RFID Chip Image Abnormal Detection Method and System Based on Intelligent Learning

Abstract

The present invention relates to the technical field of RFID chip manufacturing, and particularly to a method and system for abnormal detection of RFID chip images based on intelligent learning. The method includes the following steps: obtaining a real-time RFID chip sampling image set and performing feature alignment based on a registration template, generating an average grayscale image using the aligned image set, and obtaining a chip edge mask image set using an edge search algorithm; segmenting the RFID chip edge mask image set to obtain an RFID chip effective edge mask variance image set; integrating the variance image set to obtain a pre-generated variance map; obtaining RFID chip on-site environmental data and performing sensitive coefficient estimation to obtain an edge sensitive coefficient; performing convolutional filtering and smoothing processing on the pre-generated variance map to obtain a chip edge variance map, and constructing an RFID chip image abnormal detection model based on the edge variance map and the average grayscale image. The present invention improves the accuracy and stability of detection.

Description

Technical Field

[0001] The present invention relates to the technical field of RFID chip manufacturing, and particularly to a method and system for abnormal detection of RFID chip images based on intelligent learning. Background Art

[0002] Regarding the identification of abnormal regions in RFID, traditional intelligent learning algorithms automatically generate an average image and a variance image based on standard samples. If the tension of the on-site chips is unstable or there are differences in the quality of the backing paper, etc., resulting in a slight deformation of the final image compared with the standard samples, the intelligent learning algorithm will lead to strict image judgment (false alarms). Especially when conducting quality inspections on transparent PET chips, a large number of false alarms will seriously affect the online production. The currently commonly used intelligent learning algorithm is to continuously collect on-site qualified samples and generate an average image and a variance image based on the brightness change and stability change of the samples themselves. The intelligent learning algorithm constructed in this mode can detect obvious abnormal situations such as ink dots, surface tapes, and dark stains, etc., but it is easy to miss some abnormal regions with lighter colors and less obvious features. If the color difference range for algorithm inspection is adjusted smaller, the corresponding false alarm rate will increase. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a method and system for abnormal detection of RFID chip images based on intelligent learning to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for abnormal detection of RFID chip images based on intelligent learning includes the following steps:

[0005] Step S1: Obtain a real-time RFID chip sampling image set; register a feature template for the real-time RFID chip sampling images and perform image alignment based on the feature template to obtain an aligned image set, and calculate the average grayscale image of the aligned image set to obtain an average grayscale image; perform RFID chip edge search based on the aligned image set to obtain an RFID chip edge mask image set;

[0006] Step S2: Perform edge mask image grayscale enhancement on the RFID chip edge mask image set to obtain an edge mask grayscale enhanced image set, and perform self-adaptive threshold segmentation on the edge mask grayscale enhanced image set to obtain an RFID chip effective edge mask image set;

[0007] Step S3: Perform an effective mask graph union process on the RFID chip effective edge mask image set to obtain an RFID chip mask union image, and perform common edge feature integration based on the RFID chip mask union image to obtain a pre-generated variance image;

[0008] Step S4: Obtain the on-site environmental data of the RFID chip, and evaluate the coil density according to the on-site environmental data of the RFID chip, so as to obtain the on-site environmental coil density data; estimate the edge sensitivity coefficient according to the on-site environmental coil density data, so as to obtain the edge sensitivity coefficient;

[0009] Step S5: Perform convolution filtering and smoothing on the pre-generated variance map according to the edge sensitivity coefficient, so as to obtain the chip edge variance map, and construct an RFID chip image anomaly detection model based on the chip edge variance map and the average grayscale map; detect the anomaly regions of the sample images in the real-time RFID chip sampled image set through the RFID chip image anomaly detection model, so as to obtain the RFID chip anomaly region image set.

[0010] The present invention ensures that the detection system can promptly reflect the actual situation on the production line by collecting RFID chip images in real time, effectively improving the timeliness of detection. By aligning and calculating the average grayscale images of the real-time RFID chip sampling image set, the influence of factors such as illumination changes and angle deviations during the image acquisition process on the image quality can be eliminated, thereby providing more stable and unified basic data for subsequent image analysis. By finding edges and generating an edge mask image set, the contour features of the RFID chip can be effectively extracted, which is crucial for the precise positioning and defect detection of the chip. Further, by enhancing the grayscale and performing adaptive threshold segmentation on the edge mask image, the contrast of the edge information can be enhanced, enabling more subtle abnormal features (such as defects with lighter colors and less obvious contrast) to be more easily detected, reducing the risk of missed detections. By performing a union operation on the effective mask image set, the edge information in multiple sampling images can be integrated, the influence of noise in individual images can be eliminated, and a more stable and consistent pre-generated variance map can be obtained. The pre-generated variance map provides reliable basic data for subsequent anomaly detection and can more accurately reflect the overall characteristics and changes of the image. Obtaining on-site environmental data and evaluating the coil density can effectively consider the influence of environmental factors on the detection results. By evaluating the coil density, the degree to which the image quality may be affected can be understood. By calculating the edge sensitivity coefficient, the edge response ability under different environmental conditions can be quantified, and then the subsequent image processing strategy can be adjusted to ensure that the algorithm can operate stably in various situations. Using the edge sensitivity coefficient to perform convolution filtering on the pre-generated variance map can effectively smooth the image, suppress the influence of random noise, and enhance the clarity of the edge features. The anomaly detection model constructed based on the processed chip edge variance map has higher accuracy, can promptly and effectively identify the abnormal areas in the real-time RFID chip image set, reduce the false alarm rate, and improve production efficiency. In summary, this series of steps forms a complete RFID chip anomaly detection process by combining real-time data collection, image processing algorithms, and environmental factor analysis. This process not only improves the accuracy and stability of detection but also effectively reduces the false alarms and missed detections that may occur in the practical application of traditional algorithms. Through refined processing and intelligent analysis, this technology can significantly improve the production quality of RFID chips and provide strong support for the development of the industry.

[0011] Optionally, step S1 is specifically as follows:

[0012] Step S11: Obtain a real-time RFID chip sampling image set;

[0013] Step S12: Perform preprocessing on the real-time RFID chip sampling image set to obtain an RFID chip sampling image set;

[0014] Step S13: Randomly select pictures from the real-time RFID chip sampling image set to obtain a random sampling image set, and perform feature registration based on the random sampling image set to obtain a feature template;

[0015] Step S14: Align the remaining images in the real-time RFID chip sampling image set according to the feature template to obtain an aligned image set, and calculate the average grayscale image of the aligned image set to obtain an average grayscale image;

[0016] Step S15: Perform edge detection on the aligned image set to obtain an RFID chip edge feature image set;

[0017] Step S16: Perform binary edge mask image conversion according to the RFID chip edge feature image set to obtain an RFID chip edge mask image set.

[0018] In the present invention, by preprocessing the real-time RFID chip sampling image set, noise or unnecessary interference factors in the original image can be removed, making subsequent image processing more accurate. Random picture selection and feature registration help obtain a more representative image template, ensuring the accurate extraction of image features, and thus providing a high-quality data basis for subsequent image alignment and processing. The step of aligning the remaining images in the real-time RFID chip sampling image set minimizes the differences between different sampling images, thereby improving the accuracy of subsequent analysis, especially playing an important role in image registration and unified feature extraction. By calculating the average grayscale image, the overall brightness and contrast of the image can be standardized, enabling images under different lighting conditions to be effectively compared and analyzed. Edge detection can highlight the key features in the RFID chip image, helping to identify the contour and other important detail information of the chip, which is crucial for chip positioning and feature extraction. The binary edge mask image conversion step helps to further remove irrelevant background information and only retain the edge features related to the RFID chip, thus providing more accurate image data for subsequent analysis and processing of the chip.

[0019] Optionally, step S2 is specifically as follows:

[0020] Step S21: Perform pixel binary statistics according to the RFID chip edge mask image set to obtain a high-gradient change region image set and a low-gradient change region image set;

[0021] Step S22: Perform local adaptive histogram equalization on the high-gradient change region image set to obtain a high-gradient change region grayscale enhanced image set; perform edge contrast enhancement on the low-gradient change region image set to obtain a low-gradient change region enhanced image set;

[0022] Step S23: Merge the grayscale-enhanced image sets of the high-gradient change regions and the enhanced image sets of the low-gradient change regions to obtain the edge mask grayscale-enhanced image set;

[0023] Step S24: Perform adaptive threshold segmentation on the edge mask grayscale-enhanced image set to obtain the effective edge mask image set of the RFID chip.

