Watch case vacuum coating defect detection method and system based on image processing
Through image processing technology, including multi-angle image acquisition, full-scale contrast enhancement and optical flow motion estimation, the subjectivity and inefficiency of traditional watch case coating defect detection methods are solved, and the rapid, accurate detection and quantitative coating defects are achieved. Evaluate to support improvements in manufacturing processes.
Patent Information
- Application Number
- CN202411114023.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-08-14
AI Technical Summary
Traditional method of detecting defects in watch case coating relies on manual visual inspection, which has problems such as strong subjectivity, low efficiency and inaccurate defect monitoring.
Using the vacuum coating defect detection method of watch case based on image processing, by obtaining multi-angle real-time case coating image stream, full-scale contrast enhancement, optical flow motion estimation, timing frame segmentation, frame-by-frame surface feature analysis, fuzzy defect detection, Super-resolution imaging and fine defect secondary detection generate coating defect detection results.
It realizes rapid and accurate detection of case coating defects, improves the objectivity and efficiency of detection, can identify subtle texture defects, provide quantitative defect evaluation data, and supports improvement and optimization of the manufacturing process.
Smart Images

Figure CN118628500B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing, and in particular to a method and system for detecting vacuum coating defects of a watch case based on image processing. Background Art
[0002] The watch case is one of the core components of precision optoelectronic products, and its surface quality is directly related to the product's aesthetic appearance and performance. As consumers' requirements for watch quality continue to increase, how to quickly and accurately detect various defects produced during the vacuum coating process of the watch case has become an important challenge facing watch manufacturers. Traditional methods of case coating defect detection mainly rely on manual visual inspection, which has the problems of strong subjectivity, low efficiency and inaccurate defect monitoring. With the rapid development of computer vision technology, using image processing to identify surface defects of coatings has become an effective way to solve the problem. Summary of the invention
[0003] In order to solve the above technical problems, the present invention proposes a method and system for detecting defects of vacuum coating of watch cases based on image processing, so as to solve at least one of the above technical problems.
[0004] To achieve the above object, the present invention provides a method for detecting defects of vacuum coating of a watch case based on image processing, comprising the following steps:
[0005] Step S1: acquiring a multi-angle real-time watch case coating image stream; performing full-scale contrast enhancement on the multi-angle real-time watch case coating image stream to construct a full-scale convolution coating image stream;
[0006] Step S2: performing optical flow motion estimation on the full-scale convolution coating image stream and performing time-series frame segmentation processing to obtain a coating image frame sequence;
[0007] Step S3: Analyze the surface features of the coating image frame sequence frame by frame to construct multiple static texture feature maps; perform fuzzy defect detection on the multiple static texture feature maps to obtain texture defect data;
[0008] Step S4: extracting a defective unit image based on the texture defect data; performing step-by-step iterative decoding on the defective unit image, and performing super-resolution imaging to construct a super-resolution defective unit image;
[0009] Step S5: performing secondary fine defect detection on the super-resolution defect unit image to obtain fine texture defect data; performing defect morphology analysis on the fine texture defect data to generate defect morphology data and defect distribution density value;
[0010] Step S6: Perform unit defect comprehensive evaluation processing on the defect morphology data and defect distribution density value to generate coating defect detection results.
[0011] The present invention provides a comprehensive perspective and information source by acquiring a multi-angle real-time case coating image stream, comprehensively understands the coating condition of the case, performs full-scale contrast enhancement, enhances the contrast of the image, highlights details and features, constructs a full-scale convolution coating image stream, and improves the accuracy of analysis and detection of the coating condition. Through optical flow motion estimation and time-series frame segmentation processing, the motion information and time-series changes in the image are captured, the dynamic features in the coating image are identified, and the coating image frame sequence is obtained, which helps track the changes and development trends of the coating condition and provides basic data for subsequent analysis and detection. The frame-by-frame surface feature analysis reveals the subtle features and changes of the coating surface, accurately captures the detailed information of the coating, obtains texture defect data through fuzzy defect detection, effectively identifies defects and anomalies on the coating surface, and provides an important reference for subsequent defect location and repair. The defect unit is extracted The image highlights the local details of the defect, which facilitates a more in-depth analysis and processing of the defect. Through step-by-step iterative decoding and super-resolution imaging, the clarity and detail of the image are improved, and the detection and identification capabilities of defects are further strengthened. The secondary detection of fine defects further improves the accuracy and reliability of defect detection, ensuring comprehensive coverage and identification of subtle texture defects. The defect morphology analysis combines the defect distribution density value to provide comprehensive defect information, helping to understand the type, morphology and distribution of defects, and providing guidance and basis for subsequent processing and repair. Through comprehensive evaluation and processing of defect morphology data and defect distribution density values, the weights and impact levels of different types of defects are comprehensively considered to obtain comprehensive coating defect detection results, which help determine the overall quality status of the coating and provide quantitative defect evaluation data, providing important reference for improvement and optimization in the manufacturing process.
[0012] Preferably, step S1 comprises the following steps:
[0013] Step S11: Acquire multi-angle real-time watch case coating image stream;
[0014] Step S12: Calculate the global average brightness of the multi-angle real-time case coating image to obtain the global brightness value of the image;
[0015] Step S13: performing dark area recognition on the multi-angle real-time case coating image to extract the dark area of the image;
[0016] Step S14: optimizing the dark pixel aberration of the dark area of the image based on the global brightness value of the image, thereby constructing a brightness optimized coating image stream;
[0017] Step S15: Perform full-scale contrast enhancement on the brightness optimized coating image to construct a global convolution coating image stream.
[0018] The present invention obtains all-round and multi-angle information on the coating of the case by acquiring a multi-angle real-time coating image stream of the case, comprehensively understands the coating situation, improves the comprehensiveness and accuracy of the detection, and quantifies the overall brightness level of the image by calculating the global average brightness value of the image, thereby providing basic data for subsequent brightness adjustment and optimization, identifying the dark area to accurately locate the shadow part in the coating image, extracting the dark details in the image, and providing a basis for subsequent brightness adjustment and optimization. By optimizing the pixel aberration of the dark area, the dark details are made clearer and more prominent, and the brightness balance and visual quality of the image are improved. A brightness-optimized coating image stream is constructed to make the image more balanced and clear in terms of brightness, accurately analyze and detect the features and defects of the coating surface, enhance the contrast and clarity of the image through full-scale contrast enhancement, highlight the details and features of the image, and more accurately analyze and detect the situation of the coating surface. A global convolution coating image stream is constructed to integrate the optimized images at various angles, and provide more comprehensive and clearer coating surface information, laying a foundation for subsequent defect detection and analysis.
[0019] Preferably, step S15 specifically comprises the following steps:
[0020] Perform multi-scale spatial frequency domain decomposition on the brightness optimized coating image stream to obtain multiple component images with different spatial frequencies;
[0021] Convolving the brightness optimized coating image stream component by component according to a plurality of component images of different spatial frequencies to obtain convolution response data of each component image;
[0022] Performing nonlinear local mapping processing on the convolution response data of each component image to obtain convolution local mapping features;
[0023] Based on the convolution local mapping feature, convolution contrast optimization is performed on multiple component images of different spatial frequencies to generate multiple convolution contrast optimized component images;
[0024] Perform full-scale image reconstruction on multiple convolution contrast-optimized component images and construct a full-scale convolution coating image stream.
