A method for detecting sealing failure of biological products based on image recognition

Abstract

The present invention discloses a method for detecting sealing failure of biological products based on image recognition, and relates to the technical field of sealing detection. The method comprises: sealing the biological product by a first flexible sealing material and a second flexible sealing material and transmitting the sealing product to a first detection platform, and the biological product has a sealing area positioning coordinate on the first detection platform; activating a camera, and collecting the sealing area image information of the biological product based on the sealing area positioning coordinate; processing the sealing area image information by a defect identifier, and extracting the false defect area; performing texture consistency analysis on the false defect area to obtain the selected defect area; when the size specification of the selected defect area meets the failure size specification, marking the sealing failure of the biological product. The method solves the technical problem of poor accuracy of sealing failure detection of biological products in the prior art, and achieves the technical effect of improving detection efficiency and accuracy through image recognition technology.

Description

Technical Field

[0001] The present invention relates to the technical field of seal detection, and in particular to a method for detecting seal failure of biological products based on image recognition. Background Art

[0002] During the production and storage of biological products, the integrity of the seal is crucial to ensuring the safety, effectiveness and stability of the product. Biological products, including but not limited to drugs, vaccines, biotechnology products, etc., usually need to be stored and transported in a sterile and sealed environment to prevent microbial contamination and product degradation. However, defects in the sealing material or mistakes in the sealing process may cause the seal to fail, thereby affecting the quality and safety of the product. Traditional seal detection methods usually use defect recognition models for identification, which can identify defects in non-sealed junction areas. However, the sealing area is usually accompanied by complex wrinkles and textures. These characteristics make it difficult for general defect recognition models to accurately distinguish between real defects and normal structural features. For example, the natural wrinkles of the seal may be misjudged as defects, while real fine cracks or leaks may be missed due to the insensitivity of the model. Summary of the Invention

[0003] The present application provides a method for detecting sealing failure of biological products based on image recognition, which solves the technical problem of poor accuracy in the prior art of detecting sealing failure of biological products.

[0004] In view of the above problems, the present application provides a method for detecting sealing failure of biological products based on image recognition.

[0005] The present application provides a method for detecting sealing failure of biological products based on image recognition, the method comprising:

[0006] The biological product is sealed by a first flexible sealing material and a second flexible sealing material and transmitted to a first detection platform, wherein the biological product has a sealing area positioning coordinate on the first detection platform, wherein the sealing area is the sealing boundary area of ​​the first flexible sealing material and the second flexible sealing material; the camera is activated to collect image information of the sealing area of ​​the biological product based on the sealing area positioning coordinate; the sealing area image information is processed by a defect identifier to extract a false defect area; the texture consistency analysis is performed on the false defect area to obtain a selected defect area; when the size specification of the selected defect area meets the failure size specification, the biological product is marked as a sealing failure.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] First, the biological product is sealed by a first flexible sealing material and a second flexible sealing material and transmitted to a first detection platform. The biological product has a sealing area positioning coordinate on the first detection platform, wherein the sealing area is the sealing boundary area of ​​the first flexible sealing material and the second flexible sealing material. Next, the camera is activated to collect image information of the sealing area of ​​the biological product based on the sealing area positioning coordinates. Then, the sealing area image information is processed by a defect identifier to extract the false defect area. Finally, a texture consistency analysis is performed on the false defect area to obtain a selected defect area; when the size specification of the selected defect area meets the failure size specification, the biological product is marked as sealed. The technical problem of poor accuracy in the detection of sealing failure of biological products in the prior art is solved, and the technical effect of improving detection efficiency and accuracy is achieved through image recognition technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic flow chart of a method for detecting sealing failure of biological products based on image recognition provided in an embodiment of the present application;

[0011] Figure 2 A schematic diagram of the process of performing texture consistency analysis in the method for detecting sealing failure of biological products based on image recognition provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The present application solves the technical problem of poor accuracy in the prior art of biological product sealing failure detection by providing a biological product sealing failure detection method based on image recognition.