[0024] Through pixel binary statistics on the edge mask image set of the RFID chip, the present invention can accurately divide the high-gradient change regions and low-gradient change regions in the image. This process can effectively distinguish the important edge information and relatively flat background regions in the image, laying a good foundation for subsequent image processing. When performing local adaptive histogram equalization on the high-gradient change regions, the grayscale distribution of the image is enhanced, which helps to improve the visibility of edge details in the image and makes the edge contour of the chip clearer. By enhancing the edge contrast of the low-gradient change regions, the contrast of the regions in the image that are relatively blurred or have fewer details can be effectively improved, enhancing the details of the overall image and further improving the image quality. Then, the edge mask grayscale-enhanced image set obtained by merging the images of the two types of regions helps to comprehensively reflect the advantageous information of the high-gradient and low-gradient regions, thereby obtaining a more balanced and accurate overall image, providing good input conditions for subsequent threshold segmentation processing. Through the adaptive threshold segmentation technology, the effective edge region of the RFID chip can be accurately extracted from the enhanced image. This process can significantly improve the recognition effect of the effective edges in the image and ensure more accurate positioning and reading of the RFID chip. Overall, the design of this series of steps not only improves the grayscale and contrast of the image but also ensures the processing effect of the edge mask image set of the RFID chip through multi-level and multi-angle image enhancement, thereby effectively improving the accuracy and stability of RFID chip recognition and reading.

[0025] Optionally, step S24 is specifically as follows:

[0026] Step S241: Calculate the local thresholds of the gradient change regions for the edge mask grayscale-enhanced image set to obtain the local threshold data of the high-gradient change regions and the local threshold data of the low-gradient change regions;

[0027] Step S242: Perform threshold segmentation on the edge mask grayscale-enhanced image set according to the local threshold data of the high-gradient change regions to obtain the functional region threshold segmentation image set; perform boundary region threshold segmentation on the edge mask grayscale-enhanced image set according to the local threshold data of the low-gradient change regions to obtain the boundary region threshold segmentation image set;

[0028] Step S243: Perform an intersection operation on the boundary region threshold segmentation image set and the functional region threshold segmentation image set to obtain the background-entangled functional region image set;

[0029] Step S244: Perform logical operation pixel merging on the background-entangled functional region image set and the functional region threshold segmentation image set to obtain the RFID chip effective edge mask image set.

[0030] The present invention calculates the local threshold of the gradient change region for the edge mask gray-scale enhanced image set, which can process different regions with high and low gradient changes in the image separately. This makes the threshold segmentation of the image more flexible and adaptable. High-gradient regions usually represent obvious edges or features, while low-gradient regions may contain smoother or background parts. By calculating the local threshold for these two different regions separately, it can ensure more precise segmentation in regions with rich details, and perform loose threshold processing in regions with background or blurred edges, thus avoiding over-segmentation or omission while retaining key features. By performing functional region threshold segmentation on the edge mask gray-scale enhanced image set based on the local threshold data of the high-gradient change region, the regions representing the core functional parts in the image can be extracted from the complex background, ensuring that the extracted functional regions are of high quality and accurate. At the same time, using the local threshold data of the low-gradient region to perform threshold segmentation on the boundary region of the image further enhances the extraction effect of the edge part in the image, avoiding misjudgment or omission problems caused by image noise, and thus ensuring that the finally extracted boundary region is more accurate and clear. By performing an intersection operation on the functional region and the boundary region, background interference and blurred boundary regions can be removed to obtain the background-entangled functional region image set. This step helps to focus on the most representative and relevant parts of the image, further optimizing the image quality and laying a foundation for subsequent pixel merging operations. When performing logical operation pixel merging on the background-entangled functional region image set and the functional region threshold segmentation image set, the effective information of the two can be accurately fused to obtain the RFID chip effective edge mask image set. This step ensures the accuracy of the edge and functional regions in the final image through intelligent pixel merging, removes redundant or unnecessary regions, and significantly improves the effectiveness and accuracy of the image. This result can not only provide high-quality image data in subsequent image analysis and recognition, but also enhance the reliability and efficiency of the entire system in practical applications, especially having important application value in the precise positioning and recognition of RFID chips.

[0031] Optionally, step S3 is specifically as follows:

[0032] Step S31: Perform pixel binary statistics on the RFID chip effective edge mask image set to obtain the pixel binary data of the effective edge image set;

[0033] Step S32: Select the union layer of the RFID chip effective edge mask image set according to the pixel binary data of the effective edge image set to obtain the effective edge mask image set to be unioned;

[0034] Step S33: Set the geometric tolerance according to the pixel binary data of the effective edge image set to obtain the union geometric tolerance data;

[0035] Step S34: Perform the effective mask image union processing on the effective edge mask image set to be unioned according to the union geometric tolerance data to obtain the RFID chip mask union image;

[0036] Step S35: Integrate the common edge features according to the RFID chip mask union image to obtain the pre-generated variance map.

[0037] Through the pixel binary statistics on the effective edge mask image set, the present invention can clearly identify and quantify the effective edges of the RFID chip, eliminate noise and interference, and improve the accuracy of image analysis. The obtained binary data provides an important quantitative basis for subsequent processing, making the features of each edge more distinct and helping to formulate a more reasonable image processing strategy. Through the selection of the union layer of the effective edge image set, the effective edge information in different images can be merged to form a more complete edge mask image set. This integration helps to extract more valuable information, can effectively reduce the interference of duplicate information, and helps to improve the efficiency of subsequent processing. By setting the geometric tolerance for the pixel binary data of the effective edge image set, the allowable range of edge features can be precisely regulated according to different manufacturing and detection requirements, thereby improving product consistency. The setting of geometric tolerance enables the system to adapt to the changes in different manufacturing processes and materials, improving the adaptability of the RFID chip in different application scenarios. By processing the effective edge mask image set to be unioned, the effective mask image of the RFID chip can be further optimized, improving the clarity and usability of the image and enhancing the accuracy of subsequent analysis and recognition. After the union processing of the effective mask image, the system can work at a higher precision, improving the performance of the entire detection process. Integrating the common edge features of the RFID chip mask union image can more comprehensively identify the key features of the RFID chip and provide a more representative sample. The finally generated pre-generated variance map can not only reflect the stability of the edge features, but also provide a basis for subsequent quality control and anomaly detection, helping to timely discover potential problems.

[0038] Optionally, step S35 is specifically as follows:

[0039] Step S351: Extract the edge structure features of the RFID chip mask union image to obtain image edge structure data;

[0040] Step S352: Calculate the similarity based on the image edge structure data to obtain edge structure similarity data;

[0041] Step S353: Perform edge structure clustering calculation on the image edge structure data based on the edge structure similarity data to obtain edge structure clustering data, and perform common edge feature integration based on the edge structure clustering data to obtain common edge feature data;

[0042] Step S354: Calculate the edge feature variance based on the common edge feature data to obtain feature variance data, and visualize the feature variance data to obtain a pre-generated variance map.

[0043] By extracting the edge structure features, the present invention can more accurately identify important information in the image and reduce the misrecognition rate. The extraction of edge features can reduce the amount of image data, facilitating subsequent processing and storage. Under conditions such as light changes and noise interference, the edge feature extraction algorithm is usually more robust, helping to improve the overall reliability of the system. The similarity data can help classify similar images into the same category, improving the classification efficiency. By calculating the similarity between images, relevant images can be quickly retrieved in a large-scale image library, enhancing the user experience. By setting a similarity threshold, abnormal or unexpected images can be quickly identified, thereby improving the security of the system. Through clustering, similar edge features can be integrated, improving the data representation ability and compactness. Through clustering, multiple similar features are grouped into one category, which can reduce the complexity of subsequent calculations and improve the processing speed. Clustering analysis can help identify potential patterns or structures in the image, enhancing the depth and breadth of data analysis. Through the analysis of feature variance, the stability of features under different conditions can be evaluated, helping to judge the reliability of features. Visualizing the feature variance can intuitively display the dispersion of different features, facilitating subsequent decision-making and optimization. By analyzing the variance, the feature selection process can be optimized to select the most representative features, enhancing the system performance.

[0044] Optionally, step S4 is specifically as follows:

[0045] Step S41: Obtain the RFID chip on-site environment data;

[0046] Step S42: Extract the RFID chip coil distribution features from the RFID chip on-site environment data to obtain RFID chip coil distribution data;

[0047] Step S43: Evaluate the regional coil density based on the RFID chip coil distribution data to obtain the regional coil density data;

[0048] Step S44: Perform spatial association on the regional coil density data and the RFID chip edge feature image set to obtain the coil density-edge feature association data;

[0049] Step S45: Estimate the sensitivity coefficient based on the coil density-edge feature association data to obtain the edge sensitivity coefficient.

[0050] The present invention collects on-site environmental data (such as temperature, humidity, metal interference, etc.), which can help understand the impact of the environment on RFID signal transmission, provide a basis for subsequent data analysis, ensure that the entire evaluation process is evidence-based, and thus improve the overall performance of the system. By extracting the coil distribution characteristics, it is possible to identify the working state and signal propagation characteristics of RFID chips under different environmental conditions, provide feedback on coil design and deployment, and help optimize the layout of RFID devices to reduce signal interference and improve the recognition rate. Evaluating the coil density can reveal the impact of different regions on signal strength and quality, provide a basis for resource allocation, help identify signal differences between dense and sparse regions, facilitate corresponding adjustments in design, and ensure the balance and reliability of the system. Through spatial association analysis, it is possible to deeply understand the mutual influence between regional coil density and environmental characteristics, and reveal the law of signal transmission in complex environments. It provides important context information for subsequent sensitivity coefficient estimation, making data analysis more targeted and effective. The estimation of the sensitivity coefficient can quantify the impact of environmental factors on the performance of RFID chips and help identify key interference sources.