[0025] The present invention decomposes the image into components of different frequencies through spatial frequency domain decomposition, captures the details and features of the image at different spatial scales, and provides more information for subsequent processing. By performing convolution processing on different frequency component maps, the features at different scales are highlighted, and the detail information in the image is enhanced. The nonlinear local mapping process highlights the local details in the image, increases the contrast and clarity of the image, and makes the image easier to analyze and detect defects. Through convolution contrast optimization, the contrast of the image is further enhanced, making the image clearer and easier to analyze, and highlighting the features and defects of the coating surface. The full-scale image reconstruction integrates the optimized component maps of different frequencies to generate a more detailed and clear coating image, providing more comprehensive and accurate image data for defect detection.
[0026] Preferably, step S2 specifically comprises the following steps:
[0027] Step S21: performing optical flow motion estimation on the full-scale convolution coated image flow to obtain the image optical flow motion vector;
[0028] Step S22: performing motion change analysis between consecutive frames based on the image optical flow motion vector to generate image flow frame-to-frame motion change data;
[0029] Step S23: performing optical flow motion trajectory evolution on the inter-frame motion change data of the image stream to generate optical flow motion trajectory features;
[0030] Step S24: performing time-series frame segmentation processing on the full-scale convolution coating image stream based on the optical flow motion trajectory feature, thereby obtaining a coating image frame sequence.
[0031] The present invention captures pixel-level motion information in an image through optical flow motion estimation, helps analyze dynamic changes in the image, and plays an important role in detecting small or rapidly changing features involved in defects in coating of a watch case. Through continuous inter-frame motion change analysis, the motion pattern and change trend in the coating image are revealed, and small changes or defects on the coating surface are discovered. Image stream inter-frame motion change data is generated to provide a basis for subsequent motion trajectory evolution and feature extraction, and to achieve a more in-depth analysis of coating defects of the watch case. Optical flow motion trajectory evolution can convert pixel-level motion information in an image into trajectory features, help capture motion patterns and change laws in the coating image of the watch case, and generate optical flow motion trajectory features to provide a basis for subsequent time series frame segmentation processing, and more accurately distinguish different features and defects in the coating image. The time series frame segmentation processing divides the continuous image stream to achieve separation and extraction of coating images at different time points, and performs time series analysis on the changes and development of the coating surface. After obtaining a coating image frame sequence, the dynamic changes on the coating surface can be better observed, coating defects can be accurately detected and identified, and important information can be provided for quality control.
[0032] Preferably, the specific steps of step S24 are:
[0033] Performing motion pattern statistical analysis on the image optical flow motion vector to generate motion vector pattern features;
[0034] Based on the motion vector pattern features, the optical flow motion trajectory features are clustered and divided to generate multiple trajectory feature clusters;
[0035] Based on multiple trajectory feature clustering, the corresponding regions of the full-scale convolution coating image stream are identified, and the regional motion cycle period analysis is performed to obtain the main cycle motion period of the image;
[0036] Defining the candidate time window length according to the main cycle motion period of the image;
[0037] Based on the candidate time window length, the full-scale convolution coated image stream is subjected to frame-by-frame correlation calculation within the time window to obtain the inter-frame image correlation within the time window;
[0038] According to the inter-frame image correlation within the time window, the full-scale convolution coating image stream is subjected to time-series frame segmentation processing, thereby obtaining a coating image frame sequence;
[0039] The time sequence frame segmentation process is specifically as follows: when the inter-frame image correlation within the time window is lower than the preset inter-frame image correlation, it is determined as a segmentation point;
[0040] Perform image segmentation on the full-scale convolution coating image stream based on the segmentation points to obtain an image frame sequence;
[0041] The time-series frame segmentation operation is repeated until all image streams are traversed, and multiple image frame sequences are connected to obtain a coating image frame sequence.
[0042] The present invention reveals different motion patterns and trends in images by statistically analyzing the patterns of motion vectors, identifies specific dynamic features in coating images, and improves the accuracy and sensitivity of defect detection. It clusters the optical flow motion trajectory features to classify similar motion trajectories into one category, helps identify and analyze different motion patterns and features, and provides a basis for subsequent region identification and motion cycle analysis. It identifies the motion features and periodic changes in different regions, finds the main motion patterns and cycle periods in the image, and provides guidance for the subsequent time window length definition and inter-frame image correlation calculation. It better captures the motion features and change cycles in the image by defining the candidate time window length. It quantifies the similarity and correlation between image frames through frame-by-frame correlation calculation, helps identify the changes and differences between frames, and provides a basis for time series frame segmentation processing. It performs inter-frame segmentation at an appropriate time through inter-frame image correlation analysis, effectively segments the coating image stream into a frame sequence with continuity and correlation, and more finely observes and analyzes the changes on the coating surface.
[0043] Preferably, step S3 specifically comprises the following steps:
[0044] Step S31: Analyze the surface characteristics of the coating image frame sequence frame by frame to generate the surface characteristic data of the watch case;
[0045] Step S32: performing multi-frame texture feature fitting on the surface feature data of the watch case to construct multiple static texture feature maps;
[0046] Step S33: performing discontinuous mutation analysis on multiple static texture feature maps to obtain texture discontinuous mutation points;
[0047] Step S34: extracting image frames from the full-scale convolution coating image stream based on the texture discontinuity mutation point to obtain a frame image where the texture mutation occurs;
[0048] Step S35: Perform fuzzy defect detection on the frame image where the texture mutation occurs, thereby obtaining texture defect data.
[0049] The present invention captures minute changes and features between coating image frames through frame-by-frame surface feature analysis, generates detailed feature data of the case surface, understands surface texture, brightness, color and other information, and provides a basis for subsequent analysis and processing. By performing texture feature fitting on the surface feature data and integrating multi-frame information to construct a static texture feature map, the overall texture feature changes on the case surface are captured. By analyzing the discontinuous mutation points of the texture, the mutation position of the case surface texture is determined, and the existing defects or abnormal areas are revealed. By extracting frame images containing discontinuous mutation points of the texture, the focus is placed on the defective area, the complexity of processing data is reduced, and subsequent defect detection and analysis are performed more effectively. By performing fuzzy defect detection on the texture mutation frame image, potential texture defects, such as flaws and cracks, are identified and located, providing an important basis for the final defect analysis and processing.
[0050] Preferably, the specific steps of step S4 are:
[0051] Step S41: performing unit division processing on the frame image where the texture mutation occurs to obtain a plurality of unit images;
[0052] Step S42: locating defective units on a plurality of unit images based on the texture defect data, and extracting defective unit images;
[0053] Step S43: performing multi-layer feature compression coding on the defective unit image to obtain a low-dimensional feature coding of the image;
[0054] Step S44: performing step-by-step iterative decoding based on the low-dimensional feature coding of the image, and performing super-resolution imaging to construct a super-resolution defective unit image.
