In-vitro diagnostic reagent strip packaging quality control method and system

Abstract

The invention relates to the field of reagent strip packaging quality detection, in particular to an in-vitro diagnostic reagent strip packaging quality control method and system. The method comprises the following steps: acquiring a packaged in-vitro diagnostic reagent monitoring image; performing multi-scale reconstruction on the in-vitro diagnostic reagent monitoring image so as to generate a multi-scale noise reduction reconstructed monitoring image; performing image visual segmentation and image iteration packaging defect detection on the multi-scale noise reduction reconstruction monitoring image one by one so as to obtain a packaging defect part expansion monitoring image; carrying out surface defect texture mining on the packaging defect part expansion monitoring image, and carrying out deep defect factor inference so as to generate packaging defect inference factors; obtaining a reagent strip packaging log; and defect process tracing is performed on the reagent strip packaging log according to the packaging defect inference factors, so that packaging process stages causing defects are marked. The packaging parameters are automatically adjusted through real-time feedback, so that the packaging quality of the in-vitro diagnostic reagent strip is accurately improved.

Description

Technical Field

[0001] The present invention relates to the field of reagent strip packaging quality inspection, and particularly to a method and system for quality control of in vitro diagnostic reagent strip packaging. Background Art

[0002] As an indispensable tool in clinical diagnosis, in vitro diagnostic reagent strips are widely used in the detection of samples such as blood and urine, and play an important role especially in the early screening and monitoring of diseases. The quality of reagent strip packaging directly affects its stability, sensitivity and accuracy. Therefore, it is crucial to ensure the stability of reagent strip packaging quality. However, with the continuous expansion of the production scale of in vitro diagnostic reagent strips, traditional packaging quality inspection methods are difficult to meet the requirements of efficient and accurate quality monitoring.

[0003] Currently, the packaging quality inspection of in vitro diagnostic reagent strips usually relies on manual inspection and mechanical detection equipment, such as visual inspection, pressure testing, etc. Although these methods can detect some obvious packaging problems, they often have deficiencies such as insufficient sensitivity, low detection efficiency, and strong operation dependence. Manual inspection is often affected by the experience and subjective judgment of operators, which easily leads to inaccurate or inconsistent detection results; while traditional mechanical detection equipment has limited ability to perceive subtle defects and abnormal situations in packaging, and it is difficult to meet the high-precision quality control requirements.

[0004] In addition, in vitro diagnostic reagent strips are affected by various factors such as temperature and humidity changes, raw material quality, and equipment accuracy during the production process, resulting in quality problems such as air bubbles, leakage, and uneven packaging during the packaging process, which in turn affect the use effect of the reagent strips. In order to improve the detection efficiency and quality, the limitations of traditional methods are becoming more and more obvious. Therefore, there is an urgent need for a more intelligent, precise and automated method for in vitro diagnostic reagent strip packaging quality inspection and optimization to meet the growing production needs and improve the consistency and quality stability of products. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention proposes a method and system for quality control of in vitro diagnostic reagent strip packaging to solve at least one of the above technical problems.

[0006] To achieve the above object, the present invention provides a method for quality control of in vitro diagnostic reagent strip packaging, including the following steps: Step S1: Obtain the monitoring image of the in vitro diagnostic reagent after packaging; perform multi-scale reconstruction on the monitoring image of the in vitro diagnostic reagent to generate a multi-scale noise-reduced reconstructed monitoring image; Step S2: Perform image visual segmentation and iterative defect detection of each image on the multi-scale noise-reduced reconstructed monitoring image to obtain an enlarged monitoring image of the packaging defect part; Step S3: Conduct surface defect texture mining on the enlarged monitoring image of the encapsulation defect site, and infer deep defect factors, thereby generating encapsulation defect inference factors; Step S4: Obtain the reagent strip encapsulation log; trace the defective process of the reagent strip encapsulation log according to the encapsulation defect inference factors, thereby marking the encapsulation process stage that causes the defect; Step S5: Calculate the encapsulation site error based on the enlarged monitoring image of the encapsulation defect site, and conduct error compensation calculation to generate encapsulation defect compensation parameters; Step S6: Fine-tune the process parameters of the encapsulation process stage that causes the defect according to the encapsulation defect compensation parameters, and conduct adaptive learning optimization to build an intelligent encapsulation process control model.

[0007] By performing multi-scale reconstruction on the monitoring images of in vitro diagnostic reagents, the present invention can effectively remove the noise in the images and retain the key feature information. Noise is often a common problem in image processing. Especially in the detection of minute defects, noise can affect the accurate identification of defects. Multi-scale reconstruction can also effectively identify defects of different sizes, provide image details at different scales, and enhance the accuracy of subsequent defect detection. Especially when differentiating between defects of different sizes and types, multi-scale reconstruction provides a more comprehensive monitoring perspective. Image visual segmentation can distinguish different regions in the image, clarifying which parts belong to the encapsulation defect region and which regions are normal. This step helps to accurately extract and locate the defect region and improve the detection accuracy. The method of iterative detection for each image helps to gradually approach the true situation of the defect, and through the iterative process, continuously optimize the detection accuracy and reduce the risk of missed detection or false detection. By expanding the monitoring image, the detection range of the defect region can be effectively expanded, enabling the defect to be further magnified for observation, thereby more precisely detecting the encapsulation defect. This is of great significance for improving the overall encapsulation quality control. Surface defect texture mining can identify the characteristic textures of surface defects and, by analyzing the texture changes, help to distinguish different types of defects. Defects with different textures are related to different encapsulation processes or material qualities, and texture mining can reveal this information. Deep defect factor inference helps to deeply analyze the root causes of defects and infer which links or factors in the encapsulation process the defects originate from, such as temperature changes, pressure unevenness, material defects, etc. Inferring these deep factors can provide key basis for subsequent process optimization. The generated encapsulation defect inference factors help to accurately locate the defect source, thereby providing a direction for process improvement. By obtaining the encapsulation log of the reagent strip and tracing each link in the encapsulation process, it can be identified which link has the problem. The tracing system can quickly find out the stage where the defect occurs, help the enterprise trace back to the specific operations and links in production, and then make targeted adjustments. This process tracing mechanism is the key to encapsulation quality control. When defects occur, it can quickly trace the problem source, avoid the recurrence of similar defects, and improve the reliability and consistency of the encapsulation process. Through error calculation, the errors existing in the encapsulation process can be accurately identified and the impact of these errors on the encapsulation quality of the reagent strip can be quantified. This precise error analysis helps to optimize the encapsulation process and reduce the quality problems caused by errors in production. Through error compensation calculation, compensation is carried out according to the actually monitored defect data, directly adjusting the process parameters in the encapsulation process. This compensation process will ensure that each batch of reagent strips can meet high quality standards and reduce the fluctuations in production. Process parameter fine-tuning can accurately adjust the process links that need to be improved in the encapsulation process according to the encapsulation defect compensation parameters calculated in the previous step, improve the stability and consistency of the process, and ensure that the same defects will not appear in subsequent production batches. Adaptive learning optimization enables the system to continuously learn and improve according to different production data.As data accumulates during the production process, the system becomes more intelligent and can flexibly adjust control strategies under different production conditions. The constructed intelligent encapsulation process control model not only improves the automation level of the encapsulation process but also can self-optimize the process based on real-time data, continuously improve the encapsulation quality, and reduce the need for manual intervention.

[0008] In this specification, an in vitro diagnostic reagent strip encapsulation quality control system is provided for implementing the method described above, including: A multi-scale reconstruction module for obtaining a monitoring image of the in vitro diagnostic reagent after encapsulation; performing multi-scale reconstruction on the monitoring image of the in vitro diagnostic reagent to generate a multi-scale noise-reduced reconstructed monitoring image; An encapsulation defect detection module for performing image visual segmentation and iterative encapsulation defect detection on each image of the multi-scale noise-reduced reconstructed monitoring image to obtain an enlarged monitoring image of the encapsulation defect location; A defect inference module for mining surface defect textures of the enlarged monitoring image of the encapsulation defect location and inferring deep defect factors to generate encapsulation defect inference factors; A process traceability module for obtaining the reagent strip encapsulation log; performing defect process traceability on the reagent strip encapsulation log based on the encapsulation defect inference factors to mark the encapsulation process stage that caused the defect; A defect compensation module for calculating the encapsulation location error based on the enlarged monitoring image of the encapsulation defect location and performing error compensation calculation to generate encapsulation defect compensation parameters; An encapsulation process optimization module for fine-tuning the process parameters of the encapsulation process stage that caused the defect based on the encapsulation defect compensation parameters and performing adaptive learning optimization to construct an intelligent encapsulation process control model.