[0051] Optionally, step S42 is specifically:

[0052] Step S421: Extract the RFID chip signal environment characteristics and physical environment characteristics based on the RFID chip on-site environmental data to obtain the RFID chip signal environment data and the physical environment data;

[0053] Step S422: Perform statistics on the signal strength distribution according to the RFID chip signal environment data to obtain the RFID chip signal strength distribution data;

[0054] Step S423: Perform temporal association on the physical environment data and the RFID chip signal environment data to obtain the physical environment-signal environment association data, and estimate the physical environment impact factor based on the physical environment-signal environment association data to obtain the physical environment impact factor;

[0055] Step S424: Correct the signal strength distribution data of the RFID chip according to the physical environment impact factor, so as to obtain the actual signal strength distribution data of the RFID chip;

[0056] Step S425: Estimate the coil distribution of the RFID chip according to the actual signal strength distribution data of the RFID chip, so as to obtain the coil distribution data of the RFID chip.

[0057] By extracting the RFID chip signal environment data and physical environment characteristics, the present invention can accurately obtain the real environment of signal propagation. This provides basic data for subsequent analysis. The signal environment data helps to understand phenomena such as signal attenuation, reflection and multipath effect, and improves the signal analysis ability. Statistical signal strength distribution can help researchers intuitively understand the strength change of RFID signals at different positions, so as to better design and optimize the RFID system. Through distribution statistics, weak signal areas can be identified, providing a basis for subsequent optimization and signal enhancement measures. Through the temporal correlation between the physical environment data and the signal environment data, the specific impact of environmental changes on signal strength can be deeply understood. Estimating the physical environment impact factor enables the system to quantify the impact of environmental factors on signal propagation, providing a basis for formulating corresponding improvement measures. Through the correction of the physical environment impact factor, the accuracy of the actual RFID signal strength distribution data can be significantly improved, thereby enhancing the reliability of the RFID system. The corrected signal strength data can effectively reduce the errors caused by environmental factors and improve the performance of the system in actual applications. By estimating the coil distribution of the RFID chip through the actual signal strength distribution data, it can help designers optimize the coil layout, ensure the comprehensiveness of RFID system coverage and the uniformity of signals. A reasonable coil distribution can improve the identification efficiency of the RFID chip, reduce the identification time and cost, and improve the operation efficiency of the entire system.

[0058] Optionally, step S5 is specifically as follows:

[0059] Step S51: Perform convolution filtering smoothing processing on the pre-generated variance map according to the edge sensitivity coefficient, so as to obtain the chip edge variance map;

[0060] Step S52: Extract statistical features according to the chip edge variance map, so as to obtain the statistical feature data of the edge variance map;

[0061] Step S53: Perform edge anomaly recognition on the chip edge variance map according to the statistical feature data of the edge variance map, so as to obtain the abnormal image edge data and the normal image edge data;

[0062] Step S54: Construct an RFID chip image anomaly detection model according to the abnormal image edge data and the normal image edge data;

[0063] Step S55: Detect the abnormal regions of the sample images in the real-time RFID chip sampling image set through the RFID chip image anomaly detection model, so as to obtain the RFID chip abnormal region image set.

[0064] Through convolutional filtering of the pre-generated variance map, the present invention can effectively remove random noise in the image and improve the clarity of the edges. After smoothing processing, the features of the chip edge are more prominent, laying a foundation for subsequent feature extraction. By extracting the statistical features of the edge variance map, a quantitative description of the chip edge state can be obtained, and this data helps subsequent anomaly detection. Through effective feature extraction, the amount of data to be processed is reduced, improving the efficiency of subsequent processing. Based on the data analysis of statistical features, the abnormal conditions of the chip edge can be accurately identified, helping to quickly locate problems. By effectively dividing abnormal and normal data, the accuracy and reliability of the overall detection system are improved. By using abnormal and normal image edge data for training, the adaptability of the model to unknown data can be improved. After establishing an efficient detection model, real-time monitoring can be achieved, reducing manual intervention and improving production efficiency. Through real-time abnormal region detection, potential problems can be quickly discovered and processed, preventing a large number of unqualified products from entering the market. Through the feedback of real-time detection results, the model is continuously optimized to improve the performance and accuracy of the overall detection system.

[0065] Optionally, this specification also provides an RFID chip image anomaly detection system based on intelligent learning for executing the RFID chip image anomaly detection method based on intelligent learning as described above. The RFID chip image anomaly detection system based on intelligent learning includes:

[0066] An edge finding module, configured to obtain the real-time RFID chip sampling image set; register a feature template for the real-time RFID chip sampling image and perform image alignment based on the feature template to obtain an aligned image set, and calculate the average grayscale image of the aligned image set to obtain an average grayscale image; find the RFID chip edge according to the aligned image set to obtain the RFID chip edge mask image set;

[0067] A threshold segmentation module, configured to enhance the grayscale of the edge mask image in the RFID chip edge mask image set to obtain an edge mask grayscale enhanced image set, and perform self-adaptive threshold segmentation on the edge mask grayscale enhanced image set to obtain the RFID chip effective edge mask image set;

[0068] A common edge feature integration module is used to perform an effective mask union process on a set of RFID chip effective edge mask images to obtain an RFID chip mask union image, and perform common edge feature integration based on the RFID chip mask union image to obtain a pre-generated variance map;

[0069] An edge sensitivity coefficient estimation module is used to obtain RFID chip on-site environment data, and perform coil density evaluation based on the RFID chip on-site environment data to obtain on-site environment coil density data; perform edge sensitivity coefficient estimation based on the on-site environment coil density data to obtain an edge sensitivity coefficient;

[0070] An anomaly detection model construction module is used to perform convolutional filtering and smoothing on the pre-generated variance map according to the edge sensitivity coefficient to obtain a chip edge variance map, and construct an RFID chip image anomaly detection model based on the chip edge variance map and the average grayscale map; detect the anomaly regions of sample images in a real-time RFID chip sampled image set through the RFID chip image anomaly detection model to obtain an RFID chip anomaly region image set. Description of the Drawings

[0071] Other features, objectives, and advantages of the present invention will become more apparent by reading the detailed description of the non-restrictive embodiments with reference to the following drawings:

[0072] Figure 1 It is a schematic flowchart of the steps of the RFID chip image anomaly detection method based on intelligent learning of the present invention;

[0073] Figure 2 It is a detailed schematic flowchart of step S1 in the present invention;

[0074] Figure 3 It is a detailed schematic flowchart of step S2 in the present invention;

[0075] The realization, functional features, and advantages of the objectives of the present invention will be further described with reference to the embodiments and the drawings. Detailed Embodiments

[0076] The technical method of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0077] In addition, the attached drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0078] It should be understood that although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed related items.

[0079] To achieve the above object, please refer to Figures 1 to 3 , the present invention provides a method for detecting abnormal RFID chip images based on intelligent learning, and the method includes the following steps:

[0080] Step S1: Obtain a real-time RFID chip sampling image set; register a feature template for the real-time RFID chip sampling images and perform image alignment based on the feature template to obtain an aligned image set, and calculate an average grayscale image for the aligned image set to obtain an average grayscale image; perform RFID chip edge detection on the aligned image set to obtain an RFID chip edge mask image set;

[0081] In this embodiment, the images of the RFID chips are obtained in real time through a high-resolution camera or an RFID reading device, and the image sampling frequency is set to 10 frames per second to ensure clear capture of dynamic objects. Image processing algorithms are used to register the feature templates for these images. Through template-based alignment technology, images from different perspectives can be accurately aligned. After image alignment, an average grayscale image reflecting the overall brightness distribution is generated by calculating the average grayscale values of these images. Then, edge detection algorithms (such as Canny edge detection) are used to perform edge detection on the aligned image set, and finally the chip edge mask images are extracted.

[0082] Step S2: Perform edge mask image grayscale enhancement on the RFID chip edge mask image set to obtain an edge mask grayscale enhanced image set, and perform self-adaptive threshold segmentation on the edge mask grayscale enhanced image set to obtain an RFID chip effective edge mask image set;

[0083] In this embodiment, histogram equalization and gamma correction techniques are applied to the obtained edge mask image set for gray-scale enhancement to improve the contrast and recognizability of the edges. For example, the gamma value of the image is adjusted to 2.2 to enhance the details of the bright part. Then, the Otsu method is used for adaptive threshold segmentation to obtain the effective edge mask image set of the RFID chip. This step ensures that the clear edges of the chip are extracted and the interference noise is filtered out.