[0055] The present invention decomposes the defect image into multiple unit images through unit division processing, refines the defect area, improves the positioning accuracy of the defect, and provides a more refined data basis for subsequent defect analysis and processing. By positioning the defective units of multiple unit images, the texture defects are accurately located and extracted, which helps to achieve effective recognition and segmentation of defects and provides key support for subsequent feature extraction and processing. Through multi-layer feature compression coding, the information of the defective unit image is effectively compressed and encoded to reduce data complexity, extract key low-dimensional features, and provide efficient data representation for subsequent image processing and analysis. Through step-by-step iterative decoding and super-resolution imaging, high-quality defective unit images are reconstructed from low-dimensional feature coding, the clarity and detail display of the image are improved, defects are analyzed and identified more accurately, and support is provided for quality control and improvement.
[0056] Preferably, the specific steps of step S5 are:
[0057] Step S51: performing visual feature recognition on the super-resolution defective unit image to generate defective unit visual feature data;
[0058] Step S52: performing secondary fine defect detection on the defect unit visual feature data, thereby obtaining fine texture defect data;
[0059] Step S53: performing defect edge contour analysis on the subtle texture defect data and marking the defect edge contour line;
[0060] Step S54: performing defect morphology analysis on the defect edge contour line to generate defect morphology data;
[0061] Step S55: Calculate the defect distribution density of the super-resolution defective unit image to generate a defect distribution density value.
[0062] The present invention extracts key visual feature information, such as texture, shape, color, etc., through visual feature recognition, to provide an important data basis for the feature description and recognition of defects, accurately distinguish different types of defects, further refine and confirm defects through fine defect secondary detection, extract subtle texture defect data, increase sensitivity and accuracy to defects, and more comprehensively understand and process defects. Through defect edge contour analysis, the edge contour line of the defect is accurately located and marked, which helps to reveal the shape and size characteristics of the defect and provide a basis for further morphological analysis and processing. Through defect morphological analysis, the morphological characteristics of the defect, such as size, shape, distribution, etc., are deeply understood, which provides important information for the classification and evaluation of defects, identifies and processes defects of different forms, and evaluates the distribution density of defects in the image through defect distribution density calculation, which reveals the distribution law and density of defects.
[0063] Preferably, the specific steps of step S6 are:
[0064] Step S61: Calculate the defect position of the full-scale convolution coating image stream based on the subtle texture defect data to generate unit defect position coordinates;
[0065] Step S62: performing stereo distortion correction on the unit defect position coordinates to obtain the precise position of the watch case defect;
[0066] Step S63: Perform unit defect comprehensive evaluation processing on the defect morphology data and defect distribution density value according to the precise position of the defect of the watch case to generate a coating defect detection result.
[0067] The present invention applies the subtle texture defect data to the full-scale coating image stream through defect position calculation, realizes the precise positioning of the defect position, generates the accurate position coordinates of the unit defect, and provides an accurate data basis for subsequent processing and analysis. Through stereo distortion correction, the errors caused by the viewing angle and distortion in the image are eliminated, and the accurate correction of the unit defect position coordinates is realized to obtain the accurate position information of the watch case defect. Through the comprehensive evaluation and processing of unit defects, the defects are comprehensively evaluated by combining the precise position, morphological data and distribution density value of the defects, and the size, shape, position and density of the defects are comprehensively considered. Finally, the coating defect detection results are generated, which provides an important basis for quality control and improvement.
[0068] In this specification, a watch case vacuum coating defect detection system based on image processing is provided, which is used to execute the watch case vacuum coating defect detection method based on image processing as described above, comprising:
[0069] An image enhancement module is used to obtain a multi-angle real-time watch case coating image stream; perform full-scale contrast enhancement on the multi-angle real-time watch case coating image stream to construct a full-scale convolution coating image stream;
[0070] A time-series frame segmentation module is used to perform optical flow motion estimation on the full-scale convolution coating image stream and perform time-series frame segmentation processing to obtain a coating image frame sequence;
[0071] The fuzzy defect detection module is used to analyze the surface features of the coating image frame sequence frame by frame to construct multiple static texture feature maps; perform fuzzy defect detection on the multiple static texture feature maps to obtain texture defect data;
[0072] A super-resolution imaging module is used to extract defective unit images based on texture defect data; iteratively decode the defective unit images step by step, and perform super-resolution imaging to construct super-resolution defective unit images;
[0073] The fine defect detection module is used to perform fine defect secondary detection on the super-resolution defect unit image to obtain fine texture defect data; and is used to perform defect morphology analysis on the fine texture defect data to generate defect morphology data and defect distribution density values;
[0074] The defect assessment module is used to perform comprehensive evaluation of unit defects on defect morphology data and defect distribution density values to generate coating defect detection results.
[0075] The present invention collects the coating image stream of the watch case in real time from multiple angles to comprehensively capture various detailed information of the surface, laying a solid foundation for subsequent defect analysis. It adopts full-scale contrast enhancement technology to effectively improve the clarity and contrast of the coating image and highlight the tiny defect information on the surface. It uses optical flow motion estimation technology to perform time-series frame segmentation on the coating image stream to obtain a continuous coating image frame sequence. This time-series analysis method effectively eliminates the visual interference caused by movements such as case rotation and improves the reliability of subsequent defect detection. Through frame-by-frame surface feature analysis, multiple static texture feature maps are constructed to comprehensively depict the detailed information of the coating surface. Fuzzy defect detection is performed based on these texture feature maps, and some tiny defects that are difficult to detect with the naked eye are preliminarily found. The defect area obtained by fuzzy defect detection , extract the defect unit image and perform super-resolution reconstruction. This super-resolution imaging technology effectively amplifies and clarifies the defect information, providing higher quality input for fine defect detection. Through step-by-step iterative decoding, the detailed features of the defects are restored to the greatest extent. Based on the super-resolution defect unit image, secondary fine defect detection is performed to find some subtle texture defects that are difficult to detect with the naked eye. The detected subtle defects are morphologically analyzed to obtain rich quantitative data such as the size and shape of the defects. Combined with the defect distribution density, it provides a basis for subsequent comprehensive evaluation. Through comprehensive evaluation of defect morphological data and distribution density, the quality status of the coating surface is comprehensively judged. This quantitative evaluation result not only intuitively reflects the defect situation, but also provides data support for production process optimization. BRIEF DESCRIPTION OF THE DRAWINGS
[0076] Figure 1 A schematic flow chart of the steps of a method for detecting defects in vacuum coating of a watch case based on image processing according to the present invention;
[0077] Figure 2 Detailed implementation flow chart of step S1;
[0078] Figure 3 Detailed implementation flow chart of step S2;
[0079] Figure 4 is a schematic diagram of the detailed implementation steps of step S3, DETAILED DESCRIPTION
[0080] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0081] The present application example provides a method and system for detecting defects of vacuum coating on watch cases based on image processing. The execution subjects of the method and system for detecting defects of vacuum coating on watch cases based on image processing include but are not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as general computing nodes of the present application. The data processing platform includes but is not limited to: at least one of an audio image management system, an information management system, and a cloud data management system.