[0009] The present invention performs noise reduction on the monitored images through multi-scale reconstruction technology, which can effectively remove background noise and enhance the image quality of the target area, providing higher-quality image data for subsequent defect detection. This makes the identification of defects more accurate and improves the accuracy of subsequent processing. Multi-scale reconstruction can process images at different resolutions and levels, helping to extract more detailed information, especially for packaging defects with different sizes or shapes, improving the comprehensiveness of detection. Image vision segmentation technology helps to divide the reagent strip monitored images into multiple meaningful regions, focusing on the regions with defects, and improving the accuracy and efficiency of defect detection. Iterative packaging defect detection for each image helps to systematically analyze each defect and avoid omission. By enlarging the monitored images, the defective parts can be presented more clearly, making subsequent analysis and processing more accurate. Especially in the discovery of small or hidden defects, the enhanced images provide stronger visual support. Through surface defect texture mining, the details of packaging surface defects can be deeply analyzed to identify different types of defect textures. And through in-depth defect factor inference, the causes of defects are deeply explored, providing a scientific basis for optimizing the packaging process. By inferring the deep factors of packaging defects, it helps to find potential and overlooked defect sources. This provides data support for subsequent process traceability and optimization, and helps to discover systematic problems existing in the packaging process. By combining the inferred factors of packaging defects with packaging logs, the defects are traced and tracked, and the process stage where the defects occur is accurately located. This provides a clear basis for subsequent process improvement and quality control, avoiding empirical guesses and uncertainties. Through traceability analysis, not only can specific defects be marked, but also the deficiencies in the process can be pointed out, ensuring that the causes of defects are clearly identified, thus avoiding recurrence. By calculating the errors of the packaging parts and compensating them, the packaging deviations caused by reasons such as process instability and equipment errors are effectively corrected. This process helps to reduce product quality fluctuations and improve the accuracy and consistency of packaging. The generation and application of error compensation parameters provide a basis for real-time adjustment in the packaging process, quickly correcting the deviations in the process, and ensuring that the quality of each batch of products is within the standard range. According to the defect compensation parameters, the packaging process is fine-tuned, enabling the process to be dynamically adjusted during actual production, and enhancing the adaptability and flexibility of the packaging process. The adaptive learning optimization mechanism helps the system automatically adjust the packaging parameters according to historical data and real-time feedback, enabling the packaging process to be automatically optimized under different production conditions. This intelligent control can effectively improve production efficiency and reduce human intervention. By constructing an intelligent packaging process control model, the packaging process can be continuously optimized in the long term, not only solving the current defect problems but also preventing future problems, providing guarantee for long-term quality control. Brief Description of the Drawings

[0010] Figure 1Schematic diagram of the step process of a method for quality control of in vitro diagnostic reagent strip packaging; Figure 2 Schematic diagram of the detailed implementation steps of step S1; Figure 3 Schematic diagram of the detailed implementation steps of step S2; Figure 4 Schematic diagram of the detailed implementation steps of step S3. Detailed implementation manner

[0011] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0012] The embodiments of the present application provide a method and system for quality control of in vitro diagnostic reagent strip packaging. The execution subjects of the method and system for quality control of in vitro diagnostic reagent strip packaging include, but are not limited to, the following that carry this system: mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc., which can be regarded as general computing nodes of the present application. The data processing platform includes, but is not limited to, at least one of: audio and image management systems, information management systems, and cloud data management systems.

[0013] Please refer to Figures 1 to 4 , the present invention provides a method for quality control of in vitro diagnostic reagent strip packaging, and the method for quality control of in vitro diagnostic reagent strip packaging includes the following steps: Step S1: Obtain the monitoring image of the in vitro diagnostic reagent after packaging; perform multi-scale reconstruction on the monitoring image of the in vitro diagnostic reagent to generate a multi-scale noise-reduced reconstructed monitoring image; Step S2: Perform image visual segmentation and iterative defect detection of each image on the multi-scale noise-reduced reconstructed monitoring image to obtain an enlarged monitoring image of the packaging defect part; Step S3: Mine the surface defect texture of the enlarged monitoring image of the packaging defect part and infer the deep defect factors to generate the inferred factors of the packaging defect; Step S4: Obtain the reagent strip packaging log; trace the defective process of the reagent strip packaging log according to the inferred factors of the packaging defect to mark the packaging process stage that causes the defect; Step S5: Calculate the error of the packaging part according to the enlarged monitoring image of the packaging defect part and perform error compensation calculation to generate packaging defect compensation parameters; Step S6: Fine-tune the process parameters of the packaging process stage that causes the defect according to the packaging defect compensation parameters and perform adaptive learning optimization to construct an intelligent packaging process control model.

[0014] By performing multi-scale reconstruction on the monitoring images of in vitro diagnostic reagents, the present invention can effectively remove the noise in the images and retain the key feature information. Noise is often a common problem in image processing. Especially in the detection of tiny defects, noise will affect the accurate identification of defects. Multi-scale reconstruction can also effectively identify defects of different sizes, provide image details at different scales, and enhance the accuracy of subsequent defect detection. Especially when distinguishing between defects of different sizes and types, multi-scale reconstruction provides a more comprehensive monitoring perspective. Image visual segmentation can distinguish different regions in the image, clarifying which parts belong to the encapsulation defect region and which regions are normal. This step helps to accurately extract and locate the defect region and improve the detection accuracy. The method of iterative detection for each image helps to gradually approach the true situation of the defect, and through the iterative process, continuously optimize the detection accuracy and reduce the risk of missed detection or false detection. By expanding the monitoring image, the detection range of the defect region can be effectively expanded, enabling the defect to be further magnified for observation, thereby more accurately detecting the encapsulation defect. This is of great significance for improving the overall encapsulation quality control. Surface defect texture mining can identify the characteristic textures of surface defects and, by analyzing the texture changes, help distinguish different types of defects. Defects with different textures are related to different encapsulation processes or material qualities, and texture mining can reveal this information. Deep defect factor inference helps to deeply analyze the root causes of defects and infer which links or factors in the encapsulation process the defects originate from, such as temperature changes, pressure unevenness, material defects, etc. Inferring these deep factors can provide key bases for subsequent process optimization. The generated encapsulation defect inference factors help to accurately locate the defect source, thus providing a direction for process improvement. By obtaining the encapsulation log of the reagent strip, tracing each link in the encapsulation process, and identifying which link has problems. The tracing system can quickly find out the stage where the defect occurs, help the enterprise trace back to the specific operations and links in production, and then make targeted adjustments. This process tracing mechanism is the key to encapsulation quality control. When defects occur, it can quickly trace back to the problem source, avoid the recurrence of similar defects, and improve the reliability and consistency of the encapsulation process. Through error calculation, the errors existing in the encapsulation process can be accurately identified and the impact of these errors on the encapsulation quality of the reagent strip can be quantified. This precise error analysis helps to optimize the encapsulation process and reduce the quality problems caused by errors in production. Through error compensation calculation, compensation is performed according to the actually monitored defect data, directly adjusting the process parameters in the encapsulation process. This compensation process will ensure that each batch of reagent strips can meet high-quality standards and reduce the fluctuations in production. Process parameter fine-tuning can accurately adjust the process links that need to be improved in the encapsulation process according to the encapsulation defect compensation parameters calculated in the previous step, improve the stability and consistency of the process, and ensure that the same defects will not occur in subsequent production batches. Adaptive learning optimization enables the system to continuously learn and improve according to different production data.As data accumulates during the production process, the system becomes more intelligent and can flexibly adjust control strategies under different production conditions. The constructed intelligent encapsulation process control model not only improves the automation level of the encapsulation process but also can self-optimize the process based on real-time data, continuously improve the encapsulation quality, and reduce the need for manual intervention.

[0015] In an embodiment of the present invention, refer to Figure 1 , which is a schematic flow chart of the steps of a method for controlling the encapsulation quality of an in vitro diagnostic reagent strip according to the present invention. In this example, the steps of the method for controlling the encapsulation quality of the in vitro diagnostic reagent strip include: Step S1: Obtain a monitoring image of the in vitro diagnostic reagent after encapsulation; perform multi-scale reconstruction on the monitoring image of the in vitro diagnostic reagent to generate a multi-scale noise-reduced reconstructed monitoring image; In this embodiment, a suitable image acquisition device is selected. Usually, a high-resolution camera or an industrial camera is used to ensure that the details of the in vitro diagnostic reagent can be captured. When selecting a camera, the number of pixels, imaging speed, and optical performance should be considered. It is recommended to use a camera with at least 1600x1200 pixels to obtain clear monitoring images. During the acquisition process, a suitable light source needs to be set to ensure uniform illumination of the image. LED light sources or fluorescent lamps are used to avoid shadows and reflections. Ensure that the color temperature of the light source is between 5000K and 6500K to maintain the true color of the image. During the acquisition process, ensure that the reagent strip is placed stably and flatly to avoid image blurring caused by shaking or tilting. Use a fixed bracket or fixture to ensure the stability of the reagent strip. Image acquisition should be carried out in a vibration-free environment to ensure image quality. Set the shutter speed and ISO value of the camera to adapt to different lighting conditions. It is recommended to use a lower ISO value (such as 100 or 200) to reduce image noise, and at the same time adjust the shutter speed to obtain an appropriate exposure time (for example, 1 / 60 second) to ensure that the image is neither overexposed nor underexposed. After the acquisition is completed, store the monitoring images in a specified database or file system. It is recommended to use a lossless image format (such as TIFF or PNG) to retain the image quality. Each image should be accompanied by relevant metadata, including the acquisition time, device parameters, reagent strip type, etc., for subsequent analysis and processing. Select a suitable multi-scale reconstruction method. Commonly used methods include wavelet transform and Laplacian pyramid. Wavelet transform is especially suitable for multi-level analysis of images and can effectively capture details at different frequencies. The Laplacian pyramid can handle different scales of images and is suitable for hierarchical reconstruction of images. Before performing multi-scale reconstruction, first preprocess the monitoring images. This includes denoising the images and enhancing the contrast. Use a Gaussian filter to smooth the images, set the convolution kernel size to 5x5 to effectively remove noise while retaining edge information. Implement the wavelet transform or Laplacian pyramid method to decompose the monitoring images into sub-images at multiple scales. In wavelet transform, usually select the Daubechies wavelet as the basis function. Set the decomposition level to 3 layers to capture image features at different frequencies. In the Laplacian pyramid, generate the high-frequency and low-frequency images of each layer by gradually reducing the resolution of the image. Record the decomposition results of each layer for subsequent reconstruction and analysis. Reconstruct the sub-images at multiple scales into a denoised monitoring image through inverse transformation. Confirm the clarity and contrast of the reconstructed image to ensure that the detailed parts are effectively retained. Evaluate the reconstruction effect by calculating indicators such as PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index) to ensure that the PSNR value is higher than 30dB and the SSIM value is close to 1.

