy, image quality assurance and data standardization. The acquisition rhythm is synchronized with the production line, and the trigger device and beat adaptation are used to avoid missed acquisition and wrong acquisition, which meets the needs of efficient detection of industrial production lines; multi-view acquisition combined with mechanical rotation control breaks through the limitations of a single perspective, covers the complex surface of the workpiece, provides a comprehensive image basis for subsequent defect detection, and reduces the risk of missed detection due to missing perspectives; in terms of image quality, through initialization parameter execution and dynamic adjustment of clarity, it balances the retention of original information and clear imaging requirements, BMP / TIFF format retains pixel details, and real-time monitoring and exposure adjustment deal with interference such as reflection and motion blur on the workpiece surface; data management specifications, naming rules, database storage and metadata annotation make image data traceable and easy to search, and associate the entire process information of the workpiece, assist in subsequent defect analysis and process optimization, lay the foundation for data interoperability between the inspection system and the production management system, and improve the overall quality control efficiency.

[0025] Furthermore, based on the original image stored in the database, an adaptive filtering algorithm is used to denoise the original image to obtain a denoised image. According to the grayscale distribution characteristics of the denoised image, the image contrast is enhanced by the CLAHE algorithm to obtain an enhanced image. Then, an edge segmentation algorithm is used to extract the region of interest (ROI) of the workpiece, which specifically includes: Read the original image I from the database and calculate the noise type judgment index N of the image; Among them, the calculation formula of noise type judgment index N is: , median (・) is the median calculation function. When N<N1, it is judged as slight noise, and when N≥N1, it is judged as impulse noise. N1 is the preset noise threshold; If the noise is slight, use Gaussian filtering algorithm to remove the noise: The filter kernel function is , where σ is the Gaussian standard deviation, the value is σ=0.5×s0, s0 is the size of the initialized filter window, and the denoised image I1=I*G is obtained, which represents the convolution operation; If it is impulse noise, use the median filter algorithm to denoise: Denoised image I1(x,y)=median(I(xs:x+s,ys:y+s)), where s=floor(s0 / 2), is obtained. Calculate the grayscale histogram H (k) of the denoised image I1, k = 0, 1, ..., 255, where H (k) is the number of pixels with grayscale value k; If the image grayscale distribution is concentrated, the CLAHE algorithm is used to enhance the contrast and calculate the cumulative distribution function. ,in is the total number of pixels in the image; The enhanced gray value k'=floor (255×CDF (k)) is obtained to obtain the enhanced image I2; If the image has insufficient local contrast, the CLAHE algorithm is used to enhance the contrast. The image is divided into 8×8 sub-blocks. The histogram of each sub-block is calculated and truncation is performed. The truncation threshold is T=α×mean(H(k)), where α is the adjustment coefficient and ranges from 1.0≤α≤3.0, and mean(·) is the mean function. The sub-blocks are spliced ​​together by bilinear interpolation to obtain the enhanced image I2. The OTSU threshold segmentation algorithm is used to separate the background of the enhanced image I2 and calculate the inter-class variance: ,in 、 are the ratio of foreground and background pixels, 、 The grayscale mean of the foreground and background are respectively, and the threshold t corresponding to the maximum inter-class variance is found to segment the image into foreground and background; Combined with the edge segmentation algorithm, the Canny operator is used to extract the edge contour of the foreground area; Among them, the high threshold T of edge detection H =0.3×max (G), low threshold T L=0.15×max (G), G is the image gradient amplitude, and the region of interest (ROI) of the workpiece is determined by the edge contour. The coordinate range of the ROI is (xmin, xmax, ymin, ymax).

[0026] Specifically, based on the original image stored in the database, an adaptive filtering algorithm is used to denoise the original image to obtain a denoised image. According to the grayscale distribution characteristics of the denoised image, the image contrast is enhanced by the CLAHE algorithm to obtain an enhanced image. Then, an edge segmentation algorithm is used to extract the region of interest (ROI) of the workpiece, realizing the preprocessing of the cutting workpiece image and the extraction of the region of interest (ROI). The core is to eliminate interference, enhance defect features and focus on the detection target through multi-step processing. Specifically, the method includes: first reading the original image from the database, distinguishing slight noise from impulse noise by calculating the noise type judgment index N (based on the sum of the differences between pixel grayscale and median), and performing targeted denoising using Gaussian filtering (based on a specific kernel function and standard deviation) and median filtering (based on the initialization window size) to obtain a denoised image; then, through grayscale histogram analysis, for images with concentrated grayscale distribution or insufficient local contrast, the CLAHE algorithm (the former directly calculates the cumulative distribution function, while the latter divides sub-blocks and truncates the histogram) is used to enhance the contrast and generate an enhanced image; finally, the OTSU threshold segmentation algorithm (determines the optimal threshold by calculating the inter-class variance) is used to separate the foreground and background, and the Canny operator (based on the high and low thresholds of the gradient amplitude) is combined to extract the edge contour, and finally the ROI coordinate range of the workpiece is determined.