[0084] Step S3: Perform a union operation on the effective mask images of the RFID chip to obtain the union image of the RFID chip mask, and integrate the common edge features according to the union image of the RFID chip mask to obtain the pre-generated variance map;

[0085] In this embodiment, for the effective edge mask image set, a logical "AND" operation is performed to form the union image of the RFID chip mask. In this process, by selecting appropriate union operation rules, it is ensured that the edge information in all images is integrated. At the same time, by extracting edge features (such as edge length, curvature, etc.), the common edge features of each mask are calculated. The finally generated pre-generated variance map will show the edge changes of the RFID chip in different images, and image statistical analysis tools (such as mean and standard deviation calculations) are used to quantify the stability and variability of the edges.

[0086] Step S4: Obtain the on-site environment data of the RFID chip, and evaluate the coil density according to the on-site environment data of the RFID chip to obtain the on-site environment coil density data; estimate the edge sensitivity coefficient according to the on-site environment coil density data to obtain the edge sensitivity coefficient;

[0087] In this embodiment, sensors are deployed to collect on-site environment data, such as temperature, humidity, signal strength, etc. Combining the electromagnetic field strength in the area where the RFID chip is located, the coil density is evaluated through known algorithms (such as electromagnetic field simulation models). Assuming there are 5 RFID readers in a certain area, the coil density can be set to 5 / area (unit: square meter). According to the coil density data, the edge sensitivity coefficient is calculated. For example, using linear regression analysis, the edge sensitivity coefficient value is obtained as a parameter for subsequent processing.

[0088] Step S5: Perform convolution filtering and smoothing on the pre-generated variance map according to the edge sensitivity coefficient to obtain the chip edge variance map, and construct an RFID chip image anomaly detection model based on the chip edge variance map; detect the anomaly regions of the sample images in the real-time RFID chip sampling image set through the RFID chip image anomaly detection model to obtain the RFID chip anomaly region image set.

[0089] In this embodiment, convolution filtering is applied to the pre-generated variance map based on the obtained edge sensitivity coefficient. A suitable convolution kernel (such as a 3x3 Gaussian filter) is selected to smooth the image and reduce the influence of noise, enhancing the saliency of edge features. After the filtering process, a chip edge variance map is obtained. Then, a deep learning model (such as a convolutional neural network) is used to construct an RFID chip image anomaly detection model. By training the model and inputting a real-time RFID chip sampling image set, the model will generate a grayscale threshold range map. If the grayscale of some regions is not within the grayscale threshold range after the new image is aligned, such regions are anomaly regions, so as to detect the anomaly regions in the image, and finally output an RFID chip anomaly region image set, providing data support for subsequent analysis and processing.

[0090] Optionally, step S1 is specifically as follows:

[0091] Step S11: Obtain a real-time RFID chip sampling image set;

[0092] In this embodiment, a high-definition camera or an industrial camera is used, and through a real-time image acquisition device, an image of an object equipped with an RFID chip is captured. This device needs to be configured to be able to capture high-quality images under different lighting conditions and have a high resolution (such as 1080p or higher resolution) to ensure that the details of the RFID chip surface and its surrounding environment can be clearly captured. To ensure real-time performance, the image acquisition device can be connected to a computer or an embedded processing system to transmit the image data stream to the data processing system in real time. During the image acquisition process, the system should mark a time stamp for each frame of the image for sequential processing during subsequent image analysis. Suppose in a logistics warehouse, goods equipped with RFID tags are moving on a conveyor belt. An industrial camera installed above the conveyor belt is used to obtain RFID chip images at a speed of 10 frames per second. Each captured image can be stored in PNG format for use in subsequent image processing stages.

[0093] Step S12: Perform preprocessing on the real-time RFID chip sampling image set to obtain an RFID chip sampling image set;

[0094] In this embodiment, the acquired image set is preprocessed to enhance the image quality and improve the effect of subsequent feature extraction. The preprocessing steps include denoising, normalization, and cropping. Gaussian blur or median filtering algorithm is used for denoising to reduce the random noise in the image. Normalization adjusts the pixel values of the image to a unified range (such as between 0 and 1) to avoid the influence of image brightness differences on subsequent processing. Cropping removes the background part unrelated to the RFID chip in the image and only retains the RFID chip area. The preprocessed images can be unified to a fixed size (such as 256x256 pixels), and ensure that the area of the RFID chip in the image is at the center position for subsequent analysis. For example, in the captured image, there may be other interfering objects or overexposed areas around the RFID chip. By using a 3x3 Gaussian filter for denoising, the gray-scale fluctuations in the image are reduced; then histogram equalization processing is applied to enhance the image contrast and improve the visibility of the RFID chip surface.

[0095] Step S13: Randomly select pictures from the real-time RFID chip sampling image set to obtain a random sampling image set, and perform feature registration according to the random sampling image set to obtain a feature template;

[0096] In this embodiment, to reduce the computational complexity and improve the representativeness of feature extraction, a part of the images is randomly selected from the preprocessed image set (including N frames of images). Suppose the image set contains 1000 frames of images, and 200 frames are randomly selected for feature extraction. The randomly selected images can be realized by a random number generator to ensure that the distribution of the selected images in the entire image set is uniform. After the image selection, a feature matching algorithm (such as SIFT, SURF, etc.) is used for feature registration. The purpose of feature registration is to match the similar feature points in these images to generate a template feature set for subsequent image alignment and analysis. For example, 200 frames are randomly selected from a 1000-frame image set. By calculating the similarity between the images, key feature points are extracted from these images using the SIFT algorithm. Then, these feature points are matched to generate a feature template. This template will be used for subsequent image alignment and analysis.

[0097] Step S14: Align the remaining images in the real-time RFID chip sampling image set according to the feature template to obtain an aligned image set, and calculate the average grayscale image of the aligned image set to obtain the average grayscale image;

[0098] In this embodiment, the obtained feature template is used to perform image registration or alignment on the remaining images. Through a feature matching algorithm, such as an image registration method based on the RANSAC algorithm, the transformation matrix between the images is calculated to ensure that the positions of the RFID chips in each frame of the image are aligned with the template. The process of image alignment can involve geometric transformations such as rotation, scaling, and translation, with the aim of making the RFID chip regions in each frame of the image highly coincide with the feature template. The average grayscale image of the aligned images can be calculated as an evaluation criterion for the alignment quality. Assuming that the template image is a sample image of an RFID chip after feature registration, the transformation matrix is calculated and applied through the RANSAC algorithm and applied to other sampled images. The aligned images will present a clear structure of the RFID chip region, facilitating subsequent grayscale calculation and feature extraction.

[0099] Step S15: Perform edge detection on the aligned image set to obtain an RFID chip edge feature image set;

[0100] In this embodiment, edge detection is performed on the aligned image set using classical edge detection algorithms such as Canny edge detection, Sobel operator, or Laplacian operator. By calculating the gradient of the image, regions with large grayscale changes in the image are extracted, especially the edge parts of the RFID chips. Edge detection helps to highlight important features in the image, such as the contours of RFID chips, metal pins, label printing, etc. In practical applications, the edge detection algorithm can be optimized by adjusting the threshold to obtain the best edge feature map. For example, using the Canny edge detection algorithm to process the aligned RFID chip image, setting the low threshold to 50 and the high threshold to 150, effectively extracts the edge information of the RFID chip. The edge image will show the contour of the chip and other obvious features, facilitating further analysis.

[0101] Step S16: Perform binary edge mask image conversion according to the RFID chip edge feature image set to obtain an RFID chip edge mask image set.

[0102] In this embodiment, by binarizing the edge detection result, the edge feature map is converted into an edge mask map. Binarization can be achieved by setting a fixed threshold or an automatic threshold algorithm (such as the Otsu algorithm), setting the edge part in the image to white and the non-edge part to black. The obtained edge mask image set can effectively identify the edge region of the RFID chip and is used for subsequent morphological analysis, size detection, etc. The binary mask image is of great significance for feature extraction, morphological analysis, and image segmentation during the processing. For example, the Otsu algorithm is used to automatically calculate the optimal threshold for the edge detection image, and then the edge region is set to white and the non-edge region is set to black to form a clear edge mask image set. These mask images can help the system accurately identify the shape and structure of the RFID chip and ensure more accurate subsequent processing.

[0103] Optionally, step S2 is specifically as follows:

[0104] Step S21: Perform pixel binary statistics based on the RFID chip edge mask image set to obtain a high-gradient change region image set and a low-gradient change region image set;

[0105] In this embodiment, the gradient of each pixel in the RFID chip edge mask image set is calculated, and the gradient value is obtained by calculating the difference between adjacent pixels. According to the set gradient threshold, the pixels in the image are divided into a "high-gradient change region" and a "low-gradient change region". The high-gradient region usually corresponds to the parts with obvious edges or rapid changes in the image, such as the contact points or structural details of the RFID chip; the low-gradient region is the part with slow changes or relatively flat areas in the image, such as the background region or large internal flat areas. Through this operation, the image is divided into two sets: one is the high-gradient change region image set, and the other is the low-gradient change region image set.