[0016] Step S2: Perform image visual segmentation and defect detection for each image iteration packaging on the multi-scale noise-reduced reconstructed monitoring image, so as to obtain an enlarged monitoring image of the packaging defect part; In this embodiment, before performing image vision segmentation, selecting a suitable segmentation algorithm is the key. Common segmentation methods include threshold segmentation, region growing, K-means clustering, and deep learning segmentation networks (such as U-Net or Mask R-CNN). Considering the complexity and variability of packaging defects, it is recommended to use the U-Net model for segmentation, which performs well in medical images and industrial inspections and can accurately detect different defect regions. Before implementing the segmentation, corresponding training datasets need to be prepared. These datasets should contain annotated images that clearly indicate the regions of each defect. Use existing packaging defect images for data augmentation, such as rotation, flipping, and brightness adjustment, to increase the diversity of training samples. Set training parameters, such as the learning rate (0.001), batch size (16), and number of training epochs (50 epochs), to ensure that the model can effectively learn defect features. Train the U-Net, optimize it using the cross-entropy loss function, monitor the change in the loss value during training, and ensure that the model converges stably. Verify the performance of the model on the test set to ensure that the segmentation accuracy reaches over 85%. Use the trained U-Net model to segment the multi-scale denoising and reconstruction monitoring images. After inputting the image, the model will output the segmentation result, generating a binary image, where the defect regions are marked as 1 and the background is 0. Set a suitable threshold (such as 0.5) to determine whether each pixel belongs to the defect region. Post-process the segmented image to eliminate small noises and artifacts. For example, use morphological operations (such as opening and closing operations) to smooth the boundaries and fill small holes, enhancing the accuracy of segmentation. Record the quality of the segmentation result to ensure that the segmented image has high clarity and accuracy. Once the segmented image is obtained, perform packaging defect detection on each image. Select a suitable detection algorithm, commonly including feature-based detection (such as Haar features) or deep learning object detection models (such as YOLO or Faster R-CNN). Considering the real-time and accuracy of detection, it is recommended to use the YOLO model. Perform defect detection on each segmented image and record the detected defect positions and types. Set the parameters of the detection model, such as the confidence threshold (such as 0.5) and NMS (non-maximum suppression) threshold, to reduce duplicate detections. Through multiple iterations, optimize the detection results to ensure that the detection accuracy can be gradually improved in each detection cycle. According to the detection results, expand each defect region to generate an enlarged monitoring image of the packaging defect site. Set the expansion ratio (such as 2 times or 3 times) to ensure that the details of the defect can be clearly shown. Use bilinear interpolation to expand the image to maintain the image quality. After completing the enlarged monitoring image, record the specific positions, types, and enlarged images of each defect for subsequent analysis and processing. Ensure that the finally generated enlarged monitoring image has sufficient resolution and clarity to support subsequent defect analysis and quality control.

[0017] Step S3: Conduct surface defect texture mining on the enlarged monitoring image of the encapsulation defect site, and perform deep defect factor inference to generate encapsulation defect inference factors. In this embodiment, before conducting surface defect texture mining, selecting an appropriate texture analysis method is crucial. Commonly used texture extraction methods include Gray Level Co-occurrence Matrix (GLCM), Local Binary Pattern (LBP), and Gabor filter. Considering the complexity and diversity of texture features, the GLCM method is recommended because it can effectively capture the texture information in the image and provide multiple statistical features. Before texture extraction, preprocess the enlarged monitoring image. This includes denoising, enhancing contrast, and normalizing the image size. Use a Gaussian filter for denoising, set the convolution kernel size to 5x5 to reduce the noise in the image. Contrast enhancement is achieved through histogram equalization to ensure that texture features are more obvious in the image. Use the GLCM method to extract the texture features of the image. First, convert the image to grayscale, and calculate the gray level co-occurrence matrices in multiple directions (0°, 45°, 90°, 135°) and distances (usually set to 1 pixel). According to the generated GLCM, calculate multiple statistical features, including contrast, correlation, energy, and entropy, etc. Set the extraction parameters, such as calculating the values of each texture feature and recording the feature vectors of each defect area. Ensure that the extracted feature values can effectively reflect the nature of the defects. Usually, the correlation between feature values is required to be less than 0.5 to avoid redundant information. After completing the texture feature extraction, select an appropriate deep defect factor inference method. Commonly used inference methods include causal inference, logistic regression analysis, and Bayesian network. Considering the complex relationships between multiple factors, the Bayesian network is recommended. This method can handle uncertainty and provide intuitive causal relationships. Integrate the extracted texture features with other relevant data (such as process parameters, environmental conditions, etc.) to construct an inference model. Use the texture features of each defect area and the corresponding process parameters as the input of the model, and set the corresponding prior probabilities. Ensure the diversity and integrity of the dataset to improve the accuracy of the inference model. Use the constructed Bayesian network to infer the deep defect factors. By inputting the known texture features and process parameters, calculate the causes of the defects. Analyze the posterior probabilities output by the model to identify the factors that have the greatest impact on the defects. Usually, factors with a posterior probability higher than 0.7 are considered important inference factors. According to the results of the inference analysis, generate encapsulation defect inference factors. Record the identified important factors, including their impact degree and correlation with the defects. Ensure that each factor has a clear description and explanation for subsequent quality improvement and process optimization.

[0018] Step S4: Obtain the reagent strip encapsulation log; perform defect process traceability on the reagent strip encapsulation log according to the encapsulation defect inference factors to mark the encapsulation process stage that causes the defects. In this embodiment, first, it is necessary to obtain the encapsulation logs of the reagent strips. These logs are usually automatically generated by the encapsulation equipment and record every link in the production process, including timestamps, operators, equipment status, process parameters, etc. To ensure the integrity and accuracy of the log data, it is necessary to interface with the control system of the production line and set up an automated data acquisition system. During the data acquisition process, it is necessary to ensure that the obtained log information includes the following key parameters: Timestamp: Records the start and end times of each process stage.

[0019] Temperature and pressure: During the encapsulation process, temperature and pressure are important parameters affecting quality.

[0020] Operator: Records operator information for responsibility tracing.

[0021] Equipment status: Includes equipment operating status, fault information, etc., for analyzing the impact of the equipment on defects. All obtained log data should be stored in a database in a structured format, such as CSV or SQL, for subsequent data analysis and query. Integrate the previously generated encapsulation defect inference factors with the encapsulation logs. The inference factors usually include key process parameters that cause defects, such as abnormal temperature, pressure fluctuations, improper operations, etc. Set the impact thresholds for these factors, for example: Temperature: Set the allowable temperature range as (20°C, 25°C). If the recorded temperature is outside this range, it is regarded as a potential defect cause.

[0022] Pressure: Set the pressure fluctuation range as (1.5 bar, 2.0 bar).

[0023] Match these inference factors with the relevant parameters in the encapsulation logs to identify the process stages related to defects. Select a suitable traceability analysis method. Adopt the causal relationship analysis or root cause analysis method. Causal relationship analysis helps to identify the direct factors leading to defects, while root cause analysis can dig deeper into the root causes of problems. It is recommended to use the fishbone diagram (cause-and-effect diagram) as an analysis tool, which can effectively visualize the problem and its causes. When conducting defect process tracing, first conduct a correlation analysis between the encapsulation logs and the inference factors. Screen out the records related to defects in the logs and mark information such as timestamps, operators, and equipment status. For each defect, extract the corresponding process parameters from the encapsulation logs according to its occurrence timestamp. Through the fishbone diagram method, classify the potential causes of each defect and analyze the impacts of different process stages. For example, classify the resulting defects into "equipment factors", "operation factors", and "environmental factors" and mark them one by one in the diagram. According to the analysis results, determine the encapsulation process stage that causes defects. Record the specific process stages of each defect, including: Temperature control stage: Such as temperature exceeding the standard or being unstable.

[0024] Pressure control stage: If the pressure is abnormal, it will affect the encapsulation effect.

[0025] Operation stage: Defects caused by improper operation of the operator.

[0026] Ensure that the marked process stage can correspond to the factors for inferring encapsulation defects, providing a basis for subsequent improvement measures.

