An image vision-based magnesium metal workpiece processing intelligent monitoring system

Abstract

The application provides an image vision-based magnesium metal workpiece processing intelligent monitoring system, the core of which is that a workpiece cutting image and a workpiece infrared image of a cutting surface are collected through an image collection module; a cutting heat flow feature of the workpiece infrared image is extracted through an image processing module to obtain a heat flow feature matrix corresponding to the workpiece infrared image; a dust adhesion connected domain is constructed through an image feature extraction module to extract a dust infrared temperature feature of the workpiece infrared image; the image feature extraction module determines a connected domain image feature of the workpiece cutting image according to the dust adhesion connected domain; an intelligent monitoring module determines a dust adhesion score of the magnesium metal workpiece according to the dust infrared temperature feature and the connected domain image feature, and performs intelligent monitoring and alarm according to the dust adhesion score, can recognize the dust adhesion connected domain according to the abnormal heat flow gradient of the magnesium metal workpiece image, realizes intelligent monitoring of the workpiece adhesion dust, and improves the processing precision of the magnesium metal workpiece.

Description

Technical Field

[0001] The present application relates to the technical field of image vision processing, and more specifically, to an intelligent monitoring system for magnesium metal workpiece processing based on image vision. Background Art

[0002] Magnesium metal processing (such as milling, turning, drilling, and grinding) generates large amounts of fine dust particles. This dust not only affects the processing environment but also poses safety and health risks, as well as various impacts on equipment and product quality. Dust adheres to the processed surface, reducing the adhesion of subsequent processes such as spraying, electroplating, and anodizing, leading to surface scratches, contamination, and dimensional deviations.

[0003] Existing technologies primarily monitor dust concentration in the machining environment to indirectly assess dust adhesion on workpiece surfaces. However, due to the low color contrast between dust and the metal surface, direct image monitoring of dust on the cutting surface is difficult and often relies on extensive manual sampling inspections, resulting in low monitoring efficiency and difficulty in achieving real-time, intelligent management. Therefore, intelligent monitoring of dust adhesion on the cutting surface of magnesium metal workpieces has become a pressing technical challenge. Summary of the Invention

[0004] The present application provides an intelligent monitoring system for magnesium metal workpiece processing based on image vision, which can identify dust adhesion connected domains according to abnormal heat flow gradients in the magnesium metal workpiece image, realize intelligent monitoring of dust adhesion on the workpiece, and improve the processing accuracy of the magnesium metal workpiece.

[0005] The present application provides an intelligent monitoring system for magnesium metal workpiece processing based on image vision, which includes:

[0006] An image acquisition module is used to acquire a workpiece cutting image and a workpiece infrared image of the cutting surface after the magnesium metal workpiece is cut;

[0007] an image processing module, configured to extract a boundary mask of a cutting surface based on the workpiece cutting image to obtain a cutting surface mask, and extract cutting heat flow features from the workpiece infrared image based on the cutting surface mask to obtain a heat flow feature matrix corresponding to the workpiece infrared image;

[0008] an image feature extraction module, configured to extract a set of heat flow gradient vectors from the heat flow feature matrix, construct a dust attachment connected domain based on each heat flow gradient vector in the heat flow gradient vector set, perform infrared feature extraction on the dust attachment connected domain, and obtain infrared temperature features of dust in the infrared image of the workpiece;

[0009] The image feature extraction module is further configured to determine the connected domain image features of the workpiece cutting image based on the dust adhesion connected domain;

[0010] An intelligent monitoring module is used to determine the dust adhesion score of the magnesium metal workpiece according to the dust infrared temperature characteristics and the connected domain image characteristics, and to perform intelligent monitoring and alarm according to the dust adhesion score.

[0011] In this embodiment, extracting a boundary mask of the cutting surface according to the workpiece cutting image to obtain a cutting surface mask body includes:

[0012] Extracting a boundary feature point set of the workpiece cutting image using a boundary feature extraction algorithm;

[0013] Selecting a connected area based on the boundary feature point set, and taking the largest connected area as the cutting surface area;

[0014] A binary mask image is constructed according to the region boundary of the cutting surface region as the cutting surface mask.

[0015] In this embodiment, the cutting heat flow feature extraction is performed on the workpiece infrared image according to the cutting surface mask to obtain the heat flow feature matrix corresponding to the workpiece infrared image, specifically including:

[0016] Mapping the cutting surface mask to an infrared image coordinate system according to a preset infrared mapping matrix;

[0017] Extracting information of each pseudo-color pixel point within the cutting surface mask in the infrared image of the workpiece according to the position coordinates of the cutting surface mask and performing heat flow temperature mapping to obtain heat flow temperature values ​​corresponding to each pseudo-color pixel point;

[0018] The heat flow characteristic matrix is ​​determined according to the heat flow temperature values ​​corresponding to each pseudo-color pixel point.

[0019] In this embodiment, extracting the heat flow gradient vector set through the heat flow feature matrix specifically includes:

[0020] Performing gradient calculation on the heat flow characteristic matrix to determine the horizontal gradient value and the vertical gradient value corresponding to each matrix element;

[0021] For any matrix element in the heat flow characteristic matrix, a heat flow gradient vector is constructed according to the horizontal gradient value and the vertical gradient value of the matrix element, and a heat flow gradient vector set is formed according to the heat flow gradient vectors corresponding to each matrix element.