[0106] Step S22: Perform local adaptive histogram equalization on the high-gradient change region image set to obtain a high-gradient change region gray-scale enhanced image set; perform edge contrast enhancement on the low-gradient change region image set to obtain a low-gradient change region enhanced image set;

[0107] In this embodiment, different image enhancement processes are respectively performed on the high-gradient change region image set and the low-gradient change region image set to improve the image quality for subsequent analysis. For the high-gradient change region image set, the Adaptive Histogram Equalization (AHE) method is used to enhance the image contrast. The core idea of the AHE algorithm is to divide the image into multiple small regions (e.g., local windows), calculate the histogram and perform equalization processing within each local region. Different from the traditional global histogram equalization, AHE can perform local adjustment according to the brightness distribution of different regions, thereby improving the detail performance of the high-gradient region. The size of each window (e.g., 16x16 or 32x32 pixels) can be set, and the pixel gray values of each window are adjusted according to the local histogram, making the detail regions in the image more prominent, especially at the tiny structures and edges of the RFID chip. For the low-gradient change region image set, edge contrast enhancement technology is used. The purpose of this step is to improve the clarity of the image by increasing the contrast of the edge part in the image. Common edge contrast enhancement methods include edge enhancement techniques based on the Laplacian operator or Laplacian of Gaussian (LoG). The edge information of the image is extracted through a filtering algorithm, the Laplacian operator is used to detect the edges of the image, and then the contrast of the edge part is enhanced to highlight the details. For example, the weighted average method can be used to adjust the brightness of the edge pixels, and the details of the image are enhanced by increasing the contrast of these regions. First, the low-frequency part (background) of the image is extracted, and then the contrast of the low-gradient region is increased through methods such as contrast stretching or gamma correction, so as to more clearly display the details and edges on the surface of the RFID chip.

[0108] Step S23: Perform corresponding image merging on the high-gradient change region gray-scale enhanced image set and the low-gradient change region enhanced image set to obtain an edge mask gray-scale enhanced image set;

[0109] In this embodiment, the high-gradient change region gray-scale enhanced image set and the low-gradient change region enhanced image set are merged. An image weighted fusion method is adopted, such as simple weighted summation or a fusion algorithm based on pixel similarity. A merging coefficient can be defined (e.g., 0.7 for the high-gradient change region image and 0.3 for the low-gradient change region image), thereby generating an edge mask gray-scale enhanced image set. For example, it can be clearly seen from the merged image that the clarity of the edge region is improved and the details of the overall image are richer, which is helpful for subsequent processing.

[0110] Step S24: Perform functional region adaptive threshold segmentation according to the edge mask gray-scale enhanced image set to obtain an RFID chip effective edge mask image set.

[0111] In this embodiment, adaptive threshold segmentation of functional regions is performed based on the edge mask gray-scale enhanced image set. Here, the Otsu's method or an adaptive segmentation algorithm based on region growing can be adopted to dynamically select the threshold according to the gray-scale distribution of the image. During implementation, first, the gray-scale histogram of each image is calculated, and then the best threshold is selected for segmentation according to the principle of maximizing the variance between classes. Finally, an effective edge mask image set of the RFID chip is obtained. For example, if the effective edge region of the RFID chip is segmented from the processed image and the background noise is successfully suppressed, the contour of the effective edge mask image is clear and suitable for further analysis and application.

[0112] Optionally, step S24 is specifically as follows:

[0113] Step S241: Calculate the local threshold of the gradient change region for the edge mask gray-scale enhanced image set, so as to obtain the local threshold data of the high-gradient change region and the local threshold data of the low-gradient change region;

[0114] In this embodiment, the edge detection algorithm (such as Sobel operator, Canny operator, etc.) is used to process the gray-scale enhanced image set, and the gradient value of each pixel is calculated. During this process, first, the gradient amplitude of each pixel point in the image is obtained. Then, a gradient threshold is set, and the regions with larger gradient amplitudes are calculated and classified as high-gradient change regions, and the regions with smaller gradient amplitudes are classified as low-gradient change regions. The high-gradient change regions are usually the edges or details of objects, while the low-gradient change regions are the background or flat regions. Next, through local threshold calculation, the local threshold of each region is determined to adapt to the image features of different regions. For example, in an image containing an RFID chip, the gradient change at the edge is large, while the flat regions such as the background have a small gradient change. In this way, two data sets can be obtained: the local threshold data of the high-gradient change region and the local threshold data of the low-gradient change region.

[0115] Step S242: Perform threshold segmentation of the functional regions on the edge mask gray-scale enhanced image set according to the local threshold data of the high-gradient change region, so as to obtain a functional region threshold segmentation image set; perform threshold segmentation of the boundary regions on the edge mask gray-scale enhanced image set according to the local threshold data of the low-gradient change region, so as to obtain a boundary region threshold segmentation image set;

[0116] In this embodiment, based on the calculated local threshold data of the high-gradient change region, these data are used to perform threshold segmentation on the edge mask grayscale enhanced image to distinguish the functional region and the boundary region. The local threshold data of the high-gradient change region is used to binarize the image, and the functional region in the image is extracted. The functional region usually contains the target object in the image, such as the identification region or pattern of the RFID chip. In this process, by setting an appropriate threshold range, the target object can be extracted and separated from the background. The local threshold data of the low-gradient change region is used for binarization to segment the boundary region. The boundary region usually refers to the relatively flat or less detailed part of the image, such as the background region or the edge part around the target object. By analyzing the gradient magnitude of the image, an appropriate threshold can be set, for example, the threshold is set to 50. If the gradient value of a certain pixel is greater than 50, then it is considered as the high-gradient change region, which usually corresponds to the edge part of the object. The gradient threshold is used to binarize the image to extract the functional region. For example, the identification region of the RFID chip contains many complex details, and the gradient values corresponding to these details are relatively high. Therefore, after binarization, the external structure and detailed parts of the chip can be successfully extracted. For the low-gradient change region, a lower gradient threshold can be set, for example, the threshold is set to 10. If the gradient value of a certain pixel is less than 10, then this region is considered as the low-gradient change region, which usually corresponds to the background part of the image. By setting the low-gradient threshold, the background region in the image is extracted. These regions are usually flat and have little change, such as the white background part in the RFID chip image or other background regions of the image.

[0117] Step S243: Perform an intersection operation on the boundary region threshold segmentation image set and the functional region threshold segmentation image set to obtain the background-entangled functional region image set;

[0118] In this embodiment, the obtained boundary region threshold segmentation image set and the functional region threshold segmentation image set are subjected to an intersection operation. The purpose of the intersection operation is to obtain the part of the image that belongs to both the functional region and the boundary region. The intersection operation is performed by performing a pixel-by-pixel operation on the binarization results of the functional region and the boundary region, and comparing the "1"s (pixels indicating the existence of an object or region) in the two regions. If a pixel is "1" in both regions, it means that this pixel belongs to both the functional region and the boundary region. For example, in the image of the RFID chip, these regions usually represent the outer frame or circuit part of the chip.

[0119] Step S244: Perform logical operation pixel merging on the background-entangled functional region image set and the functional region threshold segmentation image set to obtain the RFID chip effective edge mask image set.

[0120] In this embodiment, the obtained background entanglement functional region image set and the functional region threshold segmentation image set are pixel - level merged using logical operations (such as "AND" operation or "OR" operation). The purpose is to eliminate the incorrect segmentation results caused by background interference and retain the truly effective regions in the image. For example, after the intersection operation, the obtained background entanglement functional region image set may contain some incorrect edge parts, while through pixel merging with logical operations, only the pixels containing the effective functional regions can be retained. For example, in the image of an RFID chip, after merging, the finally obtained image will clearly show the effective region of the chip, while removing the irrelevant background or noise. Finally, through processing, an effective edge mask image set of the RFID chip is obtained, which can effectively provide accurate image boundary data for subsequent image analysis or recognition. This step plays a key role in the final segmentation and target extraction of the image.

[0121] Optionally, step S3 is specifically as follows:

[0122] Step S31: Perform pixel - point binary statistics on the effective edge mask image set of the RFID chip to obtain the pixel - point binary data of the effective edge image set;

[0123] In this embodiment, pixel - point binary statistics are performed on the effective edge mask image set of the RFID chip to count the number of white pixels (value is 1) and black pixels (value is 0) of the effective edge. This process not only helps to identify the number of effective edges but also provides basic data for subsequent steps. For example, if the number of white pixels of the effective edge in a certain mask image is counted as 5000 and the number of black pixels is 15000, it indicates that the effective edge feature of this mask image is relatively obvious.

[0124] Step S32: Perform union layer selection on the effective edge mask image set of the RFID chip according to the pixel - point binary data of the effective edge image set to obtain the effective edge mask image set to be unioned;

[0125] In this embodiment, according to the obtained pixel - point binary data, a union operation is performed on the effective edge mask image set of the RFID chip. Select the images with a higher effective edge pixel density and stack these images layer by layer. To ensure the accuracy of the data, a threshold can be set (for example, only the images with the number of effective edge pixels greater than 3000 can be included in the union calculation). Finally, a new effective edge mask image set to be unioned is formed, which is convenient for subsequent geometric tolerance setting. For example, if there are three mask images and the effective edge pixels of two of them exceed the threshold, their union will retain all white pixels while removing the black pixels.