[0082] See also Figures 1 to 4 The present invention provides a method for detecting defects of vacuum coating on watch cases based on image processing, and the method for detecting defects of vacuum coating on watch cases based on image processing comprises the following steps:
[0083] Step S1: acquiring a multi-angle real-time watch case coating image stream; performing full-scale contrast enhancement on the multi-angle real-time watch case coating image stream to construct a full-scale convolution coating image stream;
[0084] Step S2: performing optical flow motion estimation on the full-scale convolution coating image stream and performing time-series frame segmentation processing to obtain a coating image frame sequence;
[0085] Step S3: Analyze the surface features of the coating image frame sequence frame by frame to construct multiple static texture feature maps; perform fuzzy defect detection on the multiple static texture feature maps to obtain texture defect data;
[0086] Step S4: extracting a defective unit image based on the texture defect data; performing step-by-step iterative decoding on the defective unit image, and performing super-resolution imaging to construct a super-resolution defective unit image;
[0087] Step S5: performing secondary fine defect detection on the super-resolution defect unit image to obtain fine texture defect data; performing defect morphology analysis on the fine texture defect data to generate defect morphology data and defect distribution density value;
[0088] Step S6: Perform unit defect comprehensive evaluation processing on the defect morphology data and defect distribution density value to generate coating defect detection results.
[0089] The present invention provides a comprehensive perspective and information source by acquiring a multi-angle real-time case coating image stream, comprehensively understands the coating condition of the case, performs full-scale contrast enhancement, enhances the contrast of the image, highlights details and features, constructs a full-scale convolution coating image stream, and improves the accuracy of analysis and detection of the coating condition. Through optical flow motion estimation and time-series frame segmentation processing, the motion information and time-series changes in the image are captured, the dynamic features in the coating image are identified, and the coating image frame sequence is obtained, which helps track the changes and development trends of the coating condition and provides basic data for subsequent analysis and detection. The frame-by-frame surface feature analysis reveals the subtle features and changes of the coating surface, accurately captures the detailed information of the coating, obtains texture defect data through fuzzy defect detection, effectively identifies defects and anomalies on the coating surface, and provides an important reference for subsequent defect location and repair. The defect unit is extracted The image highlights the local details of the defect, which facilitates a more in-depth analysis and processing of the defect. Through step-by-step iterative decoding and super-resolution imaging, the clarity and detail of the image are improved, and the detection and identification capabilities of defects are further strengthened. The secondary detection of fine defects further improves the accuracy and reliability of defect detection, ensuring comprehensive coverage and identification of subtle texture defects. The defect morphology analysis combines the defect distribution density value to provide comprehensive defect information, helping to understand the type, morphology and distribution of defects, and providing guidance and basis for subsequent processing and repair. Through comprehensive evaluation and processing of defect morphology data and defect distribution density values, the weights and impact levels of different types of defects are comprehensively considered to obtain comprehensive coating defect detection results, which help determine the overall quality status of the coating and provide quantitative defect evaluation data, providing important reference for improvement and optimization in the manufacturing process.
[0090] In the embodiment of the present invention, refer to Figure 1 , is a schematic flow chart of the steps of a method for detecting vacuum coating defects of a watch case based on image processing of the present invention. In this example, the steps of the method for detecting vacuum coating defects of a watch case based on image processing include:
[0091] Step S1: acquiring a multi-angle real-time watch case coating image stream; performing full-scale contrast enhancement on the multi-angle real-time watch case coating image stream to construct a full-scale convolution coating image stream;
[0092] In this embodiment, multiple high-resolution, high-frame-rate industrial cameras are used to perform real-time imaging and acquisition of different angles of the watch case. The camera is installed on a precise rotating platform to scan the surface of the case in all directions. A special optical lens system is used to achieve high-definition capture of micron-level details. The collected raw image data is preprocessed by color correction, noise filtering, etc. to improve image quality. A full-scale contrast stretching algorithm, such as histogram equalization, is used to enhance the visibility of subtle features in the image. Through multi-scale Laplacian pyramid decomposition, the image detail information is independently enhanced at different spatial frequencies. The preprocessed and contrast-enhanced multi-angle images are fused and reconstructed through a convolutional neural network.
[0093] The network structure adopts a fully convolutional architecture such as U-Net, which can retain the detailed information in the image and output a full-scale coating image stream with both macroscopic structure and microscopic detail information.
[0094] Step S2: performing optical flow motion estimation on the full-scale convolution coating image stream and performing time-series frame segmentation processing to obtain a coating image frame sequence;
[0095] In this embodiment, a dense optical flow algorithm, such as the Farneback algorithm or the Horn-Schunck algorithm, is used to estimate the motion vector field of the full-scale convolution coating image stream frame by frame. By analyzing the optical flow vector changes between adjacent frames, the subtle motion information of the coating surface is obtained. The optical flow algorithm makes full use of multi-scale information to improve the accuracy and robustness of motion estimation. According to the time variation characteristics of the optical flow vector field, the image frames are segmented using a clustering algorithm, and continuous frames with similar motion characteristics are classified into the same segment to form a time sequence frame sequence in the coating process. By analyzing the characteristics of each time segment, such as motion speed, acceleration, etc., the key moments of the coating process are identified. From the full-scale convolution coating image stream, the image frame sequences of each coating stage obtained by the above-mentioned time sequence segmentation are extracted. Each frame sequence represents a continuous change process of the surface state during the coating process. These high-quality coating image frame sequences provide key data input for subsequent defect detection analysis.
[0096] Step S3: Analyze the surface features of the coating image frame sequence frame by frame to construct multiple static texture feature maps; perform fuzzy defect detection on the multiple static texture feature maps to obtain texture defect data;
[0097] In this embodiment, for each frame of the coating image, Gabor filtering, LBP, GLCM and other methods are used to extract multi-scale and multi-directional texture features. These texture features can describe the microstructure and detail information of the coating surface. The calculation results of different features are combined into a multi-channel feature map to fully retain the rich information of the surface texture. The multi-channel texture feature map of each frame of the image is stacked and aggregated to construct a series of static texture feature maps. These static feature maps reflect the overall texture characteristics of the coating surface that change over time. By comparing and analyzing these feature maps, abnormal areas of the surface texture are found. A semantic segmentation model based on a convolutional neural network is used to perform pixel-by-pixel defect detection on the above static texture feature maps. This deep learning-based method can accurately identify unclear defect areas with irregular boundaries on the coating surface. The trained model is used to automatically generate a defect segmentation mask for each frame of the image to obtain complete defect data.