[0022] In this embodiment, infrared feature extraction is performed on the dust attachment connected domain to obtain the dust infrared temperature feature of the workpiece infrared image, specifically including: traversing the pseudo-color pixel points of the workpiece infrared image in the dust attachment connected domain, determining the heat flow temperature value corresponding to each pseudo-color pixel point and extracting temperature statistical features, forming a feature vector according to the label order of the temperature statistical features, and using the feature vector as the dust infrared temperature feature.

[0023] In this embodiment, determining the connected domain image features of the workpiece cutting image based on the dust adhesion connected domain specifically includes:

[0024] Obtaining an infrared mapping matrix, and converting pixel coordinates of the dust attachment connected domain in the infrared image of the workpiece into a coordinate system of the workpiece cutting image based on the infrared mapping matrix to obtain a corresponding dust attachment connected domain mask in the workpiece cutting image;

[0025] Each image pixel point of the workpiece cutting image within the dust adhesion connected domain mask is obtained, and the statistical distribution characteristics of each image pixel point are extracted. A feature vector is composed according to the label order of the statistical distribution characteristics, and the feature vector is used as the connected domain image feature.

[0026] In this embodiment, in the process of determining the dust adhesion score of the magnesium metal workpiece based on the dust infrared temperature characteristics and the connected domain image characteristics, a pre-trained convolutional neural network is used to perform cluster analysis on the dust infrared temperature characteristics and the connected domain image characteristics, and the corresponding dust adhesion score is determined based on the clustering results.

[0027] In this embodiment, before extracting the boundary mask of the cutting surface based on the workpiece cutting image and obtaining the cutting surface mask, the method further includes: performing image preprocessing on the workpiece cutting image.

[0028] In this embodiment, performing image preprocessing on the workpiece cutting image specifically includes: performing grayscale processing on the workpiece cutting image, and performing image denoising by median filtering.

[0029] In this embodiment, the workpiece cutting image is collected by a visible light industrial camera.

[0030] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0031] The present application provides an intelligent monitoring system for magnesium metal workpiece processing based on image vision. After the magnesium metal workpiece is cut, an image acquisition module collects a workpiece cutting image and a workpiece infrared image of the cutting surface; an image processing module extracts a boundary mask of the cutting surface based on the workpiece cutting image to obtain a cutting surface mask, and performs cutting heat flow feature extraction on the workpiece infrared image based on the cutting surface mask to obtain a heat flow feature matrix corresponding to the workpiece infrared image; an image feature extraction module extracts a heat flow gradient vector set through the heat flow feature matrix, constructs a dust attachment connected domain based on each heat flow gradient vector in the heat flow gradient vector set, performs infrared feature extraction on the dust attachment connected domain, and obtains dust infrared temperature features of the workpiece infrared image; the image feature extraction module determines the connected domain image features of the workpiece cutting image based on the dust attachment connected domain; the intelligent monitoring module determines the dust attachment score of the magnesium metal workpiece based on the dust infrared temperature features and the connected domain image features, and performs intelligent monitoring and alarming based on the dust attachment score.

[0032] Therefore, it can be seen that the present application extracts the cutting surface boundary through the visible light image, constructs a mask and maps it to the infrared image coordinate system, which can accurately limit the subsequent heat flow analysis to the cutting area, avoid background interference, and improve the robustness and positioning accuracy of the analysis. The dust attachment connected domain is identified by judging the abnormal amplitude of the heat flow gradient vector and the condition of the vector orientation, and extracts the thermal anomaly aggregation area of ​​dust attachment. Compared with the traditional binary segmentation, it is more physically reasonable and improves the accuracy of dust identification. By fusing infrared and geometric image features, a dust attachment scoring model is established, which realizes the quantitative evaluation and graded processing of dust impact. It has intelligent and automated processing capabilities, does not require manual sampling, and can automatically determine whether there is dust attachment exceeding the limit after cutting is completed, realize real-time monitoring and alarm reminders, and greatly improve the processing accuracy of magnesium metal workpieces.

[0033] In summary, the present application can identify the dust attachment connected domain based on the abnormal heat flow gradient of the magnesium metal workpiece image, realize the intelligent monitoring of the dust attached to the workpiece, and improve the machining accuracy of the magnesium metal workpiece. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0035] Figure 1 This is a module structure diagram of an intelligent monitoring system for magnesium metal workpiece processing based on image vision provided by the present application;

[0036] Figure 2 is a schematic diagram of the process of extracting the cutting surface mask provided by this application;

[0037] Figure 3 It is a schematic diagram of the process of determining the heat flow characteristic matrix provided by this application. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0039] The present application provides an intelligent monitoring system for magnesium metal workpiece processing based on image vision. The core of the system is to collect a workpiece cutting image and a workpiece infrared image of the cutting surface after the magnesium metal workpiece is cut through the image acquisition module; an image processing module extracts a boundary mask of the cutting surface based on the workpiece cutting image to obtain a cutting surface mask, and performs cutting heat flow feature extraction on the workpiece infrared image based on the cutting surface mask to obtain a heat flow feature matrix corresponding to the workpiece infrared image; an image feature extraction module extracts a set of heat flow gradient vectors from the heat flow feature matrix, constructs a dust adhesion connected domain based on each heat flow gradient vector in the heat flow gradient vector set, and performs infrared feature extraction on the dust adhesion connected domain to obtain dust infrared temperature features of the workpiece infrared image; the image feature extraction module determines the connected domain image features of the workpiece cutting image based on the dust adhesion connected domain; the intelligent monitoring module determines the dust adhesion score of the magnesium metal workpiece based on the dust infrared temperature features and the connected domain image features, and performs intelligent monitoring and alarm based on the dust adhesion score. The system can identify the dust adhesion connected domain based on the abnormal heat flow gradient of the magnesium metal workpiece image, realize intelligent monitoring of workpiece dust adhesion, and improve the processing accuracy of the magnesium metal workpiece.