[0126] Step S33: Set geometric tolerances based on the binary data of the pixel points in the set of valid edge images, so as to obtain union geometric tolerance data;

[0127] In this embodiment, geometric tolerances are set according to the binary data of the pixel points in the set of valid edge images. Geometric tolerances mainly refer to the allowable error range during manufacturing and inspection. Parameter settings can be made based on the shape characteristics (such as width, thickness, etc.) of the valid edges. For example, the maximum tolerance is set to 2 pixels and the minimum tolerance is set to 1 pixel. Using these tolerance data, a geometric tolerance data set can be created to provide a reference for subsequent mask image processing. For example, if the actual width of a certain valid edge is 10 pixels, after geometric tolerance setting, the valid edge will allow an error between 8 and 12 pixels.

[0128] Step S34: Perform valid mask image union processing on the set of valid edge mask images to be unioned according to the union geometric tolerance data, so as to obtain the RFID chip mask union image;

[0129] In this embodiment, the obtained geometric tolerance data is used to process the set of valid edge mask images to be unioned. According to the geometric tolerance setting, it is judged whether the geometric shape of the valid edge in each mask image meets the requirements. If it meets the requirements, the valid edges of the mask images are merged to form a union image of the RFID chip mask. For example, if there are multiple mask images and the valid edge of a certain mask image is within the geometric tolerance range, its valid edge will be superimposed on the union image, thereby enhancing the overall characteristics of the image.

[0130] Step S35: Integrate the common edge features according to the RFID chip mask union image, so as to obtain a pre-generated variance map.

[0131] In this embodiment, the common edge features are integrated according to the RFID chip mask union image. This process includes performing edge detection (such as the Canny edge detection algorithm) on the union image to extract the common edge features. According to the shape and distribution of the common edges, a pre-generated variance map is generated for subsequent analysis and optimization. If it is found that a certain specific edge feature appears frequently in multiple images, then this feature can be focused on for further quality control and optimization. For example, if in the pre-generated variance map, the variance value of a certain edge is lower than the set threshold, it indicates that this edge feature is relatively stable during the production process and is worth retaining.

[0132] Optionally, step S35 is specifically:

[0133] Step S351: Extract the edge structure features of the RFID chip mask union image, so as to obtain image edge structure data;

[0134] In this embodiment, the mask image of the RFID chip is processed. The Canny edge detection algorithm can be used to extract the edge structure in the image. By setting appropriate high and low thresholds, the Canny algorithm can effectively identify obvious edges and form an edge image. Then, morphological processing, such as erosion and dilation, is performed on the extracted edge image to eliminate noise and enhance edge features. Finally, the generated edge structure data will contain the coordinate information and intensity values of the edges, facilitating subsequent similarity calculation.

[0135] Step S352: Calculate the similarity based on the image edge structure data to obtain edge structure similarity data;

[0136] In this embodiment, after completing the extraction of edge structure features, the next task is to calculate the similarity. The Euclidean distance or cosine similarity can be used to evaluate the edge structure similarity between different images. For example, select the edge structure data of two images, convert it into vector form, and then calculate their Euclidean distance. If the distance is less than the set threshold, it is considered that the two images are similar in edge structure. The results of the similarity calculation will be saved in the form of a matrix, where each element represents the similarity value between two images.

[0137] Step S353: Perform edge structure clustering calculation on the image edge structure data based on the edge structure similarity data to obtain edge structure clustering data, and perform common edge feature integration based on the edge structure clustering data to obtain common edge feature data;

[0138] In this embodiment, after obtaining the edge structure similarity data, a clustering algorithm is used to analyze it. For example, the K-means clustering algorithm can be applied to cluster the edge similarity matrix. Set the number of clusters K, and the appropriate K value can be selected according to the elbow method. After clustering, edge structure clustering data will be obtained, including the center points of each cluster and information about the member images. Subsequently, based on these clustering results, common edge features can be integrated to identify edge features that appear repeatedly in multiple images. This process involves feature selection techniques, such as using principal component analysis (PCA) to reduce the feature dimension and enhance the calculation efficiency.

[0139] Step S354: Calculate the edge feature variance based on the common edge feature data to obtain feature variance data, and visualize the feature variance data to obtain a pre-generated variance map.

[0140] In this embodiment, feature variance calculation is performed based on the common edge feature data. By calculating the variance of each common edge feature, its stability in different images is analyzed. The numpy library can be used to calculate the variance and generate corresponding statistical data. Subsequently, visualization tools such as Matplotlib are used to display the feature variance data in the form of a heat map, forming a pre-generated variance map. This map can intuitively show which features have a high degree of variation in the image set, thereby providing a basis for subsequent feature selection or classification.

[0141] Optionally, step S4 is specifically as follows:

[0142] Step S41: Obtain the on-site environmental data of the RFID chip;

[0143] In this embodiment, appropriate environmental monitoring devices are selected, such as temperature and humidity sensors, light sensors, and electromagnetic sensors. These devices are placed at the shooting site where the RFID chip is used to collect environmental data in real time. The data collection frequency should be set to once every 5 minutes to ensure the timeliness of the data. During the data collection process, the data is transmitted to the central processing system through a wireless network. The data format can be selected as JSON format, which is convenient for subsequent processing and analysis. In addition, a data quality monitoring mechanism should be established to ensure the accuracy and reliability of the data. For example, thresholds are set to filter out outliers.

[0144] Step S42: Extract the coil distribution features of the RFID chip from the on-site environmental data of the RFID chip, thereby obtaining the RFID chip coil distribution data;

[0145] In this embodiment, machine learning algorithms are used to analyze the collected environmental data. First, the environmental data is preprocessed, including normalization and denoising. Then, a clustering algorithm (such as K-means) is applied to identify the coil distribution features of the RFID chip. By mapping the coil distribution onto a two-dimensional plane, a coil distribution map can be generated using a visualization tool to show the relative positions and densities between the coils. Finally, the coil distribution data is output, including the center coordinates, radii, and relative density values of each coil, which will provide the necessary basis for subsequent steps.

[0146] Step S43: Evaluate the regional coil density based on the RFID chip coil distribution data, thereby obtaining the regional coil density data;

[0147] In this embodiment, the coil distribution data is divided into different regions. A fixed region size (e.g., 10m x 10m) can be set, and the number and distribution of RFID chip coils within each region are counted. The coil density of the region is calculated using the formula: density equals the area of the region divided by the number of coils within the region. Then, the results are stored in a database, and a coil density distribution map is generated to visualize the coil distribution of the region.

[0148] Step S44: Perform spatial association on the regional coil density data and the RFID chip edge feature image set to obtain coil density-edge feature association data;

[0149] In this embodiment, the edge features are combined with the coil density data, and a spatial statistical method (such as spatial regression analysis) is used to evaluate the relationship between the coil density and the edge features. By calculating the correlation coefficient, a coil density-edge feature association map is generated. In implementation, the SciPy library in Python can be used for data analysis, and the Matplotlib library can be used for data visualization to facilitate observing the relationship between the two.

[0150] Step S45: Estimate the sensitivity coefficient based on the coil density-edge feature association data to obtain the edge sensitivity coefficient.

[0151] In this embodiment, based on the coil density-edge feature association data, a regression analysis method (such as linear regression or multiple regression) is used to estimate the edge sensitivity coefficient. First, a mathematical model is established, incorporating the coil density, edge features, and other variables that may affect the sensitivity into the model. The training data is used for model fitting to obtain the regression coefficients, which will be used as the estimated values of the sensitivity coefficient. The R language or the statsmodels library in Python can be used for model construction and evaluation, and the cross-validation method can be used to improve the accuracy of the model. Finally, the numerical value of the sensitivity coefficient and its corresponding statistical significance are output to assist subsequent analysis and decision-making.

[0152] Optionally, step S42 is specifically:

[0153] Step S421: Extract the RFID chip signal environment features and physical environment features based on the RFID chip on-site environment data to obtain the RFID chip signal environment data and the physical environment data;

[0154] In this embodiment, the collected RFID signal data includes signal strength, noise level, transmission frequency, etc. Then, physical environment data is obtained through the data collected by environmental sensors (such as temperature, humidity, light, and object occlusion). Signal feature extraction uses time-domain and frequency-domain analysis methods, including Fourier transform, etc., to obtain the spectral features of the signal. At the same time, through statistical analysis, the influence of different environmental factors on signal strength is identified. Finally, a comprehensive data set of RFID chip signal environment data and physical environment data is formed.

[0155] Step S422: Conduct a statistical analysis of the signal strength distribution based on the RFID chip signal environment data, so as to obtain the RFID chip signal strength distribution data;

[0156] In this embodiment, based on the collected signal strength data, a statistical analysis of the signal strength distribution is carried out. Methods such as histograms or distribution maps can be used to divide the signal strength into multiple intervals and calculate the signal strength frequency within each interval. By using a clustering algorithm (such as K-means clustering) to identify the concentrated areas of signal strength and label them as "strong signal" or "weak signal" areas. These statistical results are presented in a visual way, for example, using a heat map to represent the signal strength distribution at different positions. This process helps to identify the distribution characteristics of RFID signals in different physical environments.