[0098] Step S4: extracting a defective unit image based on the texture defect data; performing step-by-step iterative decoding on the defective unit image, and performing super-resolution imaging to construct a super-resolution defective unit image;
[0099] In this embodiment, according to the obtained texture defect data, the corresponding defect area coordinates are located in the original coating image frame sequence, and a sub-image of a certain size is cut out with these defect area coordinates as the center as the defect unit image. The size of the defect unit image needs to be adjusted according to the actual defect size and image resolution to ensure that it contains complete defect information. For the above-extracted defect unit image, a super-resolution reconstruction method based on deep learning is used to enlarge and supplement the details. The low-resolution defect unit image is iteratively reconstructed using pre-trained super-resolution models, such as SRCNN, ESPCN, EDSR, etc. These super-resolution models can reasonably infer and supplement the detail information according to the internal features of the image to generate high-quality super-resolution defect unit images. The reconstructed high-resolution defect unit image is fused with the original low-resolution defect unit image. The weighted average, Laplacian pyramid and other methods can be used for fusion to make full use of the advantages of both information, and finally a high-quality super-resolution defect unit image is obtained, which provides detailed visual input for subsequent intelligent analysis and recognition.
[0100] Step S5: performing secondary fine defect detection on the super-resolution defect unit image to obtain fine texture defect data; performing defect morphology analysis on the fine texture defect data to generate defect morphology data and defect distribution density value;
[0101] In this embodiment, the super-resolution defect unit image is finely analyzed at the pixel level. This fine detection method based on deep learning can accurately identify subtle texture defects on the defect unit image. Using the fine defect detection model, each super-resolution defect unit image is analyzed frame by frame to generate a fine defect segmentation mask. These fine defect segmentation results are summarized to form a series of subtle texture defect data. These subtle texture defect data contain detailed information such as the precise position, shape, size, etc. of the defects, laying the foundation for subsequent in-depth analysis. For the subtle texture defect data, image processing and computer vision methods are used to extract the morphological features of each defect. These morphological features include the area, perimeter, shape factor, aspect ratio, etc. of the defects, which can fully describe the geometric characteristics of the defects. At the same time, according to the distribution of defects on the entire coating surface, the spatial distribution density value of the defects, the defect morphological data and the distribution density value are calculated.
[0102] Step S6: Perform unit defect comprehensive evaluation processing on the defect morphology data and defect distribution density value to generate coating defect detection results.
[0103] In this embodiment, the morphological feature data of each defect unit extracted, including area, perimeter, shape factor, aspect ratio, etc., are used, and the spatial distribution density value of the defect on the entire coating surface is combined to comprehensively consider and analyze the defect morphological data and distribution density values. The severity of each defect unit is judged according to a preset threshold rule. According to the above comprehensive evaluation results, each defect unit is divided into different defect categories, such as slight defects, moderate defects, severe defects, etc., to generate a complete coating defect detection report, including the position coordinates, size, morphological features and defect category of each defect unit. The position coordinates, defect type, defect level and other information of each defect unit are integrated into a complete coating defect detection report, and the results are presented in the form of text, charts or interactive interfaces for easy viewing and analysis by users. For severe defect areas, they are marked on the original coating image to intuitively display the defect location and distribution.
[0104] In this embodiment, refer to Figure 2 , is a flowchart of detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include:
[0105] Step S11: Acquire multi-angle real-time watch case coating image stream;
[0106] Step S12: Calculate the global average brightness of the multi-angle real-time case coating image to obtain the global brightness value of the image;
[0107] Step S13: performing dark area recognition on the multi-angle real-time case coating image to extract the dark area of the image;
[0108] Step S14: optimizing the dark pixel aberration of the dark area of the image based on the global brightness value of the image, thereby constructing a brightness optimized coating image stream;
[0109] Step S15: Perform full-scale contrast enhancement on the brightness optimized coating image to construct a global convolution coating image stream.
[0110] In this embodiment, multi-angle image data of the case during the coating process is collected in real time to ensure that the position and angle of each camera can fully cover the case surface, obtain complete coating image information, and synchronously integrate the image streams collected by multiple cameras to form a continuous multi-angle coating image sequence. For each frame of the coating image, the global average brightness value is calculated, and a value representing the overall brightness level of the frame image is obtained by using methods such as histogram statistics or pixel average value. The global brightness values of all frame images are summarized to form a global brightness change curve of a time series, and each frame image is segmented using an image segmentation algorithm, such as the Otsu threshold method or K-means clustering. Automatic identification and extraction of dark areas, identify relatively dark areas in the image, mark them as dark areas, integrate the dark area information in all frame images, construct a time series of dark area distribution change process, dynamically adjust the pixel brightness value of the dark area in each frame image according to the global brightness change curve, for globally dark image frames, appropriately increase the pixel value of the dark area to enhance the overall brightness; for globally bright images, reduce the brightness of the dark area. Through this dark area optimization based on global brightness, a coating image sequence with more uniform overall brightness is constructed. A global contrast enhancement algorithm, such as the CLAHE (Contrast Limited Adaptive Histogram Equalization) method, is used to further process the brightness optimized image. By dynamically adjusting the overall histogram of the image, global contrast enhancement is achieved, detail information is highlighted, and the overall clarity of the image is improved. The enhanced image frame sequence is combined into a complete global convolution coating image stream, providing high-quality input data for subsequent defect detection.
[0111] In this embodiment, the specific steps of step S15 are:
[0112] Perform multi-scale spatial frequency domain decomposition on the brightness optimized coating image stream to obtain multiple component images with different spatial frequencies;
[0113] Convolving the brightness optimized coating image stream component by component according to a plurality of component maps of different spatial frequencies to obtain convolution response data of each component map;
[0114] Performing nonlinear local mapping processing on the convolution response data of each component image to obtain convolution local mapping features;
[0115] Based on the convolution local mapping feature, convolution contrast optimization is performed on multiple component images of different spatial frequencies to generate multiple convolution contrast optimized component images;
[0116] Perform full-scale image reconstruction on multiple convolution contrast-optimized component images and construct a full-scale convolution coating image stream.
[0117] In this embodiment, a method such as multi-scale Fourier transform or wavelet transform is used to perform frequency domain decomposition on the brightness optimized coating image stream, and the image is decomposed into different spatial frequency components, such as low-frequency, medium-frequency and high-frequency components. Each component represents the characteristics of the image in the spatial frequency domain at different scales. For each spatial frequency component map, a corresponding convolution kernel is designed, and the convolution kernel is convolved with the component map to obtain the convolution response value of the component at each position, so as to obtain the convolution response data of different spatial frequency components. For the convolution response data of each component, a nonlinear mapping function such as a sigmoid function or a tanh function is used for local processing. This nonlinear Mapping can highlight local texture and detail features and enhance the contrast of the image. The result is the convolution local mapping feature of each component. According to the convolution local mapping feature, the corresponding convolution filter is designed. These convolution filters are applied to the component images of different spatial frequencies respectively to optimize and enhance the local contrast. A series of new component images after convolution contrast optimization are obtained. The optimized component images are reorganized in the frequency domain by inverse transformation. The optimized component images are synthesized into a complete full-scale image by inverse Fourier transform or inverse wavelet transform. The reconstructed full-scale image sequence is composed of the final full-scale convolution coated image stream.