[0040] In order to better understand the above technical solution, the following will be described in detail with reference to the accompanying drawings and specific implementation methods. Figure 1 As shown in FIG, this figure is a module structure diagram of an intelligent monitoring system for magnesium metal workpiece processing based on image vision according to this embodiment of the present application. The intelligent monitoring system includes: an image acquisition module 100, an image processing module 200, an image feature extraction module 300 and an intelligent monitoring module 400, which are described as follows:

[0041] The image acquisition module 100 is used to acquire a workpiece cutting image and a workpiece infrared image of a cutting surface after the magnesium metal workpiece is cut.

[0042] It should be noted that the metal cutting image described in this application refers to a two-dimensional image captured by a visible light industrial camera, which reflects the true appearance of the workpiece cutting surface. The imaging principle is based on reflective imaging of natural light or artificial visible light. The infrared cutting image refers to an image captured by an infrared image acquisition device, which reflects the thermal distribution state of the workpiece cutting surface. The imaging principle is to detect the infrared radiation emitted from the surface of the object and convert it into a pseudo-color image. In this embodiment, the metal cutting image is collected by a visible light industrial camera, and the infrared cutting image is collected by an infrared camera. In some other embodiments, other devices or equipment that can realize the collection of visible light images and infrared images can also be used. This application does not limit this.

[0043] Preferably, in some embodiments of the present invention, the metal cutting images are captured using a visible light industrial camera, and the infrared cutting images are captured using an infrared thermal imaging camera. In other embodiments, other image capture devices capable of both visible light and infrared imaging, such as dual-spectral imaging systems or infrared-enhanced industrial cameras, may also be used for image capture, although this invention is not limited thereto.

[0044] During the specific implementation process, image acquisition equipment can be installed at the exit of the processing machine tool or at a pre-set inspection station after processing is completed. The visible light image acquisition device and the infrared image acquisition device must be arranged facing the same workpiece area and maintained at a relatively fixed installation angle to ensure that their fields of view are as consistent as possible, thereby reducing the impact of parallax on the accuracy of subsequent image registration. To achieve synchronous image acquisition, a programmable logic controller (PLC) or edge control module can simultaneously send synchronous trigger signals to the visible light camera and the infrared camera to ensure that both images are acquired at the same time, thereby improving data consistency and reliability and facilitating subsequent fusion analysis of heat flow and texture information.

[0045] It should be noted that to ensure image acquisition stability, it is preferred to wait until workpiece machining is complete and the machine tool spindle and worktable have completely stopped before capturing images to prevent interference from factors such as image blur, infrared image drift, or dust distribution disturbances. In certain embodiments of the present invention, an industrial robot or automatic transport device may be used to transfer the magnesium metal workpiece to be inspected to the image acquisition station for automatic positioning and image acquisition, thereby automating and standardizing the inspection process. The present invention does not limit the positioning method or transport mechanism.

[0046] The image processing module 200 is configured to extract a boundary mask of a cutting surface based on the workpiece cutting image to obtain a cutting surface mask, and extract cutting heat flow features from the workpiece infrared image based on the cutting surface mask to obtain a heat flow feature matrix corresponding to the workpiece infrared image.

[0047] In this embodiment, before extracting the boundary mask of the cutting surface based on the metal cutting image and obtaining the cutting surface mask, the method further includes: performing image preprocessing on the metal cutting image.

[0048] In a specific implementation, performing image preprocessing on the metal cutting image specifically includes: performing grayscale processing on the metal cutting image, and performing image denoising through median filtering.

[0049] In this embodiment, reference Figure 2 The figure is a schematic diagram of the process of extracting the cutting surface mask in some embodiments of the present application. The boundary mask of the cutting surface is extracted according to the metal cutting image to obtain the cutting surface mask body, which includes:

[0050] In step S21, a boundary feature extraction algorithm is used to extract a boundary feature point set of the metal cutting image;

[0051] In step S22, a connected area is selected based on the boundary feature point set, and the largest connected area is used as the cutting surface area;

[0052] In step S23, a binary mask image is constructed according to the region boundary of the cutting surface region as the cutting surface mask.

[0053] In specific implementation, the Canny edge detection algorithm can be used to extract boundary feature points of the preprocessed image to obtain the boundary feature point set of the metal cutting surface. This process can calibrate the appropriate detection threshold according to actual conditions or historical experience to obtain a clear and continuous edge contour. Based on the boundary feature point set, the image connectivity analysis algorithm, such as the connected domain labeling algorithm, is used to identify all connected areas in the image, and the largest connected area is selected as the target area representing the cutting surface through feature screening such as area and shape. According to the boundary of the above-mentioned largest connected area, the corresponding binary mask image is constructed as the cutting surface mask, wherein the pixel value in the mask area is set to 1, indicating the cutting surface range, and the pixel value outside the mask area is set to 0. The cutting surface mask is used for subsequent area limitation and feature extraction of infrared images.