[0157] Step S423: Perform temporal correlation on the physical environment data and the RFID chip signal environment data, so as to obtain physical environment-signal environment correlation data, and estimate the physical environment impact factors based on the physical environment-signal environment correlation data, so as to obtain the physical environment impact factors;

[0158] In this embodiment, temporal correlation analysis is performed on the previously collected physical environment data (such as temperature changes, humidity fluctuations, etc.) and RFID signal data (such as signal strength changes). Time series analysis techniques are used to compare the relationship between signal changes and physical environment changes. For example, the correlation coefficient calculation method is used to analyze the correlation between physical environment changes and signal strength changes within different time periods. Then, based on the analysis results, the main physical environment factors affecting signal strength are extracted. For example, under high humidity conditions, the signal strength may decrease, and finally, physical environment-signal environment correlation data is formed.

[0159] Step S424: Correct the signal strength distribution data of the RFID chip based on the physical environment impact factors, so as to obtain the actual RFID chip signal strength distribution data;

[0160] In this embodiment, the estimated physical environment impact factor is used to correct the signal strength distribution data of the RFID chip. For example, if it is found that the high humidity condition in a certain area causes the signal strength to decrease by 10%, this compensation factor is added to the actual RFID signal strength distribution data. A linear regression model or a polynomial fitting method is used to achieve this correction, ensuring that the corrected signal strength data is more in line with the actual situation. Finally, the generated actual RFID chip signal strength distribution data can more accurately reflect the signal condition under the influence of the environment.

[0161] Step S425: Estimate the coil distribution of the RFID chip based on the actual RFID chip signal strength distribution data, so as to obtain the RFID chip coil distribution data.

[0162] In this embodiment, based on the obtained actual RFID chip signal strength distribution data, the distribution of the RFID chip coil is estimated. Through spatial interpolation techniques (such as Kriging interpolation or inverse distance weighting method), the signal strength of the unmeasured area is estimated according to the signal strength data at the known positions. Based on the corrected actual RFID signal strength distribution data, an inversion algorithm (such as a genetic algorithm or particle swarm optimization) is used to estimate the distribution of the RFID chip coil. By analyzing the signal strength changes in different areas, the position and quantity of the coils are inferred.

[0163] Optionally, step S5 is specifically:

[0164] Step S51: Perform convolution filtering and smoothing on the pre-generated variance map according to the edge sensitivity coefficient, so as to obtain the chip edge variance map;

[0165] In this embodiment, the pre-generated variance map is convolved according to the edge sensitivity coefficient (ESC) to eliminate noise and enhance the clarity of the chip edge. The pre-generated variance map is convolved with the edge sensitivity coefficient, and a Gaussian filter is used for smoothing. By selecting an appropriate convolution kernel (such as a 3×3 or 5×5 Gaussian kernel) for convolution calculation, the noise is reduced and the edge detection accuracy of the image is improved. After the processing is completed, the generated chip edge variance map can effectively highlight the edge features and provide a high-quality data basis for the subsequent steps.

[0166] Step S52: Extract statistical features according to the chip edge variance map, so as to obtain the edge variance map statistical feature data;

[0167] In this embodiment, statistical feature extraction is performed on the obtained variance map of the chip edge, specifically including mean, variance, maximum value, minimum value, standard deviation, etc. These features can reflect the overall characteristics and distribution of the chip edge. The NumPy library in Python can be used to quickly calculate these statistical features. For example, np.mean() is used to calculate the mean, and np.var() is used to calculate the variance. At the same time, the edge intensity distribution can be visualized by plotting a histogram for analysis. The extracted statistical feature data provides an important numerical basis for subsequent anomaly detection.

[0168] Step S53: Perform edge anomaly recognition on the variance map of the chip edge according to the statistical feature data of the variance map of the chip edge, so as to obtain abnormal image edge data and normal image edge data;

[0169] In this embodiment, based on the extracted statistical feature data, a threshold method or a machine learning-based method is used to identify abnormal edges. An adaptive threshold can be set, and when the edge variance exceeds this threshold, it is judged as abnormal. In terms of implementation, first, the k-means clustering algorithm can be used to divide the edge feature data into two categories: normal and abnormal, and then a classifier (such as a random forest, support vector machine, etc.) is used to train and test the features. Finally, the identified abnormal edge data and normal edge data are stored separately for subsequent analysis and processing.

[0170] Step S54: Construct an RFID chip image anomaly detection model according to the abnormal image edge data and the normal image edge data;

[0171] In this embodiment, based on the collected normal and abnormal image edge data, a suitable algorithm (such as a convolutional neural network in deep learning) is selected to construct an anomaly detection model. The TensorFlow or PyTorch framework can be used to design the network structure, including convolutional layers, pooling layers, and fully connected layers. The training data of the model should include a large number of images labeled with normal and abnormal edges to improve the classification accuracy of the model. Through cross-validation and hyperparameter tuning, the performance of the model is gradually optimized to make it have good detection capabilities.

[0172] Step S55: Detect the abnormal area of the sample image in the real-time RFID chip sampling image set through the RFID chip image anomaly detection model, so as to obtain the RFID chip abnormal area image set.

[0173] In this embodiment, the anomaly detection model that has been constructed is used to process the real-time collected RFID chip image set. By performing edge detection and feature extraction on each frame of the image and inputting it into the trained model, the detection of the anomaly region is carried out. Image stream processing techniques (such as OpenCV) can be used to quickly process the real-time image. The anomaly region information output by the model will be marked on the image, and an anomaly region image set will be generated to facilitate subsequent further analysis and verification. Combined with the real-time monitoring system, potential anomalies can be responded to and processed in a timely manner, improving the overall reliability and security of the RFID chip.

[0174] Optionally, this specification also provides an RFID chip image anomaly detection system based on intelligent learning for performing the RFID chip image anomaly detection method based on intelligent learning as described above. The RFID chip image anomaly detection system based on intelligent learning includes:

[0175] An edge finding module, configured to obtain a real-time RFID chip sampling image set; register a feature template for the real-time RFID chip sampling image and perform image alignment based on the feature template to obtain an aligned image set, and calculate an average grayscale image for the aligned image set to obtain an average grayscale image; perform RFID chip edge finding based on the aligned image set to obtain an RFID chip edge mask image set;

[0176] A threshold segmentation module, configured to perform edge mask image grayscale enhancement on the RFID chip edge mask image set to obtain an edge mask grayscale enhanced image set, and perform self-adaptive threshold segmentation on the edge mask grayscale enhanced image set to obtain an RFID chip effective edge mask image set;

[0177] A common edge feature integration module, configured to perform an effective mask graph union process on the RFID chip effective edge mask image set to obtain an RFID chip mask union image, and perform common edge feature integration based on the RFID chip mask union image to obtain a pre-generated variance graph;

[0178] An edge sensitivity coefficient estimation module, configured to obtain RFID chip on-site environment data, and perform coil density evaluation based on the RFID chip on-site environment data to obtain on-site environment coil density data; perform edge sensitivity coefficient estimation based on the on-site environment coil density data to obtain an edge sensitivity coefficient;

[0179] An anomaly detection model construction module is used to perform convolutional filtering and smoothing on a pre-generated variance map according to an edge sensitivity coefficient to obtain a chip edge variance map, and construct an RFID chip image anomaly detection model based on the chip edge variance map and an average grayscale map; the RFID chip image anomaly detection model is used to detect anomaly regions of sample images in a real-time RFID chip sampled image set, so as to obtain an RFID chip anomaly region image set.

[0180] The RFID chip image anomaly detection system based on intelligent learning of the present invention can implement any one of the RFID chip image anomaly detection methods of the present invention, and is a medium for coordinating operations and signal transmission between various modules to complete the RFID chip image anomaly detection method based on intelligent learning. The internal modules of the system cooperate with each other, thereby improving the accuracy and stability of detection.

[0181] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be encompassed by the present invention.