[0118] In this embodiment, refer to Figure 3 , is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include:
[0119] Step S21: performing optical flow motion estimation on the full-scale convolution coated image flow to obtain the image optical flow motion vector;
[0120] Step S22: performing motion change analysis between consecutive frames based on the image optical flow motion vector to generate image flow frame-to-frame motion change data;
[0121] Step S23: performing optical flow motion trajectory evolution on the inter-frame motion change data of the image stream to generate optical flow motion trajectory features;
[0122] Step S24: performing time-series frame segmentation processing on the full-scale convolution coating image stream based on the optical flow motion trajectory feature, thereby obtaining a coating image frame sequence,
[0123] In this embodiment, an optical flow calculation method such as the Lucas-Kanade algorithm or the Horn-Schunck algorithm is used to estimate the optical flow motion vector of the full-scale convolution coating image stream frame by frame. The optical flow algorithm estimates the motion vector of each pixel by analyzing the brightness or gradient changes between adjacent frames. The obtained optical flow motion vector contains the motion information of the object in the image, reflecting the flow of the material during the coating process. The optical flow motion vectors of adjacent frames are compared, the motion change of each pixel is calculated, and the motion change of each pixel is counted to generate an inter-frame motion change histogram or heat map. This inter-frame motion change data describes the dynamic characteristics of the material flow during the coating process. The optical flow motion vectors of adjacent frames are connected in series to form a motion trajectory of each pixel. The shape, length, direction and other features of these motion trajectories are analyzed to extract meaningful motion trajectory features. The entire image sequence is divided into several meaningful time periods or stages. Each time period corresponds to a coating image frame sequence, reflecting different operating states of the coating process.
[0124] In this embodiment, the specific steps of step S24 are:
[0125] Performing motion pattern statistical analysis on the image optical flow motion vector to generate motion vector pattern features;
[0126] Based on the motion vector pattern features, the optical flow motion trajectory features are clustered and divided to generate multiple trajectory feature clusters;
[0127] Based on multiple trajectory feature clustering, the corresponding regions of the full-scale convolution coating image stream are identified, and the regional motion cycle period analysis is performed to obtain the main cycle motion period of the image;
[0128] Defining the candidate time window length according to the main cycle motion period of the image;
[0129] Based on the candidate time window length, the full-scale convolution coated image stream is subjected to frame-by-frame correlation calculation within the time window to obtain the inter-frame image correlation within the time window;
[0130] According to the inter-frame image correlation within the time window, the full-scale convolution coating image stream is subjected to time-series frame segmentation processing, thereby obtaining a coating image frame sequence;
[0131] The time sequence frame segmentation process is specifically as follows: when the inter-frame image correlation within the time window is lower than the preset inter-frame image correlation, it is determined as a segmentation point;
[0132] Perform image segmentation on the full-scale convolution coating image stream based on the segmentation points to obtain an image frame sequence;
[0133] The time-series frame segmentation operation is repeated until all image streams are traversed, and multiple image frame sequences are connected to obtain a coating image frame sequence.
[0134] In this embodiment, the direction, size and other characteristics of the optical flow motion vector are statistically calculated to establish a motion vector pattern distribution, and statistical characteristics of the motion vector pattern distribution, such as mean, variance, skewness, etc., are extracted as motion vector pattern characteristics. A clustering algorithm (such as K-means, DBSCAN, etc.) is used to cluster the optical flow motion trajectory characteristics. According to the clustering results, a plurality of trajectory feature clusters with similar motion patterns are divided, and the optical flow motion trajectory clustering results are mapped back to the original image to identify different motion regions, analyze the motion trajectory of the pixels in each region, extract its periodic change characteristics, and obtain the regional motion cycle period. According to the motion cycle period of each region, the main cycle motion period of the entire image flow is determined. Period, the main cycle motion period is divided into equal parts, and several candidate time window lengths are defined. In each candidate time window, the image correlation between adjacent frames is calculated, so that the change law of the coating image sequence in the time window is analyzed. When the image correlation between frames in the time window is lower than the preset threshold, the point is determined as the segmentation point, and the entire image stream is segmented at the segmentation point to obtain independent coating image frame sequences. When the image correlation between frames in the time window is higher than the preset threshold, it is determined to be the same sequence image with a high correlation, and no segmentation is performed. The above operation is repeated until the entire image stream is traversed, and all the segmented frame sequences are connected to finally obtain a complete coating image frame sequence.
[0135] In this embodiment, refer to Figure 4 , is a flowchart of detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include:
[0136] Step S31: Analyze the surface characteristics of the coating image frame sequence frame by frame to generate the surface characteristic data of the watch case;
[0137] Step S32: performing multi-frame texture feature fitting on the surface feature data of the watch case to construct multiple static texture feature maps;
[0138] Step S33: performing discontinuous mutation analysis on multiple static texture feature maps to obtain texture discontinuous mutation points;
[0139] Step S34: extracting image frames from the full-scale convolution coating image stream based on the texture discontinuity mutation point to obtain a frame image where the texture mutation occurs;
[0140] Step S35: Perform fuzzy defect detection on the frame image where the texture mutation occurs, thereby obtaining texture defect data.
[0141] In this embodiment, for each frame of the coating image, visual features such as surface texture, color, and brightness are extracted, and texture features are extracted using methods such as Gabor filtering and gray-level co-occurrence matrix. The feature data of each frame of the image are combined into a feature vector. The surface feature data of several consecutive frames (such as 50 frames) are fitted by PCA or other dimensionality reduction methods to obtain a set of principal components. Each principal component constructs a two-dimensional static texture feature map. These static texture feature maps reflect the average distribution characteristics of the surface texture during the coating process. The texture distribution in the static texture feature map is analyzed to identify the area where the texture changes discontinuously. The discontinuous boundaries in the image are detected using methods such as gradient or Sobel operator. These discontinuous mutation points correspond to the defective areas on the coating surface. The identified texture discontinuous mutation points correspond to the original coating image sequence, and the specific time frames where these texture mutations occur are extracted to obtain a series of "frame images where texture mutations are located". The texture mutation frames are analyzed using image processing algorithms (such as Canny edge detection, Sobel operator, etc.), and the blurred areas in the image, that is, surface texture defects, are identified. The feature data of these defective areas are extracted as texture defect data.
[0142] In this embodiment, step S4 includes the following steps:
[0143] Step S41: performing unit division processing on the frame image where the texture mutation occurs to obtain a plurality of unit images;
[0144] Step S42: locating defective units on a plurality of unit images based on the texture defect data, and extracting defective unit images;
[0145] Step S43: performing multi-layer feature compression coding on the defective unit image to obtain a low-dimensional feature coding of the image;
[0146] Step S44: performing step-by-step iterative decoding based on the low-dimensional feature coding of the image, and performing super-resolution imaging to construct a super-resolution defective unit image.
[0147] In this embodiment, the texture mutation frame image is divided into multiple small unit areas (such as squares, rectangles, etc.), and each small unit corresponds to a unit image. Through unit division, the position of the defect is more precisely located and analyzed. The previously obtained texture defect data is used to identify which unit images contain defects, and these unit images containing defects are extracted as defective unit images. A deep learning feature extraction network (such as a convolutional neural network) is used to perform multi-level feature extraction on the defective unit images. Through feature compression, the high-dimensional defective unit images are encoded into low-dimensional feature vectors. These low-dimensional feature codes more efficiently represent defect information. Super-resolution models such as generative adversarial networks (GANs) are used to gradually decode the low-dimensional feature codes and restore them to high-resolution defective unit images. Through iterative optimization, clear and detailed super-resolution defective unit images are obtained. These super-resolution defective images more finely display and analyze the detailed features of the defects.