[0054] In some optional embodiments of the present application, in order to improve the accuracy of boundary extraction, it is preferred to combine morphological operations after edge detection, and optional examples include dilation, erosion, and closing operations to fill and connect the edges.

[0055] In this embodiment, reference Figure 3 The figure is a schematic diagram of a process for determining a heat flow feature matrix in some embodiments of the present application. The heat flow feature extraction of the infrared cutting image is performed based on the cutting surface mask to obtain the heat flow feature matrix corresponding to the infrared cutting image. Specifically, the process includes:

[0056] In step S31, the cutting surface mask is mapped to an infrared image coordinate system according to a preset infrared mapping matrix;

[0057] In step S32, according to the position coordinates of the cutting surface mask, the information of each pseudo-color pixel in the cutting surface mask in the infrared cutting image is extracted and heat flow temperature mapping is performed to obtain the heat flow temperature value corresponding to each pseudo-color pixel;

[0058] In step S33, the heat flow characteristic matrix is ​​determined according to the heat flow temperature values ​​corresponding to the respective pseudo-color pixel points.

[0059] In specific implementation, during the determination of the infrared mapping matrix, a checkerboard calibration plate can be used to jointly calibrate and align the visible light camera and the infrared camera to obtain a spatial mapping relationship between the two, namely, the infrared mapping matrix; the infrared mapping matrix is ​​a perspective transformation matrix, which contains rotation, translation and scale transformation parameters, and can accurately convert the pixel coordinates in the visible light image to the corresponding coordinates in the infrared image; through the infrared mapping matrix, all valid pixel point coordinates of the cutting surface mask in the binary image in the visible light image coordinate system are converted into corresponding points in the infrared image coordinate system to obtain the cutting surface mask area in the infrared cutting image. In order to ensure the accuracy of the mapping, it is preferred to use interpolation processing on the mapped mask coordinate points, such as bilinear interpolation or bicubic interpolation, to adapt to the pixel grid of the infrared image to ensure the continuity and integrity of the temperature data.

[0060] In this embodiment, in the process of heat flow temperature mapping of each pseudo-color pixel information, the interval mapping of the pseudo-color pixel value and the temperature value can be performed according to the temperature-color mapping lookup table provided by the infrared device, so that the RGB value of the pseudo-color pixel point is converted into the corresponding heat flow temperature value, and all the corresponding temperature values ​​in the area are extracted according to the mapped cutting surface mask coordinates, and all the extracted temperature values ​​are organized into a two-dimensional matrix according to the spatial position of the pixels in the mask. This matrix is ​​the heat flow characteristic matrix of the cutting surface area.

[0061] It should be noted that in order to further improve the quality of the heat flow characteristic matrix, it is preferred to perform denoising and outlier removal during the temperature extraction process to reduce the impact of environmental interference and instrument errors on the temperature data.

[0062] The image feature extraction module 300 is used to extract a set of heat flow gradient vectors through the heat flow feature matrix, construct infrared features of the dust attachment connected domain based on each heat flow gradient vector in the heat flow gradient vector set, and obtain the infrared temperature features of the dust in the infrared image of the workpiece.

[0063] It should be noted that the dust attachment connected domain described in this application is the dust attachment area obtained by analyzing the change of heat flow gradient in the infrared image. During the cutting process of magnesium metal workpieces, the cutting surface will heat up rapidly due to high-speed cutting and friction, and the surface temperature is significantly higher than the normal temperature. However, when dust adheres to the cutting surface, due to the strong heat exchange ability between the dust particles and the air, its own temperature is usually significantly lower than the temperature of the cutting surface itself. In addition, the attached dust will interfere with the normal heat convection process between the workpiece surface and the surrounding air. On the one hand, the dust covering layer changes the thermal conductivity and thermal radiation characteristics of the original surface, and reduces the surface temperature of the local area. On the other hand, the dust Dust particles may form porous structures or suspended layers, causing abnormal local heat accumulation or loss, and forming obvious heat flow disturbance areas. In infrared thermal imaging images, this phenomenon is manifested as: the temperature value of the dust attachment area is relatively low; the heat flow gradient suddenly changes, and temperature abnormality faults or boundaries appear; the connectivity of heat flow in local areas is weakened. Based on the above physical phenomena, the present invention constructs a heat flow feature matrix and combines image processing algorithms to identify low-temperature areas and their corresponding heat flow gradient vectors. It can effectively judge the dust attachment area and realize intelligent detection and scoring of dust status. This mechanism is based on non-contact thermal perception, which not only improves detection accuracy but also avoids the limitations of traditional dust monitoring methods.

[0064] In this embodiment, extracting the heat flow gradient vector set through the heat flow feature matrix specifically includes:

[0065] Performing gradient calculation on the heat flow characteristic matrix to determine the horizontal gradient value and the vertical gradient value corresponding to each matrix element;

[0066] For any matrix element in the heat flow characteristic matrix, a heat flow gradient vector is constructed according to the horizontal gradient value and the vertical gradient value of the matrix element, and a heat flow gradient vector set is formed according to the heat flow gradient vectors corresponding to each matrix element.