[0027] Step S5: Expand the monitoring image according to the encapsulation defect location to calculate the encapsulation location error and perform error compensation calculation to generate encapsulation defect compensation parameters; In this embodiment, a suitable error calculation method is selected to accurately measure the geometric parameters of the encapsulation defect location. Commonly used error calculation methods include absolute error and relative error analysis. Absolute error represents the difference between the actual measured value and the ideal value, while relative error is the ratio of the absolute error to the ideal value, usually expressed as a percentage. During the implementation process, the ideal geometric dimensions of the encapsulation location need to be clarified, such as width, length, and height. These ideal values are set as the reference for subsequent error calculation. Geometric feature extraction is performed on the expanded monitoring image of the encapsulation defect location. Edge detection algorithms (such as the Canny algorithm) are used to identify the edges of the defect area, and then the specific shape and area of the defect are obtained through contour extraction algorithms (such as findContours in OpenCV). Parameters are set, such as the high and low thresholds for edge detection (such as 50 and 150), to ensure that the contours of the encapsulation defects can be effectively captured. The calculated geometric parameters include the area, perimeter, aspect ratio, etc. of the defect, and these parameters will be used for subsequent error calculation. After completing the error calculation, a suitable error compensation calculation method is selected. Commonly used compensation methods include linear compensation and non-linear compensation. Linear compensation methods are suitable for smaller errors, while non-linear compensation is suitable for dealing with larger and uneven errors. A compensation factor is set, such as set to 1.1 or 0.9, to adjust the geometric parameters. The selection of the compensation factor should be calculated based on the average error of the actual measurement. All compensated geometric parameters are organized into encapsulation defect compensation parameters. These parameters should include the compensation values for each encapsulation location and their corresponding ideal values for subsequent process adjustment and quality control. Ensure that the generated encapsulation defect compensation parameters can effectively reflect the problems in actual production for real-time adjustment and optimization in subsequent production processes.

[0028] Step S6: Fine-tune the process parameters of the encapsulation process stage causing the defect according to the encapsulation defect compensation parameters and perform adaptive learning optimization to construct an intelligent encapsulation process control model.

[0029] In this embodiment, before fine-tuning the process parameters, it is first necessary to select a suitable fine-tuning method. Common fine-tuning methods include the response surface method and the gradient descent method. The response surface method is suitable for dealing with multi-variable optimization problems and can consider the influence of multiple process parameters simultaneously; while the gradient descent method is suitable for simplified problems and can quickly converge to a local optimal solution. Considering the complexity of the packaging process and the influence of multiple parameters, it is recommended to use the response surface method for fine-tuning the process parameters. This method can establish a mathematical model between the process parameters and the product quality, thereby achieving effective optimization. Set the fine-tuning goal, such as reducing the defect rate, improving the packaging efficiency, or enhancing the product consistency. On this basis, determine the key process parameters that need to be fine-tuned, such as temperature, pressure, packaging time, etc. Design an experimental plan, set different combinations of process parameters, so as to obtain the best process parameter settings through the response surface method. For example, set the temperature range as (20°C, 30°C), the pressure range as (1.5 bar, 2.5 bar), and the packaging time range as (1 minute, 3 minutes), and conduct a full factorial experimental design to ensure that all parameter combinations can be covered. Conduct experiments, collect the packaging results under each experimental condition, including the defect rate, the yield rate, and other quality indicators. According to the experimental results, construct a response surface model, and use a quadratic polynomial to fit the relationship between the process parameters and the quality indicators. By analyzing the response surface, identify the process parameters that have the greatest impact on the packaging quality. For example, if the model shows that the temperature has a significant impact on the defect rate, then fine-tune the temperature parameter first. After adjustment, record the fine-tuned parameter values and their corresponding quality indicators to ensure that each fine-tuning can effectively improve the product quality. After completing the fine-tuning, construct an adaptive learning optimization model. Select a suitable machine learning algorithm, such as random forest or support vector machine (SVM), to establish the mapping relationship between the process parameters and the quality indicators. Use the previous experimental data to train the model and set parameters, such as the number of trees (which can be set to 100 for random forest) and the maximum depth (which can be set to 5). During the model training process, use the cross-validation method to evaluate the generalization ability of the model to ensure that it can maintain a high accuracy on unseen data. Record the loss value and accuracy of each training to monitor the learning process of the model. During the process production, implement real-time monitoring, record the process parameters and their corresponding product quality indicators. Use the constructed adaptive learning optimization model to dynamically adjust the process parameters according to the real-time data. For example, when it is detected that the defect rate increases, adjust the temperature or pressure parameters in real time to address potential quality problems. Through the feedback adjustment mechanism, continuously optimize the setting of the process parameters to ensure that it can respond to the changes in the production process in a timely manner. Record the effect of each adjustment to continuously improve the accuracy and practicality of the model. After completing the above steps, integrate the fine-tuned process parameters and the adaptive learning optimization model to construct an intelligent packaging process control model. Ensure that this model can not only predict the packaging quality but also make automatic adjustments according to the real-time data.Establish an evaluation mechanism for the model, verify it using actual production data, and ensure that the model can work stably under different production conditions. Record the prediction accuracy and adjustment effect of the model to evaluate the practical application value of the intelligent control model.

[0030] In this embodiment, refer to Figure 2 , which is a schematic diagram of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Step S11: Obtain the monitoring image of the in vitro diagnostic reagent after encapsulation; Step S12: Perform histogram distribution analysis on the monitoring image of the in vitro diagnostic reagent to obtain the dark area and bright area of the image; Step S13: Calculate the regional pixel brightness deviation of the dark area of the image according to the bright area to obtain the dark area pixel brightness deviation value; Step S14: Perform adaptive histogram equalization processing according to the dark area pixel brightness deviation value to obtain a brightness-optimized monitoring image; Step S15: Perform multi-scale reconstruction on the brightness-optimized monitoring image to generate a multi-scale noise-reduced reconstructed monitoring image.

[0031] In this embodiment, prepare a suitable image acquisition device, such as a high-resolution digital camera or a professional microscope imaging system. Ensure that the device can clearly capture the monitoring image of the in vitro diagnostic reagent under appropriate lighting conditions. The device should have the function of adjusting the focal length and aperture to optimize the image quality. Set up a good monitoring environment, including controlling the light source (such as using an LED light source), avoiding external interference (such as vibration, temperature change, etc.). The stability of the environment is crucial for improving the image quality. It is recommended to perform image acquisition in a laboratory with controllable temperature and humidity. After preparing the device and environment, perform image acquisition. According to needs, set different exposure times and ISO values to ensure that the brightness of the image is suitable for subsequent analysis. It is recommended to take multiple shots and save multiple image samples for subsequent selection of the best image for analysis. Save the acquired images in a lossless format (such as TIFF or PNG) to retain the details and quality of the images. Ensure that the naming of the image files is standardized for subsequent processing and management. Record the acquisition parameters of each image (such as time, lighting conditions, device settings, etc.) for subsequent tracking and analysis. Use image processing software (such as OpenCV, MATLAB, or the PIL library of Python) to calculate the histogram of the acquired monitoring image. The histogram shows the distribution of the brightness values of each pixel in the image. Usually, the brightness value range is set between 0 and 255. import cv2 import matplotlib.pyplot as plt image = cv2.imread('path_to_image') gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) histogram = cv2.calcHist([gray_image], [0], None,

[256] , [0, 256]) plt.plot(histogram) plt.title('Histogram') plt.show() Based on the distribution of the histogram, set a threshold to distinguish the dark and bright regions of the image. Common thresholding methods include the Otsu method, which automatically calculates the optimal threshold to divide the image into foreground and background. During this process, record the number of pixels and their corresponding brightness values in the dark and bright regions for subsequent analysis. Visualize the histogram and the image after thresholding to ensure good recognition of the dark and bright regions. Through visualization, it is possible to more intuitively analyze the brightness distribution of the image and the effect of threshold selection. After identifying the dark and bright regions, extract the pixel brightness values from the bright region. Use the NumPy library to calculate the mean and standard deviation of the pixel values in the bright region.

[0032] mean_bright = np.mean(gray_image[thresholded_image == 255]) std_bright = np.std(gray_image[thresholded_image == 255]) Calculate the pixel brightness deviation of the dark area based on the mean value of the bright area. Use the following formula: Deviation = Mean value of the bright area - Pixel value of the dark area. Use a loop to traverse the pixels in the dark area, calculate the deviation value of each pixel, and store it in an array. Finally, obtain the brightness deviation value of the dark area. Conduct statistical analysis on the brightness deviation value of the dark area, calculate indicators such as its mean value and variance to evaluate the overall situation of the brightness deviation. These statistical data will be used in the subsequent adaptive histogram equalization processing stage. Select a suitable adaptive histogram equalization algorithm, such as CLAHE (Contrast Limited Adaptive Histogram Equalization). This method can equalize the histogram within a local area, effectively improve the image contrast, and is particularly suitable for processing images with different lighting conditions. Set the parameters of CLAHE, including the grid size (e.g., 8x8) and the threshold for limiting the contrast (e.g., 2.0), to control the degree of detail enhancement of the algorithm. Apply CLAHE processing to the original monitoring image to generate a monitoring image with optimized brightness. Ensure that the parameters and results of each step are recorded during the processing for subsequent effect evaluation. Visualize the images before and after processing and analyze the brightness optimization effect. Evaluate the effect of adaptive histogram equalization by comparing the histograms and the visual effects of the images to determine whether it meets the requirements of subsequent processing. Select a suitable multi-scale reconstruction algorithm, such as wavelet transform or Laplacian pyramid. Wavelet transform can extract image features in different frequency ranges and effectively perform image denoising and reconstruction. Apply wavelet transform to the monitoring image with optimized brightness. Use the PyWavelets library for wavelet transform and select a suitable wavelet basis (such as Haar or Daubechies).