[0148] In this embodiment, the specific steps of step S5 are:
[0149] Step S51: performing visual feature recognition on the super-resolution defective unit image to generate defective unit visual feature data;
[0150] Step S52: performing secondary fine defect detection on the defect unit visual feature data, thereby obtaining fine texture defect data;
[0151] Step S53: performing defect edge contour analysis on the subtle texture defect data and marking the defect edge contour line;
[0152] Step S54: performing defect morphology analysis on the defect edge contour line to generate defect morphology data;
[0153] Step S55: Calculate the defect distribution density of the super-resolution defective unit image to generate a defect distribution density value.
[0154] In this embodiment, a deep learning image recognition model is used to extract visual features of super-resolution defective unit images, identify visual features such as shape, texture, and color of defects, and organize these feature information into structured defective unit visual feature data. Based on the defective unit visual feature data, a more sophisticated image analysis algorithm is further used to discover some subtle texture defects that are difficult to identify with the naked eye, obtain data information of these subtle texture defects, extract edge contours of subtle texture defects using an image edge detection algorithm, mark these defect edge contours to form defect edge contour data, analyze the shape, size, length and other features of the defect edge contours, organize these defect morphological features into structured defect morphological data, count the number and area of defects in each defective unit image, calculate the defect distribution density in the entire image area, reflect the concentration of defects, and use this defect distribution density value as an important indicator.
[0155] In this embodiment, the specific steps of step S6 are:
[0156] Step S61: Calculate the defect position of the full-scale convolution coating image stream based on the subtle texture defect data to generate unit defect position coordinates;
[0157] Step S62: performing stereo distortion correction on the unit defect position coordinates to obtain the precise position of the watch case defect;
[0158] Step S63: Perform unit defect comprehensive evaluation processing on the defect morphology data and defect distribution density value according to the precise position of the defect of the watch case to generate a coating defect detection result.
[0159] In this embodiment, computer vision and image processing technology are used to accurately determine the coordinate position of each subtle defect in the original coating image, calculate the precise coordinate position information of each defect unit, and organize these position coordinate information into a structured data set. Since the coating image has certain perspective and lens distortion, the position coordinates need to be corrected. The distortion information of the calibration image is used to perform stereo distortion correction on the defect position coordinates to obtain the precise position coordinates of the corrected watch case defect. According to the morphological characteristics, distribution density and precise position of the defect, each defect is comprehensively evaluated and classified to generate a detailed coating defect detection result report, including comprehensive information such as defect type, size, and position, to provide a basis for subsequent quality control.
[0160] In this embodiment, a watch case vacuum coating defect detection system based on image processing is provided, which is used to execute the watch case vacuum coating defect detection method based on image processing as described above, including:
[0161] An image enhancement module is used to obtain a multi-angle real-time watch case coating image stream; perform full-scale contrast enhancement on the multi-angle real-time watch case coating image stream to construct a full-scale convolution coating image stream;
[0162] A time-series frame segmentation module is used to perform optical flow motion estimation on the full-scale convolution coating image stream and perform time-series frame segmentation processing to obtain a coating image frame sequence;
[0163] The fuzzy defect detection module is used to analyze the surface features of the coating image frame sequence frame by frame to construct multiple static texture feature maps; perform fuzzy defect detection on the multiple static texture feature maps to obtain texture defect data;
[0164] A super-resolution imaging module is used to extract defective unit images based on texture defect data; iteratively decode the defective unit images step by step, and perform super-resolution imaging to construct super-resolution defective unit images;
[0165] The fine defect detection module is used to perform fine defect secondary detection on the super-resolution defect unit image to obtain fine texture defect data; and is used to perform defect morphology analysis on the fine texture defect data to generate defect morphology data and defect distribution density values;
[0166] The defect assessment module is used to perform comprehensive evaluation of unit defects on defect morphology data and defect distribution density values to generate coating defect detection results.
[0167] The present invention collects the coating image stream of the watch case in real time from multiple angles to comprehensively capture various detailed information of the surface, laying a solid foundation for subsequent defect analysis. It adopts full-scale contrast enhancement technology to effectively improve the clarity and contrast of the coating image and highlight the tiny defect information on the surface. It uses optical flow motion estimation technology to perform time-series frame segmentation on the coating image stream to obtain a continuous coating image frame sequence. This time-series analysis method effectively eliminates the visual interference caused by movements such as case rotation and improves the reliability of subsequent defect detection. Through frame-by-frame surface feature analysis, multiple static texture feature maps are constructed to comprehensively depict the detailed information of the coating surface. Fuzzy defect detection is performed based on these texture feature maps, and some tiny defects that are difficult to detect with the naked eye are preliminarily found. The defect area obtained by fuzzy defect detection , extract the defect unit image and perform super-resolution reconstruction. This super-resolution imaging technology effectively amplifies and clarifies the defect information, providing higher quality input for fine defect detection. Through step-by-step iterative decoding, the detailed features of the defects are restored to the greatest extent. Based on the super-resolution defect unit image, secondary fine defect detection is performed to find some subtle texture defects that are difficult to detect with the naked eye. The detected subtle defects are morphologically analyzed to obtain rich quantitative data such as the size and shape of the defects. Combined with the defect distribution density, it provides a basis for subsequent comprehensive evaluation. Through comprehensive evaluation of defect morphological data and distribution density, the quality status of the coating surface is comprehensively judged. This quantitative evaluation result not only intuitively reflects the defect situation, but also provides data support for production process optimization.
[0168] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.
[0169] What has been described above is only a specific implementation mode of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined in this article can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown in this article, but will conform to the widest scope consistent with the principles and novel features invented in this article.
Claims
1. A method for detecting defects in vacuum coating of watch cases based on image processing, characterized in that: The following steps are involved: Step S1: acquiring a multi-angle real-time watch case coating image stream; performing full-scale contrast enhancement on the multi-angle real-time watch case coating image stream to construct a full-scale convolution coating image stream; Step S2: performing optical flow motion estimation on the full-scale convolution coating image stream and performing time-series frame segmentation processing to obtain a coating image frame sequence; Step S3: Analyze the surface features of the coating image frame sequence frame by frame to construct multiple static texture feature maps; perform fuzzy defect detection on the multiple static texture feature maps to obtain texture defect data; Step S4: extracting a defective unit image based on the texture defect data; performing step-by-step iterative decoding on the defective unit image, and performing super-resolution imaging to construct a super-resolution defective unit image; Step S5: performing secondary fine defect detection on the super-resolution defect unit image to obtain fine texture defect data; performing defect morphology analysis on the fine texture defect data to generate defect morphology data and defect distribution density value; Step S6: Perform unit defect comprehensive evaluation processing on the defect morphology data and defect distribution density value to generate a coating defect detection result.