[0182] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein 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 these embodiments shown herein, but rather will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for detecting anomalies in RFID chip images based on intelligent learning, characterized in that: The following steps are involved: Step S1: Acquire a real-time RFID chip sampling image set; Registering feature templates for real-time RFID chip sampling images and performing image alignment based on the feature templates to obtain an aligned image set, and calculating an average grayscale image of the aligned image set to obtain an average grayscale image; Find the edge of the RFID chip according to the aligned image set, so as to obtain the RFID chip edge mask image set; Step S2: performing edge mask image grayscale enhancement on the RFID chip edge mask image set to obtain an edge mask grayscale enhanced image set, and performing self-adaptive threshold segmentation on the edge mask grayscale enhanced image set to obtain an RFID chip effective edge mask image set; Step S3: performing effective mask map union processing on the RFID chip effective edge mask image set to obtain the RFID chip mask union image, and integrating common edge features according to the RFID chip mask union image to obtain a pre-generated variance map; Step S4: Acquire the field environment data of the RFID chip, and perform coil density evaluation based on the field environment data of the RFID chip, thereby obtaining the field environment coil density data; perform edge sensitivity coefficient estimation based on the field environment coil density data, thereby obtaining the edge sensitivity coefficient; Step S4 is specifically as follows: Step S41: Acquire the on-site environment data of the RFID chip; Step S42: extracting RFID chip coil distribution features from the RFID chip field environment data, thereby obtaining RFID chip coil distribution data; Step S43: performing regional coil density evaluation according to the RFID chip coil distribution data, thereby obtaining regional coil density data; Step S44: spatially correlating the regional coil density data and the RFID chip edge feature image set, thereby obtaining coil density-edge feature correlation data; Step S45: estimating the sensitivity coefficient according to the coil density-edge feature correlation data, thereby obtaining the edge sensitivity coefficient; Step S5: Perform convolution filtering and smoothing processing on the pre-generated variance map according to the edge sensitivity coefficient to obtain the chip edge variance map, and build an RFID chip image anomaly detection model based on the chip edge variance map and the average grayscale map; perform sample image abnormal area detection on the real-time RFID chip sampling image set through the RFID chip image anomaly detection model to obtain the RFID chip abnormal area image set; Step S5 is specifically as follows: Step S51: performing convolution filtering and smoothing processing on the pre-generated variance map according to the edge sensitivity coefficient, so as to obtain a chip edge variance map; Step S52: extracting statistical features according to the chip edge variance map, thereby obtaining statistical feature data of the edge variance map; Step S53: performing edge anomaly recognition on the chip edge variance map according to the edge variance map statistical feature data, thereby obtaining abnormal image edge data and normal image edge data; Step S54: constructing an RFID chip image anomaly detection model according to the abnormal image edge data and the normal image edge data; Step S55: using the RFID chip image anomaly detection model to perform sample image abnormality region detection on the real-time RFID chip sampling image set, thereby obtaining an RFID chip abnormal region image set.

2. The RFID chip image anomaly detection method based on intelligent learning according to claim 1 is characterized in that: Step S1 is specifically as follows: Step S11: Acquire a real-time RFID chip sampling image set; Step S12: performing sample image set preprocessing on the real-time RFID chip sample image set, thereby obtaining the RFID chip sample image set; Step S13: randomly selecting pictures from the real-time RFID chip sampling image set to obtain a random sampling image set, and performing feature registration based on the random sampling image set to obtain a feature template; Step S14: performing residual image alignment on the real-time RFID chip sampling image set according to the feature template to obtain an aligned image set, and performing average grayscale image calculation on the aligned image set to obtain an average grayscale image; Step S15: performing edge detection on the aligned image set to obtain an RFID chip edge feature image set; Step S16: performing binary edge mask image conversion according to the RFID chip edge feature image set, thereby obtaining the RFID chip edge mask image set.

3. The RFID chip image anomaly detection method based on intelligent learning according to claim 1 is characterized in that: Step S2 is specifically as follows: Step S21: performing pixel binary statistics according to the RFID chip edge mask image set, thereby obtaining a high gradient change area image set and a low gradient change area image set; Step S22: performing local adaptive histogram equalization on the high gradient change area image set, thereby obtaining a high gradient change area grayscale enhanced image set; Performing edge contrast enhancement on the low gradient change region image set, thereby obtaining a low gradient change region enhanced image set; Step S23: merging the high gradient change area grayscale enhanced image set and the low gradient change area enhanced image set to obtain an edge mask grayscale enhanced image set; Step S24: performing adaptive threshold segmentation of functional regions according to the edge mask grayscale enhanced image set, thereby obtaining an effective edge mask image set of the RFID chip.

4. The RFID chip image anomaly detection method based on intelligent learning according to claim 3 is characterized in that: Step S24 is specifically as follows: Step S241: performing local threshold calculation on the edge mask grayscale enhanced image set in the gradient change region, thereby obtaining local threshold data of the high gradient change region and local threshold data of the low gradient change region; Step S242: performing functional region threshold segmentation on the edge mask grayscale enhanced image set according to the local threshold data of the high gradient change region, thereby obtaining a functional region threshold segmentation image set; performing boundary region threshold segmentation on the edge mask grayscale enhanced image set according to the local threshold data of the low gradient change region, thereby obtaining a boundary region threshold segmentation image set; Step S243: performing an intersection operation on the boundary region threshold segmentation image set and the function region threshold segmentation image set, thereby obtaining a background entanglement function region image set; Step S244: performing logical pixel merging on the background entanglement functional area image set and the functional area threshold segmentation image set, thereby obtaining an RFID chip effective edge mask image set.

5. The RFID chip image anomaly detection method based on intelligent learning according to claim 1 is characterized in that: Step S3 is specifically as follows: Step S31: performing binary statistics of pixels of the effective edge mask image set of the RFID chip, thereby obtaining binary data of pixels of the effective edge image set; Step S32: performing a merging layer selection on the RFID chip effective edge mask image set according to the binary data of the pixel points of the effective edge image set, thereby obtaining the effective edge mask image set to be merged; Step S33: setting geometric tolerance according to the binary data of the pixels in the valid edge image set, thereby obtaining the geometric tolerance data of the union set; Step S34: performing effective mask image union processing on the effective edge mask image set to be united according to the union geometric tolerance data, thereby obtaining an RFID chip mask union image; Step S35: integrating common edge features according to the RFID chip mask and image union to obtain a pre-generated variance map.

6. The RFID chip image anomaly detection method based on intelligent learning according to claim 5 is characterized in that: Step S35 is specifically as follows: Step S351: extract edge structure features from the RFID chip mask union image to obtain image edge structure data; Step S352: performing similarity calculation based on the image edge structure data, thereby obtaining edge structure similarity data; Step S353: performing edge structure clustering calculation on the image edge structure data based on the edge structure similarity data, thereby obtaining edge structure clustering data, and performing common edge feature integration according to the edge structure clustering data, thereby obtaining common edge feature data; Step S354: Calculate the edge feature variance according to the common edge feature data to obtain feature variance data, and visualize the feature variance data to obtain a pre-generated variance map.

7. The RFID chip image anomaly detection method based on intelligent learning according to claim 1 is characterized in that: Step S42 is specifically as follows: Step S421: extracting RFID chip signal environment characteristics and physical environment characteristics according to the RFID chip field environment data, thereby obtaining RFID chip signal environment data and physical environment data; Step S422: performing signal strength distribution statistics according to the RFID chip signal environment data, thereby obtaining the RFID chip signal strength distribution data; Step S423: performing time-series correlation on the physical environment data and the RFID chip signal environment data to obtain physical environment-signal environment correlation data, and estimating the physical environment impact factor according to the physical environment-signal environment correlation data to obtain the physical environment impact factor; Step S424: performing signal strength distribution correction on the RFID chip signal strength distribution data according to the physical environment influencing factors, thereby obtaining actual RFID chip signal strength distribution data; Step S425: Estimating the RFID chip coil distribution according to the actual RFID chip signal strength distribution data, thereby obtaining the RFID chip coil distribution data.

8. An RFID chip image anomaly detection system based on intelligent learning, characterized in that: The method for detecting anomalies in RFID chip images based on intelligent learning according to claim 1 comprises: The edge search module is used to obtain a real-time RFID chip sampling image set; register a feature template for the real-time RFID chip sampling image and perform image alignment based on the feature template to obtain an aligned image set, and calculate an average grayscale image of the aligned image set to obtain an average grayscale image; perform RFID chip edge search based on the aligned image set to obtain an RFID chip edge mask image set; A threshold segmentation module is used to perform edge mask image grayscale enhancement on the RFID chip edge mask image set, thereby obtaining an edge mask grayscale enhanced image set, and perform self-adaptive threshold segmentation on the edge mask grayscale enhanced image set, thereby obtaining an RFID chip effective edge mask image set; The common edge feature integration module is used to perform effective mask map union processing on the RFID chip effective edge mask image set, so as to obtain the RFID chip mask union image, and perform common edge feature integration according to the RFID chip mask union image, so as to obtain a pre-generated variance map; The edge sensitivity coefficient estimation module is used to obtain the field environment data of the RFID chip, and to perform coil density evaluation based on the field environment data of the RFID chip, thereby obtaining the field environment coil density data; and to perform edge sensitivity coefficient estimation based on the field environment coil density data, thereby obtaining the edge sensitivity coefficient; The anomaly detection model building module is used to perform convolution filtering and smoothing processing on the pre-generated variance map according to the edge sensitivity coefficient to obtain the chip edge variance map, and build an RFID chip image anomaly detection model based on the chip edge variance map and the average grayscale map; the RFID chip image anomaly detection model is used to perform sample image abnormal area detection on the real-time RFID chip sampling image set to obtain the RFID chip abnormal area image set.

Images