[0027] The advantages are: taking into account both the targetedness and accuracy of pre-processing, laying a high-quality data foundation for subsequent feature extraction and defect detection. First, adaptive denoising, by judging the matched filter algorithm by noise type, retains edge details (median filtering for impulse noise) and smoothes high-frequency interference (Gaussian filtering for slight noise), avoiding feature loss caused by a single denoising method; second, intelligent contrast enhancement, the CLAHE algorithm combines global and local processing logic to improve the overall grayscale distribution concentration problem, and solves the local lack of contrast through sub-block truncation and interpolation, effectively amplifying the grayscale difference between defects (such as cracks and burrs) and normal areas; third, precise positioning of the ROI, the combination of OTSU segmentation and Canny edge detection can efficiently separate the workpiece from the background, eliminate interference from fixtures, workbenches, etc., ensure that subsequent analysis focuses on the effective area, and reduce the impact of redundant information on detection accuracy.

[0028] Furthermore, based on the extracted region of interest (ROI), texture features, shape features, edge features, and grayscale features are extracted respectively. Based on the extracted multidimensional features, feature screening is performed using the Relief-F algorithm to obtain the optimal feature subset, which specifically includes: In the region of interest (ROI), texture features are extracted and texture parameters are calculated using the gray-level co-occurrence matrix (GLCM). For distance d = 1, angle The GLCM matrix , calculate the energy characteristics , entropy characteristics , contrast feature , correlation characteristics ,in 、 is the row and column mean, 、 is the row and column standard deviation; The LBP algorithm is used to extract local binary pattern features, and the LBP histogram is statistically analyzed to obtain the texture feature vector F1=[E,Ent,Con,Cor,...]; Extract shape features, extract contours of defect candidate areas within ROI, and , calculate the area of ​​the contour, where R is the defect area; pass , determine the contour perimeter, where x(i) and y(i) are the coordinates of the contour points and m is the number of contour points; The circularity of the contour is determined by C=4πA / L², and the aspect ratio of the contour is determined by R=W / H, where W and H are the width and height of the defect's circumscribed rectangle, respectively. The shape feature vector F2=[A,L,C,R,...] is obtained; Extract edge features and apply Sobel operator to ROI to calculate horizontal and vertical gradients 、 , gradient amplitude , gradient direction , obtain edge direction features by statistical gradient direction histogram, and calculate edge continuity features ,in is the length of the continuous edge segment, is the number of edge segments, and the edge feature vector is obtained , is the mean gradient amplitude, is the gradient direction mode; Extract grayscale features by , calculate the grayscale mean within the ROI, is the number of pixels within the ROI; pass , determine the grayscale variance; pass , determine the grayscale entropy, where p (k) is the probability that the grayscale value in the ROI is k, and obtain the grayscale feature vector ; The texture feature F1, shape feature F2, edge feature F3 and grayscale feature F4 are spliced ​​into a multidimensional feature matrix , the dimension is m0×1, m0 is the initial feature extraction dimension; The Relief-F algorithm is used to calculate the weight w(j) of each feature and update it iteratively: ;in, , is the nearest neighbor sample of the same type, is the nearest neighbor sample of different classes, n is the number of samples, p (c) is the prior probability of class c, and the weight w (j) ≥ w min The features of min is the weight threshold.

[0029] Specifically, this step achieves comprehensive extraction and optimized screening of defect features in cutting workpieces. The core is to extract features that can effectively distinguish defects from normal areas from the region of interest (ROI), and to select the most recognizable feature subset through an algorithm. Specifically, within the ROI, the gray-level co-occurrence matrix (GLCM) is used to calculate texture features such as energy, entropy, contrast, and correlation, and the LBP algorithm is used to extract local binary pattern features to form a texture feature vector; contour extraction is used to calculate shape features such as area, perimeter, circularity, and aspect ratio to form a shape feature vector; gradient information is obtained through the Sobel operator, and edge features such as edge direction and continuity are calculated to form an edge feature vector; grayscale features are extracted through parameters such as grayscale mean, variance, and entropy to form a grayscale feature vector; after splicing the four types of features into a multidimensional feature matrix, the Relief-F algorithm is used to iteratively calculate the weight of each feature, and the features with weights ≥ the threshold are selected to form the optimal feature subset.

[0030] Advantages lie in the comprehensiveness of feature extraction and targeted screening, providing high-quality input for defect detection models. The diverse feature dimensions cover four categories: texture (reflecting surface roughness and uniformity), shape (distinguishing defect geometry), edge (capturing contour features such as cracks and burrs), and grayscale (reflecting regional brightness differences). These features adapt to the physical characteristics of different defect types (such as oxidation spots, cracks, and burrs). Furthermore, algorithms such as GLCM, LBP, and Sobel are highly targeted, accurately capturing subtle defect features (such as edge continuity of tiny cracks and the aspect ratio of burrs). The Relief-F algorithm iteratively updates feature weights, retaining features highly relevant to defect classification and eliminating redundant information. This reduces the computational complexity of subsequent models while improving the robustness and accuracy of defect recognition.