[0033] import pywt coeffs = pywt.wavedec(clahe_image, 'haar', level=2) Reconstruct the image based on the wavelet coefficients and set a threshold to remove noise. Use the soft threshold or hard threshold method to ensure that the reconstructed image is clearer. Conduct quality evaluation on the reconstructed monitoring image. Use indicators such as PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index) to evaluate the reconstruction effect. Intuitively display the processing effect by visually comparing the original image, the image with optimized brightness, and the reconstructed image.

[0034] In this embodiment, the specific steps of step S15 are as follows: Detect the reagent edge details of the monitoring image with optimized brightness and mark multiple reagent edge details; Visually identify the edge details of multiple reagents to generate the visual texture features of each reagent edge; Sharpen and enhance the visual texture features of each reagent edge to generate a sharpened and enhanced monitoring image; Perform multi-scale decomposition on the sharpened and enhanced monitoring image to obtain multiple sub-bands with different frequencies; Perform per-sub-band noise analysis on the multiple sub-bands with different frequencies to generate the noise features of each frequency sub-band; Perform multi-band adaptive filtering and noise reduction on the noise features of each frequency sub-band to obtain multiple filtered and noise-reduced sub-bands; Perform multi-scale reconstruction based on the multiple filtered and noise-reduced sub-bands to generate a multi-scale noise-reduced reconstructed monitoring image.

[0035] In this embodiment, a suitable edge detection algorithm is selected to identify the edge details of the reagent. Commonly used edge detection algorithms include the Canny edge detection algorithm and the Sobel operator. The Canny algorithm is usually the first choice due to its superiority in noise suppression and edge localization. For this purpose, appropriate Gaussian filter parameters need to be set to remove the noise in the image. Apply the Canny edge detection algorithm to the monitored image after brightness optimization. Set the convolution kernel size of the Gaussian filter (such as 5x5) and the low and high thresholds (for example, the low threshold is set to 50 and the high threshold is set to 150). Through this setting, the edge details of the reagent can be effectively extracted. The result of edge detection will generate a binary image, showing the detected edge regions. Mark the detected edges. Use a marking algorithm (such as connected component analysis) to identify and mark different edge regions in the image, ensuring that the edges of each reagent are clearly identified. This process helps with subsequent visual texture recognition and analysis. Select a suitable texture feature extraction method, such as the gray-level co-occurrence matrix (GLCM) or local binary pattern (LBP). The GLCM can capture the relationships between gray levels in the image, while the LBP is sensitive to local texture changes and is suitable for texture analysis of reagent edges. Calculate the texture features for each marked reagent edge region. For example, use the GLCM to calculate features such as contrast, correlation, and homogeneity for each region, or use the LBP to calculate local texture patterns. These features will provide a basis for subsequent visual analysis. Store the extracted visual texture features in a structured database for subsequent calling and analysis. Record the identifier of each reagent and its corresponding texture feature values for subsequent comparison and research. Select a suitable image sharpening algorithm, such as Laplacian filtering or high-pass filtering. Laplacian filtering can effectively enhance the edge details of the image and improve the image clarity. Sharpen the monitored image after brightness optimization. Set the convolution kernel of the Laplacian operator (such as 3x3) and apply it to the image. During processing, pay attention to adjusting the processing intensity to avoid artifacts and noise caused by oversharpening. Conduct a visual evaluation of the sharpened and enhanced monitored image to ensure that the edge details are effectively enhanced. By comparing with the original image, observe the clarity of the edges and the retention of details. Select a suitable multi-scale decomposition method, such as wavelet transform or Laplacian pyramid. The wavelet transform is suitable for multi-scale analysis of images due to its good performance in the time and frequency domains. Apply wavelet transform to the sharpened and enhanced monitored image and select a suitable wavelet basis (such as Daubechies). Set the number of decomposition levels (such as 2 or 3 levels) to obtain subbands of different frequencies. Extract each subband after wavelet transform, including low-frequency and high-frequency subbands. Analyze the frequency characteristics of each subband to evaluate the structure and detail retention of the image. Select a suitable noise analysis method, usually using spectral analysis or noise estimation techniques. Spectral analysis can evaluate the frequency components of the noise in each subband.Extract noise features for each frequency subband and calculate its signal-to-noise ratio (SNR) or root mean square error (RMSE) and other indicators. These indicators will be used for subsequent noise filtering decisions. Record and store the noise features of each subband for subsequent analysis and comparison. Analyze the noise features of different subbands to provide a basis for adaptive filtering. Select a suitable adaptive filtering algorithm, such as Wiener filtering or bilateral filtering. Wiener filtering is suitable for situations where the signal and noise are clearly distinguished, while bilateral filtering reduces noise while maintaining the edges. Apply the selected adaptive filtering algorithm to each frequency subband and set the corresponding parameters (such as noise variance, spatial range, etc.) to ensure the optimization of the filtering effect. After filtering, evaluate the noise reduction effect of each subband. Evaluate the filtered subbands and compare the signal-to-noise ratio and visual effect before and after noise reduction. Judge the effectiveness of adaptive filtering by analyzing the results and consider whether further optimization of parameters is needed. Select a suitable reconstruction method, usually using the inverse wavelet transform or Laplace pyramid reconstruction algorithm. The inverse wavelet transform can synthesize the filtered subbands back to an approximation of the original image. Combine multiple filtered and denoised subbands and reconstruct them using the inverse wavelet transform algorithm. Ensure that as much detail information as possible is maintained during the reconstruction process while suppressing noise. Visualize the reconstructed final monitoring image and analyze its clarity, detail retention, and noise level. Evaluate the reconstruction effect by comparing it with the original monitoring image, and record relevant experimental parameters and results.

[0036] In this embodiment, refer to Figure 3 , is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Step S21: performing deep semantic visual recognition on the multi-scale noise reduction and reconstruction monitoring image to obtain multiple in vitro diagnostic reagent packaging locations; Step S22: performing image visual segmentation on multiple in vitro diagnostic reagent packaging locations, thereby generating multiple packaging local sub-images; Step S23: performing iterative packaging defect detection on multiple packaging local sub-images one by one to mark the packaging defect locations; Step S24: enlarging and extracting the package defect location, thereby obtaining an enlarged monitoring image of the package defect location.

[0037] In this embodiment, a suitable deep learning model is selected for semantic visual recognition. Commonly used models include Convolutional Neural Network (CNN), U-Net, and Mask R-CNN, etc. Due to the complexity of the task and the details to be processed, the Mask R-CNN model is recommended, which can effectively perform instance segmentation and identify different parts in the image. Before performing deep semantic visual recognition, the corresponding training dataset needs to be prepared. Use the existing labeled dataset or perform data annotation according to actual needs. It is necessary to ensure that the dataset contains encapsulated samples of various in vitro diagnostic reagents so that the model can learn the characteristics of different encapsulated parts. Use the selected deep learning model for training, and set appropriate hyperparameters, including the learning rate (such as 0.001), batch size (such as 16), and number of training epochs (such as 50 epochs). During the training process, use the cross-entropy loss function for optimization to ensure that the model can gradually improve the recognition accuracy. After the training is completed, evaluate the model performance through the validation set to ensure that the recognition accuracy rate reaches the expected target (such as more than 85%). Apply the trained model to the multi-scale denoising reconstruction monitoring image for deep semantic visual recognition. The model will automatically identify the encapsulated parts of multiple in vitro diagnostic reagents and generate the corresponding label image. Record the accuracy of the recognition results and the response time of the model for subsequent analysis. Select a suitable image segmentation algorithm. Use threshold segmentation, region growing, or deep learning-based segmentation methods (such as U-Net). Since deep semantic recognition has been performed in the early stage, the deep learning-based U-Net algorithm is recommended to improve the segmentation accuracy. Prepare data for the identified encapsulated parts to ensure that the format of the input image is suitable for the U-Net model. Standardize the image, including resizing the image (such as 256x256 pixels) and performing normalization. Apply the U-Net model for image visual segmentation to generate multiple encapsulated local sub-images. Set appropriate hyperparameters, such as the learning rate and the number of iterations, to ensure that the model can effectively segment the required regions. After the segmentation is completed, check the quality of each sub-image to ensure that it contains clear encapsulated parts. Evaluate the segmentation results, calculate the segmentation accuracy and recall rate, and ensure that the model can effectively segment each encapsulated part. Record the characteristics of each sub-image for subsequent defect detection. Select a suitable defect detection algorithm. Adopt traditional image processing methods (such as edge detection, morphological operations) or deep learning-based object detection methods (such as YOLO or Faster R-CNN). Considering accuracy and flexibility, the YOLO algorithm is recommended. If YOLO is used for detection, a labeled dataset needs to be prepared, in which the positions of defects need to be marked in each encapsulated sub-image. Perform model training, set appropriate hyperparameters, including the learning rate, batch size, and number of training epochs. After the training is completed, evaluate the accuracy and detection speed of the model. Apply the trained YOLO model to each encapsulated local sub-image for defect detection.The model will automatically identify packaging defects and mark the defective parts on the image. Record the detection time and accuracy of each image for subsequent analysis. Store the detected defect results to generate a defect report, including information such as the defect type, location, and size of each sub-image. These results will provide a basis for subsequent defect analysis and improvement. According to the detection results, extract the marked defective areas from each local sub-image of the package. Determine the coordinates and size of the extracted area to ensure that the extracted area contains sufficient context information. Select a suitable image enlargement technique. Commonly used techniques include image interpolation (such as bilinear interpolation, cubic interpolation) and image cropping. For detail enhancement, bilinear interpolation is recommended for enlargement. Enlarge the image of each extracted defective part. Set an appropriate enlargement ratio (such as 2 times or 3 times) to ensure that the details are clearly visible. During the implementation process, ensure the quality of the results for subsequent analysis. Visually evaluate the enlarged defect monitoring image to ensure that the details are clear and there is no distortion. Store the enlarged image for subsequent defect analysis and report generation.