[0067] In a specific implementation, the heat flow characteristic matrix is ​​subjected to gradient calculation to determine the horizontal gradient value and vertical gradient value of each matrix element. Preferably, the Sobel operator can be used to calculate the directional derivative of the heat flow characteristic matrix, thereby efficiently extracting the edges and transition areas of temperature changes in two-dimensional space. The Sobel operator is a commonly used edge detection algorithm with noise suppression and smoothing capabilities. It is suitable for analyzing temperature change gradients in infrared temperature maps, wherein the horizontal gradient value can be expressed as the rate of change of the derivative in the horizontal direction, and the vertical gradient value can be expressed as the rate of change of the derivative in the vertical direction. For any matrix element in the heat flow characteristic matrix, the corresponding heat flow gradient vector can be constructed through the horizontal gradient value and vertical gradient value corresponding to the matrix element. The heat flow gradient vector is a two-dimensional plane vector, and the two-dimensional elements respectively represent the gradient values ​​in the horizontal and vertical directions at the point.

[0068] Preferably, to improve computational efficiency and stability, the edges of the heat flow feature matrix can be mirrored and filled to avoid computational anomalies in the boundary region. Furthermore, for vector points with low gradient amplitudes, threshold suppression can be performed to reduce the interference of noise on the subsequent construction of the heat flow connected domain. Furthermore, in some specific implementations, the amplitude and direction of each heat flow gradient vector can be calculated for subsequent judgment of gradient strength and direction consistency. This application does not elaborate on this.

[0069] In this embodiment, constructing a heat flow interconnection domain according to each heat flow gradient vector in the heat flow gradient vector set specifically includes:

[0070] Obtain the gradient amplitude corresponding to each heat flow gradient vector;

[0071] Threshold screening is performed based on the gradient amplitudes corresponding to the respective heat flow gradient vectors, heat flow gradient vectors with amplitudes higher than a preset threshold are retained, and multiple connected regions are generated based on the pixel points of the infrared cutting image corresponding to the retained heat flow gradient vectors;

[0072] The boundary point sets of the connected areas are obtained respectively, and it is judged whether the boundary point sets of the connected areas meet the dust heat transfer conditions. The connected areas that meet the dust heat transfer conditions are retained as heat flow connected domains.

[0073] In specific implementation, each heat flow gradient vector can be subjected to a modulus operation, and the vector modulus can be used as the gradient amplitude corresponding to each heat flow gradient vector. Then, a set gradient amplitude threshold can be used, such as setting the mean plus the standard deviation of the heat flow amplitude map as the gradient amplitude threshold. Each heat flow gradient vector can be threshold-screened, and connected area analysis can be performed based on the pixel points of the infrared cutting image corresponding to the retained heat flow gradient vectors.

[0074] In connected region analysis, connectivity analysis is used to identify image regions, with the eight-neighborhood determination method being preferred to improve the accuracy of region coherence identification. Connected region analysis, based on a binary graph constructed from selected valid points, identifies multiple hotspots potentially affected by dust. For each connected region, a set of boundary pixels is extracted using edge detection or equivalent contour extraction algorithms. The geometric center of the region is then calculated based on the pixels within the region. To further determine heat flow directionality, the angle between the heat flow gradient vector corresponding to the boundary point and the unit direction vector pointing to the region's centroid is analyzed. The cosine of the angle between the two vectors is calculated. If the cosine value is above a preset threshold, it indicates that the gradient direction tends to converge inward, meaning that heat energy is concentrating from the exterior to the interior of the region. This abnormal thermal convection is consistent with localized heat conduction obstruction caused by the accumulation of cutting dust, a typical manifestation of dust adhesion thermal disturbance. The number of boundary points in each region that meet this condition is counted. If a ratio threshold is met, the connected region is considered a heat flow connected domain and used for subsequent infrared temperature feature extraction and dust adhesion scoring.

[0075] Determine whether the boundary point set of the connected area meets the dust heat transfer condition. The judgment principle is: if the directions of the heat flow gradient vectors on the boundary all point to the inside of the area, it means that the temperature gradient at the boundary is from the outside to the inside, that is, heat is transferred from the outside to the area, which satisfies the abnormal thermal convection characteristics caused by dust adhesion on the cutting surface. The following describes an optional specific implementation method: for each boundary point, obtain its gradient vector, and calculate the unit direction vector of the point pointing to the center of mass of the connected area, calculate the cosine of the angle between the gradient vector and the unit direction vector, and if the cosine of the angle is higher than the preset cosine threshold, it is determined that the connected area meets the dust heat transfer condition.

[0076] In this embodiment, infrared feature extraction is performed on the heat flow communication domain to obtain the infrared temperature feature of the dust in the infrared cutting image, specifically including: traversing the pseudo-color pixels of the infrared cutting image in the heat flow communication domain, determining the heat flow temperature value corresponding to each pseudo-color pixel and extracting temperature statistical features, forming a feature vector according to the label order of the temperature statistical features, and using the feature vector as the infrared temperature feature of the dust.