[0031] Furthermore, the optimal feature subset is input into the detection model for inference, and the preliminary defect detection results are output, including the location coordinates, type and confidence of the defect. The defect confidence threshold is preset, and the preliminary detection results are screened to obtain effective defect detection results, including: Based on the optimal feature subset X' obtained by screening, a detection sample set is constructed, where the positive sample is a feature vector containing defects, and the negative sample is a feature vector without defects. The sample number ratio satisfies 1:1≤positive sample:negative sample≤3:1; Select the SVM model for training and construct the kernel function: K (x, y) = exp (-γ||xy||²); where γ is the kernel parameter. The optimal parameters C and γ are determined through cross-validation, where C is the penalty coefficient, with a value of 1≤C≤100, and γ is 0.01≤γ≤10. Train the SVM model to obtain the classification hyperplane: ;in, is the Lagrange multiplier, is the sample label, is the bias term, is the number of support vectors; Input the optimal feature subset X' into the trained SVM model and output the predicted category of the defect And the decision value d, confidence c=1 / (1+exp (-d)), get the preliminary detection result; Extract the confidence level c from the preliminary inspection results and compare it with the preset defect confidence level threshold θ0; When c≥θ0, it is determined to be a valid defect, and the location coordinates, type and confidence information of the defect are retained; When c<θ0, it is determined to be an invalid defect and is eliminated; The information of all valid defects is summarized to form a list of valid defect detection results, where each defect entry contains a unique identifier ID, location coordinates (x, y), type label t, size parameters (s1, s2) and confidence c.

[0032] Specifically, this step achieves the detection and screening of defects in cut workpieces. The core is to use the SVM model to classify and infer the optimal feature subset and output valid defect results. This includes: constructing a test sample set based on the optimal feature subset, controlling the ratio of positive (defective) to negative (non-defective) samples between 1:1 and 3:1; selecting the SVM model, constructing a Gaussian kernel function K (x,y) = exp (-γ||xy||²), and combining cross-validation to determine the penalty coefficient C (1≤C≤100) and kernel parameter γ (0.01≤γ≤10), and training the classification hyperplane; inputting the optimal feature subset into the trained model, outputting the defect prediction category, decision value d, and confidence level c = 1 / (1+exp(−d)); filtering the results using a preset confidence threshold θ0, retaining valid defects with a confidence level greater than θ0 and eliminating invalid defects, ultimately forming a result list containing the defect ID, location, type, size, and confidence level.

[0033] The advantages lie in the targeted nature of the detection model and the reliability of its results, providing a precise basis for defect identification. By controlling the balance between positive and negative samples, the model avoids bias due to sample imbalance and improves its sensitivity to defect detection. The model also boasts strong adaptability. The SVM combined with the Gaussian kernel function effectively processes high-dimensional feature spaces. Parameter optimization through cross-validation enhances the classification capability for complex defect features (such as mixed multi-class defect scenarios). Low-confidence results are eliminated based on confidence thresholds to reduce false detections, while retaining multi-dimensional defect information (location, type, etc.), providing complete data support for subsequent analysis and feedback, and enhancing the practical value of the detection results.

[0034] Furthermore, we determine the evaluation indicators, compare them with the defect confidence threshold, locate the problem links in the detection process, and generate corresponding feedback signals, including: Collect known real defect data from the production line as a baseline, including the number of real defects , location, type and size information; Based on the effective defect detection results, count the number of real defects detected , that is, the number of defects with position errors less than ε between the detection results and the real defects and the same type, ε is the position error threshold, which satisfies 0.05mm≤ε≤0.1mm; Count the number of falsely detected defects , that is, the number of real defects that do not exist in the test results; Count the number of missed defects ; Calculating detection accuracy , where Acc∈[0,100%]; Calculating the false positive rate ,in is the total number of detection results; Calculate the false negative rate ; Furthermore, set the preset thresholds for evaluation indicators: accuracy threshold Acc0 = 99%, false positive rate threshold FPR0 = 1%, false negative rate threshold FNR0 = 1% (which can be preset by experienced personnel); When Acc < Acc0 and FNR ≥ FNR0, analyze the reasons for missed detections. If the missed detection defects are mostly detailed features such as micro cracks, it is determined that the image quality is insufficient, and a feedback signal S1 of "adjust image acquisition parameters" is generated; When FPR > FPR0, analyze the reasons for false detections. If the false detections are mostly caused by insufficient feature discrimination, it is determined that there is redundancy in the feature extraction process, and a feedback signal S2 of "optimize feature extraction dimension" is generated; When the detection accuracy of a specific type of defect continuously drops below 80%, it is determined that the generalization ability of the detection model for this type of defect is insufficient, and a feedback signal S3 of "increase defect sample training" is generated; When there are repeated detections or missed detection areas in multi-view images, it is determined that the image acquisition perspective is insufficient, and a feedback signal S4 of "adjust multi-view acquisition parameters" is generated; When the preprocessed image still has obvious noise interfering with feature extraction, it is determined that the preprocessing parameter settings are unreasonable, and a feedback signal S5 of "optimize preprocessing filtering parameters" is generated; Summarize the generated feedback signals to form a feedback signal set S = {S1, S2, S​​​​​The advantages are: the evaluation indicators are quantitative and comprehensive, and the system performance is measured in multiple dimensions such as accuracy, false detection rate, and missed detection rate, avoiding the one-sidedness caused by a single indicator. The threshold can be preset by experienced personnel to adapt to the needs of different production scenarios; the feedback signal is highly targeted, and through the correlation analysis between defect types and detection results, the problem is accurately located in specific links such as image acquisition, feature extraction, and model training, providing a clear direction for subsequent parameter adjustments and avoiding blind optimization; forming a closed loop of quality control, and promoting continuous iteration of the system through dynamic comparison of real defect data and detection results, which is especially suitable for batch production scenarios of complex cutting workpieces, and can effectively reduce production losses caused by missed and false detection of defects.