[0038] In this embodiment, refer to Figure 4 , which is a schematic diagram of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Step S31: Perform three-dimensional morphological analysis on the enlarged monitoring image of the packaging defect part to generate three-dimensional morphological features of the defect part; Step S32: Mine the surface defect texture of the enlarged monitoring image of the packaging defect part to obtain surface defect texture data; Step S33: Classify the defects according to the three-dimensional morphological features of the defect part and the surface defect texture data to obtain the current defect type label; Step S34: Infer the deep defect factors according to the current defect type label to generate packaging defect inference factors.

[0039] In this embodiment, a suitable three-dimensional morphology analysis method is selected. Commonly used methods include stereovision technology and structured light scanning. Stereovision technology calculates depth information through images from multiple viewpoints and is suitable for obtaining three-dimensional morphological information of defective parts. Structured light scanning, on the other hand, obtains three-dimensional data of the object surface by projecting a specific light pattern. The enlarged monitoring images are taken from multiple viewpoints to ensure obtaining image data of the defective part from different angles. During the shooting process, standard lighting conditions and camera parameters are set to maintain the consistency of the images. At least three images from different angles are required for subsequent three-dimensional reconstruction. The stereovision algorithm is used to process the multi-viewpoint images, and the three-dimensional model is reconstructed by matching feature points and calculating disparities. A high matching accuracy (such as 0.1 pixel) is used to ensure the accuracy of the reconstruction. The generated three-dimensional model should be able to represent the shape, size, and position of the defective part. The processing time and reconstruction accuracy are recorded for subsequent evaluation. Morphological features are extracted from the reconstructed three-dimensional model, including the volume, surface area, shape factor, etc. of the defect. Morphological analysis methods, such as calculating the geometric center and principal component analysis, are adopted to quantify the three-dimensional features of the defect. These features will serve as the basis for subsequent analysis. A suitable texture analysis method is selected, such as the gray-level co-occurrence matrix (GLCM) and local binary pattern (LBP). GLCM can capture the relationship between gray levels in the image, while LBP can efficiently extract local texture features and is suitable for analyzing surface defects. The enlarged monitoring images are subjected to texture analysis to calculate the GLCM and LBP features of the defective part. Appropriate distance and angle parameters (such as 0°, 45°, 90°, and 135°) are set to comprehensively capture the texture information in the image. The extracted features include contrast, correlation, uniformity, and entropy, etc. The extracted texture data is stored in the database to ensure that the features of each defective part are recorded. By statistically analyzing the distribution of texture features, potential defect patterns are identified. The mean and standard deviation of each feature are recorded for subsequent classification and inference. A suitable defect classification model is selected. Machine learning-based methods (such as support vector machines, random forests) or deep learning models (such as convolutional neural networks) are used. Considering the complexity of the data, the random forest model is recommended as it has good processing ability for the non-linear relationships of features. The three-dimensional morphological features and surface texture data are combined to construct a feature set. The feature set should include all the extracted morphological and texture features to ensure the comprehensiveness of information. Feature selection criteria are set to retain the features that have a greater impact on the classification results. The constructed feature set is used to train the selected classification model, and appropriate hyperparameters are set, such as the number of trees (such as 100 trees) and the maximum depth (such as 10 layers). The performance of the model is evaluated through cross-validation to ensure that the classification accuracy reaches the expected target (such as over 90%). The trained model is applied to classify new defect data to generate labels for the current defect types. The type of each defect and its corresponding features are recorded for subsequent analysis and improvement.Select appropriate inference methods, usually using causal inference models or Bayesian networks. Causal inference models can analyze the causes of defects, while Bayesian networks are suitable for dealing with uncertainties and complex relationships. Integrate defect type labels and relevant experimental data to establish a basic dataset for defect factor inference. Ensure that the dataset contains sufficient samples for effective inference analysis. Record the occurrence frequency and relevant parameters of each defect. Apply the selected inference model for analysis and set appropriate parameters (such as belief update thresholds) to ensure the reliability of the inference results. By analyzing the relationships between defect types and other factors, infer the factors that lead to defects. Summarize the inference results and generate an inference factor report for encapsulating defects. Include the causes and their impact degrees of each defect type to provide a basis for subsequent quality improvement and prevention and control measures. Record the key parameters and results during the inference process for subsequent review.

[0040] In this embodiment, step S4 includes the following steps: Step S41: Obtain the reagent strip encapsulation log; Step S42: Conduct an encapsulation process behavior analysis on the reagent strip encapsulation log and extract the encapsulation process behavior data of multiple stages; Step S43: Conduct a phased process logic analysis on the encapsulation process behavior data of multiple stages to generate an encapsulation process logic rule; Step S44: Perform multi-timepoint encapsulation process evolution according to the encapsulation process logic rule to construct a time-sequential encapsulation process behavior sequence; Step S45: Conduct defect process tracing on the time-sequential encapsulation process behavior sequence according to the encapsulation defect inference factors to mark the encapsulation process stages that cause defects.

[0041] In this embodiment, the source of the reagent strip packaging logs is confirmed. These logs are usually generated by the packaging equipment and record detailed information about each link in the production process, including equipment status, operators, timestamps, and various process parameters. Therefore, it is necessary to interface with the production line equipment to ensure that the packaging logs can be obtained in real time. Set up an automated data acquisition system to regularly extract the packaging logs from the packaging equipment. Use a database management system (such as MySQL or MongoDB) to store these log data for subsequent analysis. It is recommended to standardize the data storage format, including fields such as time, operation type, parameter values, etc., to improve the readability and processing efficiency of the data. After obtaining the packaging logs, perform data preprocessing. Clean the data to remove redundant information and duplicate records to ensure the integrity and accuracy of the log data. Set metrics for data preprocessing, such as the validity of the records (e.g., more than 90%) and the consistency of the timestamps (e.g., no time jumps). Clearly define the packaging process behaviors, including key process stages (such as heating, cooling, packaging, etc.) and related process parameters (such as temperature, pressure, time, etc.). These definitions will provide a basis for subsequent data analysis. Extract data related to the process behaviors from the packaging logs. Use data analysis tools (such as Pandas or Excel) to analyze the extracted data and identify the process behaviors at each stage. Set the extraction time window (such as every minute or every hour), and divide the data into different stages according to the time series. Analyze the process behaviors at each stage and extract relevant features (such as mean, maximum, minimum, variance, etc.). These features will lay the foundation for subsequent logical analysis to ensure that the process performance at each stage can be comprehensively evaluated. According to the extracted stage-based process data, construct a process logic model. Use models such as logistic regression or decision trees to perform correlation analysis on the process parameters and results at different stages to identify the key factors affecting the packaging quality. Through model analysis, extract the logical relationships of the packaging process. For example, the relationship between the heating time within a certain temperature range and the packaging quality. These logical relationships can clarify the influence degree of each process stage and provide a basis for subsequent optimization. According to the analysis results, generate the logical rules of the packaging process. These rules will help identify which process parameters have the greatest impact on the final product quality at different stages, and record the accuracy and applicable scope of each rule for subsequent reference. Select a suitable time series evolution model, such as a state space model or a time series prediction model. Through these models, simulate the evolution process of the packaging process at different time points and analyze the change trends of the process parameters. Organize the stage-based process behavior data into a format suitable for model input, including timestamps, process parameters, and corresponding output results. Ensure the continuity and integrity of the data to improve the prediction accuracy of the time series model. Apply the selected time series evolution model to perform multi-timepoint packaging process evolution analysis. Record the changes in the process parameters at each time point and predict the future process trends based on historical data.Evaluate the prediction ability of the model to ensure that the prediction error is within an acceptable range (e.g., within 10%). Integrate the previously generated encapsulation defect inference factors with the time-series encapsulation process behavior sequence. Ensure that each defect type has corresponding process parameters and stages for traceability analysis. Select a suitable traceability analysis method, such as causal relationship analysis or root cause analysis. Through these methods, identify the specific process stages and parameters that cause defects. Trace the time-series encapsulation process behavior sequence step by step, record the process parameters at each stage and their corresponding defect types. By analyzing the relationship between process changes and defect occurrences, mark the specific process stages that cause defects. Record the results of the traceability analysis to generate a defect process traceability report. The report should include the occurrence stage of each defect, related parameters, and their impact levels, providing a basis for subsequent process optimization and quality control.

[0042] In this embodiment, step S5 includes the following steps: Step S51: Calculate the geometric dimensions of the encapsulation part by expanding the monitoring image according to the encapsulation defect part to generate geometric parameters of the encapsulation defect part; Step S52: Calculate the parameter error of the geometric parameters of the encapsulation defect part based on the preset reagent strip encapsulation parameter threshold to obtain the error parameter of the encapsulation defect part; Step S53: Calculate the error compensation of the error parameter of the encapsulation defect part to generate the compensation parameter of the encapsulation defect.