2. The method for detecting defects of vacuum coating of watch cases based on image processing according to claim 1, characterized in that: The specific steps of step S1 are: Step S11: Acquire multi-angle real-time watch case coating image stream; Step S12: Calculate the global average brightness of the multi-angle real-time case coating image to obtain the global brightness value of the image; Step S13: performing dark area recognition on the multi-angle real-time case coating image to extract the dark area of the image; Step S14: optimizing the dark pixel aberration of the dark area of the image based on the global brightness value of the image, thereby constructing a brightness optimized coating image stream; Step S15: Perform full-scale contrast enhancement on the brightness optimized coating image to construct a global convolution coating image stream.
3. The method for detecting defects of vacuum coating of watch cases based on image processing according to claim 2, characterized in that: The specific steps of step S15 are: Perform multi-scale spatial frequency domain decomposition on the brightness optimized coating image stream to obtain multiple component images with different spatial frequencies; Convolving the brightness optimized coating image stream component by component according to a plurality of component images of different spatial frequencies to obtain convolution response data of each component image; Performing nonlinear local mapping processing on the convolution response data of each component image to obtain convolution local mapping features; Based on the convolution local mapping feature, convolution contrast optimization is performed on multiple component images of different spatial frequencies to generate multiple convolution contrast optimized component images; Full-scale image reconstruction is performed on multiple convolutional contrast-optimized component maps to construct a full-scale convolutional coated image stream.
4. The method for detecting defects of vacuum coating of watch cases based on image processing according to claim 1, characterized in that: The specific steps of step S2 are: Step S21: performing optical flow motion estimation on the full-scale convolution coated image flow to obtain the image optical flow motion vector; Step S22: performing motion change analysis between consecutive frames based on the image optical flow motion vector to generate image flow frame-to-frame motion change data; Step S23: performing optical flow motion trajectory evolution on the inter-frame motion change data of the image stream to generate optical flow motion trajectory features; Step S24: performing time-series frame segmentation processing on the full-scale convolution coating image stream based on the optical flow motion trajectory characteristics, thereby obtaining a coating image frame sequence.
5. The method for detecting defects of vacuum coating of watch cases based on image processing according to claim 4 is characterized in that: The specific steps of step S24 are: Performing motion pattern statistical analysis on the image optical flow motion vector to generate motion vector pattern features; Based on the motion vector pattern features, the optical flow motion trajectory features are clustered and divided to generate multiple trajectory feature clusters; Based on multiple trajectory feature clustering, the corresponding regions of the full-scale convolution coating image stream are identified, and the regional motion cycle period analysis is performed to obtain the main cycle motion period of the image; Defining the candidate time window length according to the main cycle motion period of the image; Based on the candidate time window length, the full-scale convolution coated image stream is subjected to frame-by-frame correlation calculation within the time window to obtain the inter-frame image correlation within the time window; According to the inter-frame image correlation within the time window, the full-scale convolution coating image stream is subjected to time-series frame segmentation processing, thereby obtaining a coating image frame sequence; The time sequence frame segmentation process is specifically as follows: when the inter-frame image correlation within the time window is lower than the preset inter-frame image correlation, it is determined as a segmentation point; Perform image segmentation on the full-scale convolution coating image stream based on the segmentation points to obtain an image frame sequence; The time-series frame segmentation process is repeated until all image streams are traversed, and multiple image frame sequences are connected to obtain a coating image frame sequence.
6. The method for detecting defects of vacuum coating of watch cases based on image processing according to claim 1, characterized in that: The specific steps of step S3 are: Step S31: Analyze the surface characteristics of the coating image frame sequence frame by frame to generate the surface characteristic data of the watch case; Step S32: performing multi-frame texture feature fitting on the surface feature data of the watch case to construct multiple static texture feature maps; Step S33: performing discontinuous mutation analysis on multiple static texture feature maps to obtain texture discontinuous mutation points; Step S34: extracting image frames from the full-scale convolution coating image stream based on the texture discontinuity mutation point to obtain a frame image where the texture mutation occurs; Step S35: performing blur defect detection on the frame image where the texture mutation occurs, so as to obtain texture defect data.
7. The method for detecting defects of vacuum coating of watch cases based on image processing according to claim 1, characterized in that: The specific steps of step S4 are: Step S41: performing unit division processing on the frame image where the texture mutation occurs to obtain a plurality of unit images; Step S42: locating defective units on a plurality of unit images based on the texture defect data, and extracting defective unit images; Step S43: performing multi-layer feature compression coding on the defective unit image to obtain a low-dimensional feature coding of the image; Step S44: performing step-by-step iterative decoding based on the low-dimensional feature coding of the image, and performing super-resolution imaging to construct a super-resolution defective unit image.
8. The method for detecting defects of vacuum coating of watch cases based on image processing according to claim 1, characterized in that: The specific steps of step S5 are: Step S51: performing visual feature recognition on the super-resolution defective unit image to generate defective unit visual feature data; Step S52: performing secondary fine defect detection on the defect unit visual feature data, thereby obtaining fine texture defect data; Step S53: performing defect edge contour analysis on the subtle texture defect data and marking the defect edge contour line; Step S54: performing defect morphology analysis on the defect edge contour line to generate defect morphology data; Step S55: performing defect distribution density calculation on the super-resolution defective unit image to generate a defect distribution density value.
9. The method for detecting defects of vacuum coating of watch cases based on image processing according to claim 1, characterized in that: The specific steps of step S6 are: Step S61: Calculate the defect position of the full-scale convolution coating image stream based on the subtle texture defect data to generate unit defect position coordinates; Step S62: performing stereo distortion correction on the unit defect position coordinates to obtain the precise position of the watch case defect; Step S63: Perform unit defect comprehensive evaluation processing on the defect morphology data and defect distribution density value according to the precise position of the watch case defect to generate a coating defect detection result.
10. A watch case vacuum coating defect detection system based on image processing, characterized in that: The method for detecting defects of vacuum coating on a watch case based on image processing as claimed in claim 1 comprises: An image enhancement module is used to obtain a multi-angle real-time watch case coating image stream; perform full-scale contrast enhancement on the multi-angle real-time watch case coating image stream to construct a full-scale convolution coating image stream; A time-series frame segmentation module is used to perform optical flow motion estimation on the full-scale convolution coating image stream and perform time-series frame segmentation processing to obtain a coating image frame sequence; The fuzzy defect detection module is used to analyze the surface features of the coating image frame sequence frame by frame to construct multiple static texture feature maps; perform fuzzy defect detection on the multiple static texture feature maps to obtain texture defect data; A super-resolution imaging module is used to extract defective unit images based on texture defect data; iteratively decode the defective unit images step by step, and perform super-resolution imaging to construct super-resolution defective unit images; The fine defect detection module is used to perform fine defect secondary detection on the super-resolution defect unit image to obtain fine texture defect data; and is used to perform defect morphology analysis on the fine texture defect data to generate defect morphology data and defect distribution density values; The defect assessment module is used to perform comprehensive evaluation of unit defects on defect morphology data and defect distribution density values to generate coating defect detection results.