[0037] Furthermore, based on the generated feedback signal, the image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model parameters are adjusted respectively. Based on the adjusted parameters, the process from image acquisition to defect detection and classification is repeated, the final detection results are output, and iterative optimization of the detection model and feature extraction method is achieved. Specifically, the following steps are performed: Analyze the feedback signal set S to determine the parameter type and adjustment direction that need to be adjusted; If there is a feedback signal S1 (adjust image acquisition parameters), adjust the exposure time according to the grayscale characteristics of the missed defects ,in is the target grayscale mean, is the grayscale mean of the current missed detection area; if the image is blurred, reduce the gain , and increase the light source intensity L1=L0×1.2, where L0 is the initial light source intensity; If there is a feedback signal S2 (optimizing feature extraction dimension), based on the feature weights of the Relief-F algorithm, increase the number of features with weight w (j) ≥ 0.8 and reduce the number of features with weight w (j) < 0.2. The adjusted feature extraction dimension m1 = m0 + Δm, where Δm = 2 × count (w (j) ≥ 0.8) - count (w (j) < 0.2); If there is a feedback signal S3 (increase defect sample training), collect the defect images and feature data of missed or misdetected defects, expand the training sample set, and the number of new samples meets , The number of original samples is used; the detection model is updated using the transfer learning method, the backbone network parameters are frozen, the head network parameters are fine-tuned, the learning rate is set to 1 / 10 of the original learning rate, and the number of training iterations is increased by 50 rounds; If there is a feedback signal S4 (adjust the multi-view acquisition parameters), increase the number of view angles corresponding to the missed detection area, and adjust the angle interval of the rotation axis Δθ=360° / (n+Δn view ), where Δn view The number of viewing angles increased, satisfying 1≤Δnview ≤2; optimize the synchronization timing of the trigger device to ensure clear image acquisition in areas with high defect incidence; If there is a feedback signal S5 (optimize preprocessing filter parameters), adjust the filter window size s1 according to the noise type. For impulse noise, s1=s0+2; for Gaussian noise, increase the Gaussian standard deviation σ1=1.2×σ0; adjust the truncation threshold α1=α0+0.5 of the CLAHE algorithm to enhance local contrast; According to the adjusted parameters, the image acquisition step is re-executed to acquire new multi-view original images; Perform preprocessing, feature extraction, defect detection and classification processes on the newly acquired original images to obtain updated effective defect detection results; Calculate the updated evaluation indicators Acc1, FPR1, and FNR1, and compare them with the preset thresholds. If Acc1 ≥ Acc0, FPR1 ≤ FPR0, and FNR1 ≤ FNR0, save the adjusted parameters; otherwise, repeat the parameter adjustment process until the evaluation indicator requirements are met. Output the final inspection results, displaying images and text reports with defect location annotations in a visual format, including workpiece ID, inspection time, defect statistics (number, type and percentage), and judgment results (pass / fail); Establish a parameter adjustment log to record the parameter name, values ​​before and after adjustment, feedback signal type, and evaluation index changes for each adjustment for subsequent process analysis; regularly (e.g., weekly) iteratively optimize the detection model and feature extraction method, incorporating new defect types and process change data to ensure the long-term stability of the detection system.

[0038] Specifically, dynamic optimization and closed-loop iteration of detection system parameters based on feedback signals are achieved. The core is to accurately adjust system parameters by analyzing the feedback signal set, and ensure that detection performance meets the standards through cyclic verification. Specifically, it includes: parsing the feedback signal set S to identify the parameter categories and directions that need to be adjusted; taking targeted measures for different signals - S1 adjusting image acquisition parameters (such as optimizing exposure time, gain and light source intensity according to the grayscale characteristics of the missed detection area), S2 optimizing feature extraction dimensions (increasing or decreasing the number of features based on the Relief-F weight), S3 expanding samples and updating the model through transfer learning (increasing missed detection / false detection samples, fine-tuning network parameters), S4 adjusting multi-view acquisition parameters (increasing the number of viewing angles, optimizing the rotation angle interval), S5 optimizing preprocessing filter parameters (adjusting the window size, standard deviation and contrast enhancement threshold according to the noise type); after adjustment, re-execute the image acquisition, preprocessing, feature extraction and detection processes, calculate the updated evaluation indicators (Acc1, FPR1, FNR1), compare them with the preset thresholds, and repeat the adjustment until the requirements are met if they are not met; finally, output visual inspection results, establish a parameter adjustment log, and regularly iterate the model and feature extraction method to adapt to new defect types and process changes.

[0039] The advantages lie in the establishment of a closed-loop optimization mechanism of "feedback-adjustment-verification" to ensure that the inspection system continuously adapts to production needs. First, parameter adjustments are highly targeted, directly locating problematic links (such as image quality and feature dimensions) through feedback signals. Adjustment measures are closely linked to defect characteristics (such as missed grayscale and noise type), avoiding blind optimization. Second, the verification mechanism is rigorous. By repeatedly executing the entire process and comparing evaluation indicators, the adjustment effects are quantifiable and traceable, ensuring that the system's performance is stable and meets standards. Third, it has strong long-term adaptability. The parameter adjustment log provides data support for process analysis, and the regular iteration mechanism can incorporate new defect types and process changes, enabling the system to maintain long-term robustness in the inspection of complex cut workpieces and effectively balancing inspection accuracy and production efficiency.