[0043] In this embodiment, a suitable geometric dimension calculation method is selected to accurately measure the geometric parameters of the encapsulated defect site. Commonly used methods include edge detection and contour extraction techniques, implemented in combination with image processing software (such as OpenCV or MATLAB). For defects with complex shapes, it is recommended to use region-based measurement methods. Using the previously obtained enlarged monitoring image, preprocessing is carried out to improve the calculation accuracy. This includes steps such as grayscale processing of the image, noise removal, and contrast enhancement. Set appropriate image processing parameters, such as the convolution kernel size of the Gaussian filter (for example, 5x5), to reduce background noise and highlight the defect edges. In the processed image, apply the Canny edge detection algorithm to extract the edges of the encapsulated defect site. Set high and low thresholds (such as 50 and 150) to ensure that the defect contour can be effectively captured. After edge detection, use the contour extraction algorithm (such as the findContours method) to obtain the contour information of the defect. According to the extracted contour information, calculate the geometric dimensions of the encapsulated defect site, including area, perimeter, aspect ratio, etc. Convert the pixel units in the image to actual physical dimensions through a calibration coefficient (for example, the actual length corresponding to each pixel). Record the calculated geometric parameters for subsequent error analysis. According to the design specifications and production standards of the reagent strip, preset the encapsulation parameter thresholds. These thresholds include the allowable ranges of each geometric parameter (for example, the area should be between 10 - 15 square millimeters, and the aspect ratio should be between 1:1 and 1:2). Ensure that these thresholds can reflect the quality standards of the actual product. Select a suitable error calculation method, usually using the calculation methods of absolute error and relative error. Calculate the absolute error for each geometric parameter by subtracting the preset value from the actual measured value. First, calculate the absolute error: Absolute Error = |Actual Value - Preset Value|. Then, calculate the relative error: Relative Error = Absolute Error / Preset Value × 100%. Record the error values of each geometric parameter and perform statistical analysis on these errors, such as calculating the average error and the maximum error, to evaluate the stability of the overall production process. Select a suitable error compensation method, commonly including linear compensation methods and non-linear compensation methods. Linear compensation is applicable to cases where the errors are small and relatively uniform, while non-linear compensation is suitable for handling larger and non-uniform errors. According to the results of error calculation, set the compensation parameters. The goal of compensation is to adjust the actual measured value to the preset target range. The compensation parameters are calculated based on the average error of the actual measurement, and a certain compensation factor (such as 1.1 times or 0.9 times) is set to adjust the geometric parameters. Perform compensation calculations for each geometric parameter using the following formula: Compensated Value = Actual Value + Compensation Factor × Absolute Error. When implementing this process, ensure that all compensation parameters comply with the preset tolerance range and record the compensated results. Evaluate the compensated geometric parameters to ensure that they are within the preset threshold range. Verify the compensation effect through statistical analysis (such as calculating the average value and standard deviation after compensation). If necessary, conduct further experiments to confirm the effectiveness of the compensation strategy.

[0044] In this embodiment, step S6 includes the following steps: Step S61: Calculate the adjustment of the current process parameters for the packaging process stage causing the defect according to the packaging defect compensation parameters, so as to obtain multiple process parameter adjustment ranges; Step S62: Simulate the adjustment of the packaging process parameters based on multiple process parameter adjustment ranges to obtain multiple parameter adjustment simulation schemes; Step S63: Evaluate the packaging efficiency of multiple parameter adjustment simulation schemes to generate the packaging efficiency of each scheme; Step S64: Analyze the optimal adjustment range according to the packaging efficiency of each scheme, so as to obtain the optimal process parameter adjustment range value; Step S65: Fine-tune the process parameters of the packaging process production line based on the optimal process parameter adjustment range value, and perform adaptive learning optimization to construct an intelligent packaging process control model.

[0045] In this embodiment, first, the encapsulation defect compensation parameters calculated previously need to be integrated with the corresponding process stages. The parameters of each process stage (such as temperature, pressure, time, etc.) should be adjusted accordingly based on the compensation parameters. It is necessary to ensure that the compensation parameters can effectively reflect the defect causes identified in the previous analysis. Select a suitable adjustment calculation method. Generally, the linear adjustment or proportional adjustment method is used. Linear adjustment is suitable for simple increases and decreases, while proportional adjustment is more suitable for more complex process parameter adjustments, especially when multiple parameters change. For each process parameter, the following formula is applied for adjustment calculation: New parameter value = Current parameter value + Compensation parameter × Current parameter value. Through this formula, the adjustment range of each process parameter is obtained. Record each adjusted process parameter and its range. These adjustment ranges will provide basic data for subsequent simulations and evaluations. Build a simulation model of the encapsulation process and select a suitable simulation tool (such as MATLAB / Simulink, AnyLogic, etc.) for parameter adjustment simulation. The simulation model should be able to reflect the actual operation of the production line, including the mutual influence between various process parameters. Generate multiple parameter adjustment simulation schemes based on the calculated adjustment ranges of multiple process parameters. Each scheme should include different parameter combinations to ensure that all adjustment methods are covered. For example, set different adjustment ranges for temperature, pressure, and time, and record the specific parameter values of each scheme. Perform simulation execution for each parameter adjustment scheme and record the output results of each scheme, including key indicators such as encapsulation efficiency, defect rate, and production cost. These indicators will provide data support for subsequent efficiency evaluations. Set the indicators for encapsulation efficiency evaluation, including production speed, product qualification rate, and resource utilization rate, etc. These indicators will help evaluate the effects of each simulation scheme and ensure that the performance of the encapsulation process can be comprehensively reflected. Select a suitable efficiency evaluation method, such as the multi-factor analysis method or the weighted scoring method. Through these methods, multiple indicators are integrated into a comprehensive efficiency score, making it easier to compare the efficiencies of different schemes. Conduct encapsulation efficiency evaluation for each simulation scheme and calculate the comprehensive efficiency score of each scheme. Record the evaluation results of each scheme, including the scores of each indicator and the total score, for subsequent analysis. Select a suitable optimal adjustment range analysis method, such as Pareto analysis or sensitivity analysis. Pareto analysis helps identify the parameters that have the greatest impact on encapsulation efficiency, while sensitivity analysis can evaluate the impact degree of different parameter changes on the overall efficiency. Integrate the encapsulation efficiency data of each scheme and identify which parameter adjustment ranges contribute the most to the improvement of encapsulation efficiency. By comparing the efficiency scores of different schemes, find the range of the optimal adjustment range. Based on the analysis results, determine the optimal process parameter adjustment range value. Record the specific value of this adjustment range and its corresponding encapsulation efficiency to ensure that the production process can be adjusted according to this parameter in the future. Based on the determined optimal adjustment range value, formulate a specific process parameter fine-tuning scheme.Ensure that the solution includes detailed adjustment steps, target parameters, and their expected effects, so that there is a basis to follow during implementation. Build an adaptive learning optimization model and select a suitable algorithm (such as reinforcement learning or genetic algorithm) for real-time optimization of process parameters. This model should be able to continuously adjust process parameters based on the feedback data during the production process to achieve the best performance. Fine-tune the process parameters of the encapsulation process production line, implement the adjusted parameters, and monitor various indicators during the production process in real time. By collecting feedback data, verify the effect of the fine-tuning, and further optimize according to the actual situation. Analyze the results of the fine-tuning and evaluate the production efficiency, product quality, and cost-effectiveness after adjustment. Summarize the experiences and lessons during the adjustment process to provide a reference for subsequent process optimization.

[0046] In this embodiment, an in vitro diagnostic reagent strip encapsulation quality control system is provided for performing the method described above, including: A multi-scale reconstruction module for obtaining an in vitro diagnostic reagent monitoring image after encapsulation; performing multi-scale reconstruction on the in vitro diagnostic reagent monitoring image to generate a multi-scale noise-reduced reconstructed monitoring image; An encapsulation defect detection module for performing image vision segmentation and individual image iterative encapsulation defect detection on the multi-scale noise-reduced reconstructed monitoring image to obtain an enlarged monitoring image of the encapsulation defect location; A defect inference module for performing surface defect texture mining on the enlarged monitoring image of the encapsulation defect location and performing deep defect factor inference to generate encapsulation defect inference factors; A process traceability module for obtaining the reagent strip encapsulation log; performing defect process traceability on the reagent strip encapsulation log according to the encapsulation defect inference factors to mark the encapsulation process stage that causes the defect; A defect compensation module for calculating the encapsulation location error based on the enlarged monitoring image of the encapsulation defect location and performing error compensation calculation to generate encapsulation defect compensation parameters; An encapsulation process optimization module for fine-tuning the process parameters of the encapsulation process stage that causes the defect according to the encapsulation defect compensation parameters and performing adaptive learning optimization to build an intelligent encapsulation process control model.