[0077] The temperature statistical features may preferably include but are not limited to the following: average temperature value: represents the overall level of temperature in the connected area. The dust adhesion area is usually lower than the average temperature of the normal cutting surface; maximum temperature value: reflects the degree of local heat concentration that may exist in the area, which helps to identify the mixed state area; minimum temperature value: used to determine whether there is a low-temperature cold spot, which is a typical infrared manifestation of dust coverage; temperature difference range: that is, the difference between the maximum temperature and the minimum temperature, reflecting the amplitude of temperature fluctuation within the area; temperature standard deviation: used to quantify the degree of discreteness of the temperature distribution. The temperature distribution in the dust adhesion area is usually more uneven and the standard deviation is relatively larger; temperature skewness: measures the asymmetry of the temperature distribution, which can reflect the imbalance of temperature distribution caused by dust; temperature kurtosis: reflects whether the temperature distribution is concentrated or flat. Dust adhesion often causes accumulation at the edge of the temperature distribution, resulting in a change in kurtosis; hot spot ratio: counts the proportion of pixels below a certain threshold of the average temperature, which assists in reflecting the density of dust cold spots; regional area: counts the number of pixels in the heat flow connected domain, used to evaluate the size of the dust impact range;

[0078] It should be noted that the aforementioned temperature statistical features are arranged in a preset label order to form a one-dimensional feature vector. This feature vector represents the infrared temperature signature of the dust in the heat flow connected domain and can subsequently be used as input for intelligent detection models to implement functions such as dust adhesion identification, scoring, and abnormality alarms.

[0079] In some embodiments, to improve feature stability, the temperature statistical feature vectors of multiple connected domains can be further normalized, or multi-frame image fusion can be performed based on a sliding window strategy to enhance the robustness and real-time performance of dust detection.

[0080] The image feature extraction module 300 is further configured to determine the connected domain image features of the workpiece cutting image based on the dust adhesion connected domain.

[0081] In this embodiment, determining the connected domain image features of the metal cutting image based on the heat flow connected domain specifically includes:

[0082] Obtaining an infrared mapping matrix, and converting pixel coordinates of the heat flow connected domain in the infrared cutting image to a coordinate system of the metal cutting image based on the infrared mapping matrix to obtain a corresponding heat flow connected domain mask in the metal cutting image;

[0083] Each image pixel point of the metal cutting image within the heat flow connected domain mask is obtained, and the statistical distribution characteristics of each image pixel point are extracted. A feature vector is formed according to the label order of the statistical distribution characteristics, and the feature vector is used as the connected domain image feature.

[0084] Among them, the statistical distribution characteristics include but are not limited to: regional area (total number of pixels), reflecting the size of the connected domain; regional perimeter, aspect ratio (length-to-width ratio) of the minimum circumscribed rectangle, convexity and concavity analysis of the boundary contour, such as the number of concave points and the convex hull area ratio, directional characteristics of the regional shape, such as the main direction angle, average grayscale value, grayscale standard deviation, grayscale median, grayscale minimum and maximum values, grayscale co-occurrence matrix texture characteristics, etc.

[0085] The intelligent monitoring module 400 is configured to determine the dust adhesion score of the magnesium metal workpiece according to the dust infrared temperature characteristics and the connected domain image characteristics, and perform intelligent monitoring and alarming according to the dust adhesion score.

[0086] In this embodiment, in the process of determining the dust adhesion score of the magnesium metal workpiece based on the dust infrared temperature characteristics and the connected domain image characteristics, a pre-trained convolutional neural network is used to perform cluster analysis on the dust infrared temperature characteristics and the connected domain image characteristics, and the corresponding dust adhesion score is determined based on the clustering results.

[0087] The following is a specific embodiment of the present application using a convolutional neural network to perform cluster analysis on the dust infrared temperature characteristics and the connected domain image characteristics:

[0088] A convolutional neural network is used to perform cluster analysis on the infrared temperature features of the dust and the connected domain image features, and the clustering results are used as the basis for evaluating the degree of dust adhesion. In specific implementation, the infrared temperature features of the dust and the connected domain image features are first fused to form a unified multi-dimensional joint feature vector. This joint feature vector contains both the spatial distribution information of temperature and image features such as the shape, size and texture of the dust connected area, which is used to comprehensively reflect the distribution and adhesion status of the dust.

[0089] The fused joint feature vector is input into a pre-designed convolutional neural network model, which contains multiple convolutional layers, pooling layers, and activation function layers. The specific structure is as follows: the input layer receives a multi-channel feature map, extracts the local spatial temperature-related feature patterns through several convolution kernels, and then reduces the dimension and enhances the translation invariance of the features through the pooling layer; the activation function layer uses activation functions such as ReLU to introduce nonlinear mapping capabilities; after the features undergo multi-layer convolution and pooling, the network integrates high-order semantic information through a fully connected layer and outputs the final clustering classification results.

[0090] The training sample data consists of a joint feature vector composed of historically collected dust infrared temperature features and connected domain image features, as well as the corresponding manual scoring results of dust adhesion; the manual scoring is obtained by professionals through on-site inspection or image annotation based on the actual degree of dust adhesion; during the training process, a supervised learning method is adopted, and the cross-entropy loss function or mean square error loss function is used to feedback the error between the network output and the manual scoring label, and the back-propagation algorithm is used to adjust the network weights and biases to optimize the model performance.