[0040] Furthermore, a cutting workpiece defect detection system based on image feature feedback is proposed, which is used to implement any of the above detection methods, including: The acquisition module is used to obtain the process characteristics of the cutting workpiece and simultaneously build a preset defect type library. It is used to synchronously acquire the original images of the cutting workpiece from multiple perspectives based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system; Main control module, which is used to initialize image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model thresholds based on the performance parameters of the hardware system and defect type characteristics, input the optimal feature subset into the detection model for inference, and output preliminary defect detection results, including the location coordinates, type and confidence of the defects. It presets a defect confidence threshold, screens the preliminary detection results, obtains effective defect detection results, and determines evaluation indicators. It compares the evaluation indicators with the defect confidence threshold, locates the problem links in the detection process, and generates corresponding feedback signals. It is also used to adjust the image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model parameters based on the generated feedback signals, and repeats the process from image acquisition to defect detection and classification according to the adjusted parameters, outputs the final detection results, and realizes iterative optimization of the detection model and feature extraction method; The data processing module is used to store the original image in a database and associate the annotation information based on the acquisition time of the original image and the workpiece ID. The module is used to use an adaptive filtering algorithm to denoise the original image based on the original image stored in the database to obtain a denoised image. The module is also used to enhance the image contrast using the CLAHE algorithm based on the grayscale distribution characteristics of the denoised image to obtain an enhanced image. The edge segmentation algorithm is then used to extract the region of interest (ROI) of the workpiece. Based on the extracted region of interest (ROI), texture features, shape features, edge features, and grayscale features are extracted respectively. Based on the extracted multidimensional features, the module performs feature screening using the Relief-F algorithm to obtain the optimal feature subset. The display module is used to present the process and results of the cutting workpiece defect detection to the user.

[0041] Furthermore, the acquisition module includes: The first acquisition unit is used to obtain the process characteristics of the cutting workpiece and simultaneously build a preset defect type library; The second acquisition unit is used to synchronously acquire original images of the cutting workpiece under multiple viewing angles based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system.

[0042] Furthermore, the main control module includes: Initialization management unit, which is used to initialize image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model thresholds based on the performance parameters of the hardware system and defect type characteristics; The detection unit is used to input the optimal feature subset into the detection model for inference, output the preliminary detection results of the defect, including the location coordinates, type and confidence of the defect, preset the defect confidence threshold, screen the preliminary detection results, and obtain the effective defect detection results; Feedback unit: The feedback unit is used to determine the evaluation index, compare the evaluation index with the defect confidence threshold, locate the problem link in the detection process, and generate the corresponding feedback signal; The optimization unit is used to adjust the image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model parameters based on the generated feedback signal. According to the adjusted parameters, the process from image acquisition to defect detection and classification is repeatedly executed to output the final detection results and realize iterative optimization of the detection model and feature extraction method.

[0043] Furthermore, the data processing module includes: A data storage unit, the data storage unit is used to store the original image in a database and associate the annotation information according to the acquisition time of the original image and the workpiece ID; The preprocessing unit is used to perform denoising on the original image based on the original image stored in the database by using an adaptive filtering algorithm to obtain a denoised image; The feature processing unit is used to enhance the image contrast by the CLAHE algorithm according to the grayscale distribution characteristics of the denoised image to obtain an enhanced image. Then, the edge segmentation algorithm is used to extract the region of interest (ROI) of the workpiece. Based on the extracted region of interest (ROI), texture features, shape features, edge features and grayscale features are extracted respectively. Based on the extracted multi-dimensional features, the Relief-F algorithm is used to perform feature screening to obtain the optimal feature subset.

[0044] The advantages of the present invention are: from hardware construction to parameter initialization, it is in line with the process characteristics and defect types of the cutting workpiece, multi-view synchronous acquisition combined with the production line rhythm to ensure image integrity, and adaptive denoising, CLAHE enhancement and ROI extraction to improve data quality; through GLCM, LBP, Sobel operators and other operators to extract texture, shape, edge, grayscale multi-dimensional features, the Relief-F algorithm to screen the optimal subset, combined with the SVM model and confidence threshold to achieve accurate detection; with the help of accuracy, false detection rate, missed detection rate and other indicators evaluation, targeted feedback signals are generated to dynamically adjust parameters, forming a "detection-evaluation-optimization" closed loop, which is both systematic, accurate and long-term adaptable, and can effectively meet the defect detection needs of complex cutting workpieces.