[0047] The present invention denoises the monitoring images through multi-scale reconstruction technology, which can effectively remove background noise, enhance the image quality of the target area, and provide higher-quality image data for subsequent defect detection. This makes the identification of defects more accurate and improves the accuracy of subsequent processing. Multi-scale reconstruction can process images at different resolutions and levels, helping to extract more detailed information, especially for packaging defects with different sizes or shapes, improving the comprehensiveness of detection. Image vision segmentation technology helps to divide the reagent strip monitoring image into multiple meaningful regions, focusing on the regions with defects, and improving the accuracy and efficiency of defect detection. Iterative packaging defect detection for each image helps to systematically analyze each defect and avoid omission. By enlarging the monitoring image, the defective part can be presented more clearly, making subsequent analysis and processing more accurate. Especially in the discovery of small or hidden defects, the enhanced image provides stronger visual support. Through surface defect texture mining, the details of packaging surface defects can be deeply analyzed to identify different types of defect textures. And through in-depth inference of deep defect factors, the causes of defects are deeply explored, providing a scientific basis for optimizing the packaging process. By inferring the deep factors of packaging defects, it helps to find potential and overlooked defect sources. This provides data support for subsequent process traceability and optimization, and helps to discover systematic problems existing in the packaging process. By combining the inferred factors of packaging defects with packaging logs, the defects are traced and the process stage where the defects occur is accurately located. This provides a clear basis for subsequent process improvement and quality control, avoiding empirical guesses and uncertainties. Through traceability analysis, not only can specific defects be marked, but also the deficiencies in the process can be pointed out, ensuring that the causes of defects are clearly identified, thus avoiding recurrence. By calculating the error of the packaging part and compensating it, the packaging deviation caused by reasons such as process instability and equipment error is effectively corrected. This process helps to reduce product quality fluctuations and improve the accuracy and consistency of packaging. The generation and application of error compensation parameters provide a basis for real-time adjustment during the packaging process, quickly correcting the deviation in the process, and ensuring that the product quality of each batch is within the standard range. According to the defect compensation parameters, the packaging process is fine-tuned, enabling the process to be dynamically adjusted during actual production, and enhancing the adaptability and flexibility of the packaging process. The adaptive learning optimization mechanism helps the system automatically adjust packaging parameters according to historical data and real-time feedback, enabling the packaging process to be automatically optimized under different production conditions. This intelligent control can effectively improve production efficiency and reduce human intervention. By constructing an intelligent packaging process control model, the packaging process can be continuously optimized for a long time, not only solving current defect problems but also preventing future problems, providing guarantee for long-term quality control.

[0048] Therefore, in all respects, the embodiments should be regarded as exemplary and non-limiting. 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 within the present invention.

[0049] As described above, these are merely 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 the embodiments shown herein, but rather will conform to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A method for controlling the packaging quality of an in vitro diagnostic reagent strip, characterized in that: The following steps are involved: Step S1: Acquire a monitoring image of the in vitro diagnostic reagent after packaging; Performing multi-scale reconstruction on the in vitro diagnostic reagent monitoring image to generate a multi-scale noise reduction and reconstructed monitoring image; Step S2: performing image visual segmentation and iterative packaging defect detection on the multi-scale noise reduction and reconstruction monitoring image, thereby obtaining an enlarged monitoring image of the packaging defect location; Step S3: mining the surface defect texture of the enlarged monitoring image of the package defect site and inferring the deep defect factors, thereby generating the package defect inference factors; Step S4: Obtaining the reagent strip packaging log; Based on the packaging defect inference factors, the defect process is traced back to the reagent strip packaging log, thereby marking the packaging process stage that caused the defect; Step S5: performing packaging part error calculation according to the enlarged monitoring image of the packaging defect part, and performing error compensation calculation to generate packaging defect compensation parameters; Step S6: fine-tuning the process parameters of the packaging process stage causing the defect according to the packaging defect compensation parameters, and performing adaptive learning optimization to construct an intelligent packaging process control model.

2. The method for controlling the packaging quality of in vitro diagnostic reagent strips according to claim 1, characterized in that: The specific steps of step S1 are: Step S11: Acquire a monitoring image of the in vitro diagnostic reagent after packaging; Step S12: performing a histogram distribution analysis on the in vitro diagnostic reagent monitoring image to obtain dark areas and bright areas of the image; Step S13: calculating the regional pixel brightness deviation of the dark area of ​​the image according to the bright area to obtain the pixel brightness deviation value of the dark area; Step S14: performing adaptive histogram equalization processing according to the brightness deviation value of the dark area pixels to obtain a brightness optimized monitoring image; Step S15: Perform multi-scale reconstruction on the brightness optimized monitoring image to generate a multi-scale noise reduction and reconstruction monitoring image.

3. The method for controlling the packaging quality of in vitro diagnostic reagent strips according to claim 2, characterized in that: The specific steps of step S15 are: Perform reagent edge detail detection on the brightness optimization monitoring image and mark multiple reagent edge details; Perform edge texture visual recognition on multiple reagent edge details to generate visual texture features of each reagent edge; Performing edge sharpening enhancement on the visual texture features of each reagent edge to generate a sharpening enhanced monitoring image; Perform multi-scale decomposition on the sharpened and enhanced monitoring image to obtain multiple sub-bands of different frequencies; Performing sub-band noise analysis on the multiple sub-bands of different frequencies one by one, thereby generating a noise feature of each frequency sub-band; Perform multi-band adaptive filtering and denoising on the noise characteristics of each frequency sub-band to obtain multiple filtering and denoising sub-bands; Multi-scale reconstruction is performed based on multiple filtered denoising sub-bands to generate a multi-scale denoising reconstructed monitoring image.

4. The method for controlling the packaging quality of in vitro diagnostic reagent strips according to claim 1, characterized in that: The specific steps of step S2 are: Step S21: performing deep semantic visual recognition on the multi-scale noise reduction and reconstruction monitoring image to obtain multiple in vitro diagnostic reagent packaging locations; Step S22: performing image visual segmentation on multiple in vitro diagnostic reagent packaging locations, thereby generating multiple packaging local sub-images; Step S23: performing iterative packaging defect detection on multiple packaging local sub-images one by one to mark the packaging defect locations; Step S24: Enlarge and extract the package defective part image, so as to obtain an enlarged monitoring image of the package defective part.

5. The method according to claim 1, characterized in that The specific steps of step S3 are: Step S31: performing a three-dimensional morphological analysis of the defective part on the enlarged monitoring image of the package defective part to generate a three-dimensional morphological feature of the defective part; Step S32: performing surface defect texture mining on the enlarged monitoring image of the package defect site to obtain surface defect texture data; Step S33: classify the defects according to the three-dimensional morphological features of the defect part and the surface defect texture data to obtain the current defect type label; Step S34: perform deep defect factor inference based on the current defect type label to generate packaging defect inference factors.

6. The method for controlling the packaging quality of in vitro diagnostic reagent strips according to claim 1, characterized in that: The specific steps of step S4 are: Step S41: Obtaining the reagent strip packaging log; Step S42: performing packaging process behavior analysis on the reagent strip packaging log, and extracting packaging process behavior data of multiple stages; Step S43: performing phased process logic analysis on packaging process behavior data of multiple phases, thereby generating packaging process logic rules; Step S44: performing multi-time point packaging process evolution according to packaging process logic rules, thereby constructing a sequential packaging process behavior sequence; Step S45: performing defect process tracing on the sequential packaging process behavior sequence according to the packaging defect inference factor, thereby marking the packaging process stage that causes the defect.

7. The method for controlling the packaging quality of in vitro diagnostic reagent strips according to claim 1, characterized in that: The specific steps of step S5 are: Step S51: calculating the geometric dimensions of the packaging defective part according to the enlarged monitoring image of the packaging defective part to generate geometric parameters of the packaging defective part; Step S52: Calculating the parameter error of the geometric parameters of the packaging defect part based on the preset reagent strip packaging parameter threshold, thereby obtaining the packaging defect part error parameter; Step S53: performing error compensation calculation on the error parameters of the package defect location to generate package defect compensation parameters.

8. The method for controlling the packaging quality of in vitro diagnostic reagent strips according to claim 1, characterized in that: The specific steps of step S6 are: Step S61: performing process parameter adjustment calculation for the current stage of the packaging process stage causing the defect according to the packaging defect compensation parameter, thereby obtaining multiple process parameter adjustment ranges; Step S62: performing packaging process parameter adjustment simulation based on multiple process parameter adjustment ranges to obtain multiple parameter adjustment simulation solutions; Step S63: Evaluate the packaging efficiency of multiple parameter adjustment simulation schemes to generate the packaging efficiency of each scheme; Step S64: performing an optimal adjustment range analysis according to the packaging efficiency of each solution, thereby obtaining an optimal process parameter adjustment range value; Step S65: fine-tune the process parameters of the packaging process production line based on the optimal process parameter adjustment amplitude value, and perform adaptive learning optimization to build an intelligent packaging process control model.

9. An in vitro diagnostic reagent strip packaging quality control system, characterized in that: The method for controlling the quality of in vitro diagnostic reagent strip packaging according to claim 1 comprises: A multi-scale reconstruction module is used to obtain a monitoring image of an in vitro diagnostic reagent after packaging; multi-scale reconstruction is performed on the monitoring image of the in vitro diagnostic reagent to generate a multi-scale noise reduction and reconstruction monitoring image; The packaging defect detection module is used to perform image visual segmentation on the multi-scale noise reduction and reconstruction monitoring image and iterative packaging defect detection on each image, thereby obtaining an enlarged monitoring image of the packaging defect part; The defect inference module is used to mine the surface defect texture of the enlarged monitoring image of the package defect site and infer the deep defect factors, thereby generating the package defect inference factors; The process tracing module is used to obtain the reagent strip packaging log; based on the packaging defect inference factors, the defective process of the reagent strip packaging log is traced, thereby marking the packaging process stage that caused the defect; A defect compensation module is used to calculate the error of the packaging part according to the enlarged monitoring image of the packaging defect part, and perform error compensation calculation to generate packaging defect compensation parameters; The packaging process optimization module is used to fine-tune the process parameters of the packaging process stage that causes the defects according to the packaging defect compensation parameters, and perform adaptive learning optimization to build an intelligent packaging process control model.

Images