[0091] During the training iteration process, when the correlation indicator between the network clustering results and the manual scoring results, such as the Pearson correlation coefficient, is lower than the preset threshold, the following optimization measures are taken: adjust the learning rate, increase or decrease the number of convolution layers, modify the convolution kernel size, adjust the batch size, add regularization terms to prevent overfitting, or use data augmentation technology to improve the model generalization ability; continue iterative training until the correlation between the clustering results and the manual scoring reaches the expected standard.

[0092] After training, the convolutional neural network model is deployed in a real-time monitoring system, which takes the infrared temperature characteristics of dust collected on-site and the connected domain image characteristics as joint input to quickly perform feature extraction and cluster analysis. The clustering result output by the model is the assessment value of the degree of dust adhesion, which provides a scientific basis for subsequent dust control, cleaning scheduling and maintenance decisions, thereby improving the safety of the industrial environment and the efficiency of equipment operation.

[0093] Through this method, a deep learning model is used to fully explore the complex relationship between dust temperature and morphological characteristics, achieve accurate classification and prediction of dust adhesion status, effectively overcome the limitations of traditional judgment based on a single feature or manual experience, and significantly improve the automation and intelligence level of dust monitoring.

[0094] In this embodiment, performing intelligent monitoring and alarming based on the dust adhesion score specifically includes:

[0095] Based on historical data and safety regulations on dust adhesion scores, multiple alarm threshold intervals are preset, for example: Green interval (safe): score is lower than threshold T1; Yellow interval (caution): score is between threshold T1 and threshold T2; Red interval (danger): score is higher than threshold T2;

[0096] The system receives the dust adhesion score in real time and compares it with the preset threshold to make a judgment: if the score is in the green range, the system maintains normal monitoring status and no alarm is issued; if the score enters the yellow range, an early warning is triggered to remind relevant personnel to pay attention to the dust accumulation trend; if the score exceeds the red range, an alarm is triggered and corresponding protection or cleaning measures are initiated.

[0097] The system can send alarm signals in a variety of ways, including text messages, emails, platform notifications, sound and light alarms, etc.; it can also link automatic sprinkler and exhaust systems or start equipment cleaning procedures to achieve rapid response; all alarm events and corresponding dust scoring data are recorded in real time to facilitate subsequent analysis and system optimization; through the feedback mechanism, the scoring thresholds and alarm strategies are continuously adjusted to improve the intelligence level of the system.

[0098] In this embodiment, the environmental parameters (such as humidity, temperature, wind speed) and the equipment operating status can be combined to further verify the dust adhesion risk, avoid false alarms or missed alarms, and enhance the accuracy and reliability of monitoring; this application is based on the dust adhesion score output by the convolutional neural network, combined with the preset multi-level threshold and linkage alarm mechanism, to achieve intelligent and dynamic monitoring and early warning of the dust status, effectively ensuring the cleanliness and safety of the industrial environment and the normal operation of the equipment.

[0099] It can be seen that in this application, the cutting surface boundary is extracted by visible light image, a mask is constructed and mapped to the infrared image coordinate system, which can accurately limit the subsequent heat flow analysis to the cutting area, avoid background interference, and improve the robustness and positioning accuracy of the analysis. The dust attachment connected domain is identified by judging the abnormal amplitude and vector orientation of the heat flow gradient vector, and the thermal anomaly aggregation area of ​​dust attachment is extracted. Compared with the traditional binary segmentation, it is more physically reasonable and improves the accuracy of dust identification. By fusing infrared and geometric image features, a dust attachment scoring model is established, which realizes the quantitative evaluation and graded processing of dust impact. It has intelligent and automated processing capabilities, does not require manual sampling, and can automatically determine whether there is dust attachment exceeding the limit after cutting is completed, realize real-time monitoring and alarm reminders, and greatly improve the processing accuracy of magnesium metal workpieces.

[0100] In summary, the present application can identify the dust attachment connected domain based on the abnormal heat flow gradient of the magnesium metal workpiece image, realize the intelligent monitoring of the dust attached to the workpiece, and improve the machining accuracy of the magnesium metal workpiece.

[0101] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0102] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0103] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

Claims

1. An intelligent monitoring system for magnesium metal workpiece processing based on image vision, characterized in that: The intelligent monitoring system includes: An image acquisition module is used to acquire a workpiece cutting image and a workpiece infrared image of the cutting surface after the magnesium metal workpiece is cut; an image processing module, configured to extract a boundary mask of a cutting surface based on the workpiece cutting image to obtain a cutting surface mask, and extract cutting heat flow features from the workpiece infrared image based on the cutting surface mask to obtain a heat flow feature matrix corresponding to the workpiece infrared image; an image feature extraction module, configured to extract a set of heat flow gradient vectors from the heat flow feature matrix, construct a dust attachment connected domain based on each heat flow gradient vector in the heat flow gradient vector set, perform infrared feature extraction on the dust attachment connected domain, and obtain infrared temperature features of dust in the infrared image of the workpiece; The image feature extraction module is further configured to determine the connected domain image features of the workpiece cutting image based on the dust adhesion connected domain; an intelligent monitoring module, configured to determine a dust adhesion score of the magnesium metal workpiece based on the dust infrared temperature characteristics and the connected domain image characteristics, and to perform intelligent monitoring and alarming based on the dust adhesion score; Constructing a heat flow interconnection domain according to each heat flow gradient vector in the heat flow gradient vector set specifically includes: Obtain the gradient amplitude corresponding to each heat flow gradient vector; Threshold screening is performed based on the gradient amplitudes corresponding to the respective heat flow gradient vectors, heat flow gradient vectors with amplitudes higher than a preset threshold are retained, and multiple connected regions are generated based on the pixel points of the infrared cutting image corresponding to the retained heat flow gradient vectors; Obtain the boundary point sets of the connected areas respectively, determine whether the boundary point sets of the connected areas meet the dust heat transfer conditions, and retain the connected areas that meet the dust heat transfer conditions as the heat flow connected domain; Among them, for each boundary point, its gradient vector is obtained, and the unit direction vector pointing from the point to the center of mass of the connected area is calculated, and the cosine of the angle between the gradient vector and the unit direction vector is calculated. If the cosine of the angle is higher than the preset cosine threshold, it is judged that the connected area meets the dust heat transfer conditions.