[0045] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting defects in cutting workpieces based on image feature feedback, characterized in that: include: Acquire the process characteristics of the cutting workpiece, simultaneously build a preset defect type library, and build a hardware system including an acquisition module, mechanical structure, and computing unit. Based on the performance parameters of the hardware system and the characteristics of the defect type, initialize the image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model thresholds; Based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system, the original images of the cut workpiece at multiple viewing angles are synchronously acquired. The original images are stored in the database and associated with the annotation information according to the acquisition time and workpiece ID of the original images. Based on the original image stored in the database, the adaptive filtering algorithm is used to denoise the original image to obtain a denoised image. According to the grayscale distribution characteristics of the denoised image, the image contrast is enhanced by the CLAHE algorithm to obtain an enhanced image. Then, the edge segmentation algorithm is used to extract the region of interest (ROI) of the workpiece. Based on the extracted region of interest (ROI), texture features, shape features, edge features, and grayscale features are extracted respectively. Based on the extracted multidimensional features, feature screening is performed using the Relief-F algorithm to obtain the optimal feature subset. The optimal feature subset is input into the detection model for inference, and the preliminary detection results of the defects are output, including the location coordinates, type and confidence level of the defects. The defect confidence threshold is preset, and the preliminary detection results are screened to obtain effective defect detection results. Determine the evaluation indicators, compare the evaluation indicators with the defect confidence threshold, locate the problem links in the detection process, and generate corresponding feedback signals; Based on the generated feedback signal, the image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model parameters are adjusted respectively. According to the adjusted parameters, the process from image acquisition to defect detection and classification is repeated, the final detection results are output, and iterative optimization of the detection model and feature extraction method is achieved.

2. The method for detecting defects in cutting workpieces based on image feature feedback according to claim 1, characterized in that: The process characteristics of the cutting workpiece are obtained, a preset defect type library is built simultaneously, and a hardware system including an acquisition module, a mechanical structure and a computing unit is built. According to the performance parameters of the hardware system and the defect type characteristics, the image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model thresholds are initialized, specifically including: Obtain the process characteristics of the cutting workpiece and call the specific definitions and characteristic parameters of surface defects, size defects and morphological defects from the preset defect type library; Determine the performance requirements of the acquisition module and build an image acquisition subsystem using an industrial camera, a lens adapted to the workpiece size, and a coaxial light source; Based on the size and shape complexity of the workpiece, a workpiece positioning table with rotation / translation axes and a trigger device linked to the production line were designed to construct a mechanical structure subsystem. The angular accuracy of the rotation axis met α≤0.1°, and the displacement accuracy of the translation axis met β≤0.01mm. Based on the real-time requirements of image processing and model inference, GPU servers are configured as computing units to ensure that their floating-point computing capability is ≥10 TFLOPS. Based on the performance parameters of the hardware system, the image acquisition parameters are initialized, including: exposure time t0, which is calculated based on the workpiece surface reflectivity as t0 = k × (1 / ρ), where k is the illumination coefficient and ρ is the surface reflectivity, with a value range of 0.2 ≤ ρ ≤ 0.

8. The gain g0 is set to an initial value of 1.0; Initialize the filter window size s0 in the preprocessing parameters according to the scale range of the defect characteristics, satisfying s0=2×floor(d / 2)+1, where d is the estimated diameter of the minimum defect and floor (・) is the floor rounding function; Based on the feature dimension of the defect, initialize the feature extraction dimension m0, which includes the number of texture features m1, the number of shape features m2, the number of edge features m3, and the number of grayscale features m4, satisfying m0=m1+m2+m3+m4, and m1≥5, m2≥3, m3≥2, and m4≥2; The initial confidence threshold of the detection model is set to θ0 = 0.8 for preliminary screening of defect detection results.

3. The method for detecting defects in cutting workpieces based on image feature feedback according to claim 1, characterized in that: The method of synchronously capturing original images of the cut workpiece from multiple perspectives based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system, storing the original images in a database and associating them with annotation information according to the acquisition time and workpiece ID of the original images, specifically includes: Based on the production line's takt time T, the image acquisition interval Δt=T is set by the trigger device to ensure synchronization with the workpiece transfer rhythm; When the workpiece arrives at the inspection station, the trigger device sends a trigger signal to the acquisition module to start the multi-view image acquisition process; Control the rotating shaft in the mechanical structure subsystem to drive the workpiece to rotate at an angle θ i =i×(360° / n), where i=1, 2, ..., n, and n is the number of viewing angles, satisfying 3≤n≤6; At each viewing angle, the industrial camera collects raw images according to the initialized exposure time t0 and gain g0. The image format is set to BMP or TIFF to preserve the original pixel information. During the collection process, , real-time monitoring of image clarity evaluation index Q, where For the image The gray value at Q<Q min When Q is ≥ Q, the exposure time t=t0+Δt is automatically adjusted until Q≥Q min , Q min is the preset clarity threshold; The collected multi-view original images are named according to the naming rule of workpiece ID - view serial number - acquisition time; Based on the database storage specifications, the original image is stored in the distributed database, and an associated index between the workpiece ID and the image file path is established; Annotation information is added to each original image, including the material of the workpiece, processing technology type, tool model, cutting parameters, and the ambient temperature and humidity during acquisition, to form complete image metadata.

4. The method for detecting defects in cutting workpieces based on image feature feedback according to claim 1, wherein: Based on the original image stored in the database, the adaptive filtering algorithm is used to denoise the original image to obtain a denoised image. According to the grayscale distribution characteristics of the denoised image, the image contrast is enhanced by the CLAHE algorithm to obtain an enhanced image. Then, the edge segmentation algorithm is used to extract the region of interest (ROI) of the workpiece, which specifically includes: Read the original image I from the database and calculate the noise type judgment index N of the image; Among them, the calculation formula of noise type judgment index N is: , median(・) is the median calculation function. When N<N1, it is judged as slight noise, and when N≥N1, it is judged as impulse noise. N1 is the preset noise threshold; If the noise is slight, use Gaussian filtering algorithm to remove the noise: The filter kernel function is , where σ is the Gaussian standard deviation, the value is σ=0.5×s0, s0 is the size of the initialized filter window, and the denoised image I1=I*G is obtained, which represents the convolution operation; If it is impulse noise, use the median filter algorithm to denoise: Denoised image I1(x,y)=median(I(xs:x+s,ys:y+s)), where s=floor(s0 / 2), is obtained. Calculate the grayscale histogram H (k) of the denoised image I1, k = 0, 1, ..., 255, where H (k) is the number of pixels with grayscale value k; If the image grayscale distribution is concentrated, the CLAHE algorithm is used to enhance the contrast and calculate the cumulative distribution function. ,in is the total number of pixels in the image; The enhanced gray value k'=floor (255×CDF (k)) is obtained to obtain the enhanced image I2; If the image has insufficient local contrast, the CLAHE algorithm is used to enhance the contrast. The image is divided into 8×8 sub-blocks. The histogram of each sub-block is calculated and truncation is performed. The truncation threshold is T=α×mean(H(k)), where α is the adjustment coefficient and ranges from 1.0≤α≤3.0, and mean(·) is the mean function. The sub-blocks are spliced ​​together by bilinear interpolation to obtain the enhanced image I2. The OTSU threshold segmentation algorithm is used to separate the background of the enhanced image I2 and calculate the inter-class variance: ,in 、 are the ratio of foreground and background pixels, 、 The grayscale mean of the foreground and background are respectively, and the threshold t corresponding to the maximum inter-class variance is found to segment the image into foreground and background; Combined with the edge segmentation algorithm, the Canny operator is used to extract the edge contour of the foreground area; Among them, the high threshold T of edge detection H =0.3×max (G), low threshold T L =0.15×max (G), G is the image gradient amplitude, and the region of interest (ROI) of the workpiece is determined by the edge contour. The coordinate range of the ROI is (xmin, xmax, ymin, ymax).

5. The method for detecting defects in cutting workpieces based on image feature feedback according to claim 1, characterized in that: Based on the extracted region of interest (ROI), texture features, shape features, edge features, and grayscale features are extracted respectively. Based on the extracted multidimensional features, feature screening is performed using the Relief-F algorithm to obtain the optimal feature subset, which specifically includes: In the region of interest (ROI), texture features are extracted and texture parameters are calculated using the gray-level co-occurrence matrix (GLCM). For distance d = 1, angle The GLCM matrix , calculate the energy characteristics , entropy characteristics , contrast feature , correlation characteristics ,in 、 row and column means, 、 is the row and column standard deviation; LBP algorithm is used to extract local binary pattern features, and LBP histogram is used to obtain texture feature vector F1=[E, Ent, Con, Cor, ...]; shape features are extracted, and contours of defect candidate areas within ROI are extracted, and then the texture feature vector F1=[E, Ent, Con, Cor, ...] is obtained by LBP algorithm. , calculate the area of ​​the contour, where R is the defect area; by , determine the contour perimeter, where x(i) and y(i) are the coordinates of the contour points and m is the number of contour points; The circularity of the contour is determined by C=4πA / L², and the aspect ratio of the contour is determined by R=W / H, where W and H are the width and height of the defect's circumscribed rectangle, respectively. The shape feature vector F2=[A,L,C,R,...] is obtained; Extract edge features and apply Sobel operator to ROI to calculate horizontal and vertical gradients 、 , gradient amplitude , gradient direction , obtain edge direction features by statistical gradient direction histogram, and calculate edge continuity features ,in is the length of the continuous edge segment, is the number of edge segments, and the edge feature vector is obtained , is the mean gradient amplitude, is the gradient direction mode; Extract grayscale features by , calculate the grayscale mean within the ROI, is the number of pixels within the ROI; , determine the grayscale variance; by , determine the grayscale entropy, where p (k) is the probability that the grayscale value in the ROI is k, and obtain the grayscale feature vector ; The texture feature F1, shape feature F2, edge feature F3 and grayscale feature F4 are spliced ​​into a multidimensional feature matrix , the dimension is m0×1, m0 is the initial feature extraction dimension; The Relief-F algorithm is used to calculate the weight w(j) of each feature and update it iteratively: ;in, , is the nearest neighbor sample of the same type, is the nearest neighbor sample of different classes, n is the number of samples, p (c) is the prior probability of class c, and the weight w (j) ≥ w min The features of min is the weight threshold.