2. The intelligent monitoring system for magnesium metal workpiece processing based on image vision according to claim 1, characterized in that: Extracting a boundary mask of a cutting surface according to the workpiece cutting image to obtain a cutting surface mask body includes: Extracting a boundary feature point set of the workpiece cutting image using a boundary feature extraction algorithm; Selecting a connected area based on the boundary feature point set, and taking the largest connected area as the cutting surface area; A binary mask image is constructed according to the region boundary of the cutting surface region as the cutting surface mask.

3. The intelligent monitoring system for magnesium metal workpiece processing based on image vision according to claim 1, characterized in that: Extracting cutting heat flow features from the infrared image of the workpiece according to the cutting surface mask to obtain a heat flow feature matrix corresponding to the infrared image of the workpiece specifically includes: Mapping the cutting surface mask to an infrared image coordinate system according to a preset infrared mapping matrix; Extracting information of each pseudo-color pixel point within the cutting surface mask in the infrared image of the workpiece according to the position coordinates of the cutting surface mask and performing heat flow temperature mapping to obtain heat flow temperature values ​​corresponding to each pseudo-color pixel point; The heat flow characteristic matrix is ​​determined according to the heat flow temperature values ​​corresponding to each pseudo-color pixel point.

4. The intelligent monitoring system for magnesium metal workpiece processing based on image vision according to claim 1, characterized in that: Extracting the heat flow gradient vector set through the heat flow characteristic matrix specifically includes: Performing gradient calculation on the heat flow characteristic matrix to determine the horizontal gradient value and the vertical gradient value corresponding to each matrix element; For any matrix element in the heat flow characteristic matrix, a heat flow gradient vector is constructed according to the horizontal gradient value and the vertical gradient value of the matrix element, and a heat flow gradient vector set is formed according to the heat flow gradient vectors corresponding to each matrix element.

5. The intelligent monitoring system for magnesium metal workpiece processing based on image vision according to claim 1, characterized in that: Performing infrared feature extraction on the dust attachment connected domain to obtain the dust infrared temperature feature of the workpiece infrared image specifically includes: traversing the pseudo-color pixel points of the workpiece infrared image in the dust attachment connected domain, determining the heat flow temperature value corresponding to each pseudo-color pixel point and extracting temperature statistical features, forming a feature vector according to the label order of the temperature statistical features, and using the feature vector as the dust infrared temperature feature.

6. The intelligent monitoring system for magnesium metal workpiece processing based on image vision according to claim 1, characterized in that: Determining the connected domain image features of the workpiece cutting image according to the dust adhesion connected domain specifically includes: Obtaining an infrared mapping matrix, and converting pixel coordinates of the dust attachment connected domain in the infrared image of the workpiece into a coordinate system of the workpiece cutting image based on the infrared mapping matrix to obtain a corresponding dust attachment connected domain mask in the workpiece cutting image; Each image pixel point of the workpiece cutting image within the dust adhesion connected domain mask is obtained, and the statistical distribution characteristics of each image pixel point are extracted. A feature vector is composed according to the label order of the statistical distribution characteristics, and the feature vector is used as the connected domain image feature.

7. The intelligent monitoring system for magnesium metal workpiece processing based on image vision according to claim 1, characterized in that: In the process of determining the dust adhesion score of the magnesium metal workpiece based on the dust infrared temperature characteristics and the connected domain image characteristics, a pre-trained convolutional neural network is used to perform cluster analysis on the dust infrared temperature characteristics and the connected domain image characteristics, and the corresponding dust adhesion score is determined based on the clustering results, wherein the training sample data of the convolutional neural network consists of a joint feature vector composed of historically collected dust infrared temperature characteristics and connected domain image features, as well as the corresponding dust adhesion manual score results.

8. The intelligent monitoring system for magnesium metal workpiece processing based on image vision according to claim 1, characterized in that: Extracting the boundary mask of the cutting surface according to the workpiece cutting image, and obtaining the cutting surface mask also include: performing image preprocessing on the workpiece cutting image.

9. The intelligent monitoring system for magnesium metal workpiece processing based on image vision according to claim 1, characterized in that: The image preprocessing for the workpiece cutting image specifically includes: grayscale processing for the workpiece cutting image, and image denoising by median filtering.

10. The intelligent monitoring system for magnesium metal workpiece processing based on image vision according to claim 1, characterized in that: The workpiece cutting image is collected by a visible light industrial camera.

Images