6. The method for detecting defects in cutting workpieces based on image feature feedback according to claim 1, characterized in that: The optimal feature subset is input into the detection model for inference, and the preliminary detection results of the defects are output, including the location coordinates, type and confidence of the defects. The defect confidence threshold is preset, and the preliminary detection results are screened to obtain effective defect detection results, which specifically include: Based on the optimal feature subset X' obtained by screening, a detection sample set is constructed, where the positive sample is a feature vector containing defects, and the negative sample is a feature vector without defects. The sample number ratio satisfies 1:1≤positive sample:negative sample≤3:1; Select the SVM model for training and construct the kernel function: K (x, y) = exp (-γ||xy||²); where γ is the kernel parameter. The optimal parameters C and γ are determined through cross-validation, where C is the penalty coefficient, with a value of 1≤C≤100, and γ is 0.01≤γ≤10. Train the SVM model to obtain the classification hyperplane: ;in, is the Lagrange multiplier, is the sample label, is the bias term, is the number of support vectors; Input the optimal feature subset X' into the trained SVM model and output the predicted category of the defect And the decision value d, confidence c=1 / (1+exp (-d)), get the preliminary detection result; Extract the confidence level c from the preliminary inspection results and compare it with the preset defect confidence level threshold θ0; When c≥θ0, it is determined to be a valid defect, and the location coordinates, type and confidence information of the defect are retained; When c<θ0, it is determined to be an invalid defect and is eliminated; The information of all valid defects is summarized to form a list of valid defect detection results, where each defect entry contains a unique identifier ID, location coordinates (x, y), type label t, size parameters (s1, s2) and confidence c.

7. A cutting workpiece defect detection system based on image feature feedback, used to implement the detection method according to any one of claims 1 to 6, characterized in that: include: An acquisition module, which is used to obtain the process characteristics of the cut workpiece and simultaneously build a preset defect type library, and is used to synchronously acquire original images of the cut workpiece from multiple perspectives based on initialized image acquisition parameters and the acquisition module and trigger device in the hardware system; A main control module, which is used to initialize image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model thresholds based on the performance parameters of the hardware system and defect type characteristics, input the optimal feature subset into the detection model for inference, output preliminary defect detection results, including the location coordinates, type and confidence of the defects, preset defect confidence thresholds, screen the preliminary detection results, obtain effective defect detection results, determine evaluation indicators, compare the evaluation indicators with the defect confidence thresholds, locate problem links in the detection process, and generate corresponding feedback signals. It is also used to adjust the image acquisition parameters, preprocessing parameters, feature extraction dimensions and detection model parameters based on the generated feedback signals, and repeatedly execute the process from image acquisition to defect detection and classification according to the adjusted parameters, output the final detection results and realize iterative optimization of the detection model and feature extraction method; a data processing module, wherein the data processing module is configured to store the original image in a database and associate the annotation information with the original image based on the acquisition time of the original image and the workpiece ID; to perform denoising on the original image based on the original image stored in the database using an adaptive filtering algorithm to obtain a denoised image; and to enhance the image contrast using a CLAHE algorithm based on the grayscale distribution characteristics of the denoised image to obtain an enhanced image; and then to extract the region of interest (ROI) of the workpiece using an edge segmentation algorithm; and based on the extracted region of interest (ROI), respectively extract texture features, shape features, edge features, and grayscale features; and to perform feature screening using a Relief-F algorithm based on the extracted multidimensional features to obtain an optimal feature subset; A display module is used to present the process and results of the cutting workpiece defect detection to the user.

8. The cutting workpiece defect detection system based on image feature feedback according to claim 7, characterized in that: The acquisition module includes: A first acquisition unit, the first acquisition unit is used to obtain the process characteristics of the cutting workpiece and simultaneously build a preset defect type library; The second acquisition unit is used to synchronously acquire original images of the cutting workpiece under multiple viewing angles based on the initialized image acquisition parameters and the acquisition module and trigger device in the hardware system.

9. The cutting workpiece defect detection system based on image feature feedback according to claim 7, characterized in that: The main control module includes: An initialization management unit, configured to initialize image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model thresholds based on hardware system performance parameters and defect type characteristics; A detection unit, which is used to input the optimal feature subset into the detection model for inference, output preliminary defect detection results, including the location coordinates, type and confidence of the defect, preset defect confidence threshold, screen the preliminary detection results, and obtain effective defect detection results; A feedback unit is used to determine an evaluation index, compare the evaluation index with a defect confidence threshold, locate the problem link in the detection process, and generate a corresponding feedback signal; An optimization unit is used to adjust image acquisition parameters, preprocessing parameters, feature extraction dimensions, and detection model parameters based on the generated feedback signal, and repeatedly execute the process from image acquisition to defect detection and classification according to the adjusted parameters, output the final detection results, and realize iterative optimization of the detection model and feature extraction method.

10. The cutting workpiece defect detection system based on image feature feedback according to claim 7, characterized in that: The data processing module includes: A data storage unit, the data storage unit being used to store the original image in a database and associate the annotation information with the original image according to the acquisition time of the original image and the workpiece ID; A preprocessing unit, configured to perform denoising on the original image using an adaptive filtering algorithm based on the original image stored in the database to obtain a denoised image; The feature processing unit is used to enhance the image contrast by using the CLAHE algorithm according to the grayscale distribution characteristics of the denoised image to obtain an enhanced image, and then use the edge segmentation algorithm to extract the region of interest (ROI) of the workpiece. Based on the extracted region of interest (ROI), texture features, shape features, edge features and grayscale features are extracted respectively. Based on the extracted multidimensional features, feature screening is performed by using the Relief-F algorithm to obtain the optimal feature subset.

Images