[0013] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0014] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0015] Examples, such as Figure 1 As shown, the embodiment of the present application provides a method for detecting sealing failure of biological products based on image recognition, wherein the method includes:

[0016] The biological product is sealed by a first flexible sealing material and a second flexible sealing material and transferred to a first detection platform. The biological product has a sealing area positioning coordinate on the first detection platform, wherein the sealing area is the sealing interface area between the first flexible sealing material and the second flexible sealing material.

[0017] During the seal testing process for biological products, the product is first sealed with a first flexible sealing material and a second flexible sealing material. After sealing is complete, the product is transferred to a dedicated first testing platform. On the first testing platform, each biological product is assigned precise seal area positioning coordinates, which indicate the location of the seal area at the interface between the first and second flexible sealing materials.

[0018] Furthermore, the biological product is sealed by the first flexible sealing material and the second flexible sealing material and transported to the first detection platform, wherein the biological product has a sealing area positioning coordinate on the first detection platform, including:

[0019] Obtain sealing control parameters of the first flexible sealing material and the second flexible sealing material for the biological product, wherein the sealing control parameters have coordinate feature identifiers; perform regional clustering analysis on the coordinate feature identifiers based on the sealing control parameters to obtain regional clustering results, wherein the same category of the regional clustering results has the same sealing control parameters; and add the regional clustering results into the sealing area positioning coordinates.

[0020] Specifically, the key control parameters for sealing biological products using the first and second flexible sealing materials are obtained. The sealing control parameters not only reflect the physical and chemical conditions of the sealing process, but also have coordinate feature identifiers, that is, their correspondence with the spatial position of the sealing area. The sealing control parameters with coordinate feature identifiers are used to perform regional clustering analysis to classify data points with similar characteristics into the same category. Optionally, based on the similarity of the sealing control parameters, K-means clustering is used to cluster different coordinate feature identifiers to form regional clustering results. Regional clustering results of the same category will have the same sealing control parameters. The regional clustering results are integrated into the sealing area positioning coordinates. By combining the regional clustering results with the sealing area positioning coordinates, the sealing effect can be more accurately evaluated.

[0021] The camera is activated to collect image information of the sealing area of ​​the biological product based on the positioning coordinates of the sealing area.

[0022] Activate a high-precision camera and use the seal area positioning coordinates to capture image information of the seal area of ​​the biological product. Image acquisition based on coordinate information ensures the accuracy and pertinence of the image data, laying a solid foundation for subsequent analysis and processing.

[0023] The image information of the sealing area is processed by a defect identifier to extract the false defect area.

[0024] During the seal inspection process of biological products, the image information of the sealing area collected by the camera will be sent to the defect identifier for in-depth processing, and then the areas with significant differences from the normal areas will be extracted, namely the false defect areas. The false defect areas do not mean that these areas are not defects, but require further verification and analysis to determine whether they fall into the category of defects.

[0025] Furthermore, the image information of the sealing area is processed by a defect identifier to extract the false defect area, including:

[0026] Obtain a defect texture set; obtain a positive sample sealing surface image of the biological product, wherein the positive sample sealing surface image has a sealing area wrinkle texture label; configure a defect compilation quantity constraint; according to the defect compilation quantity constraint, randomly compile the defect texture set to the positive sample sealing surface image, repeat several times, and obtain a negative sample sealing surface image, wherein the negative sample sealing surface image has a defect texture label; configure the defect identifier based on the positive sample sealing surface image and the negative sample sealing surface image as input data, and using the defect texture label and the sealing area wrinkle texture label as convolutional network output supervision data.

[0027] A series of defect texture images are collected from big data, representing various possible seal defects, such as damage, stains, and wrinkle abnormalities. These defect texture images are organized into a defect texture set for subsequent compilation into positive sample images. A series of biological product seal surface images are collected from big data as positive sample seal surface images. These positive sample images are annotated with seal region wrinkle texture labels to enable training to distinguish between normal and defective textures. Positive sample seal surface images should have a clear seal region with no defects or only normal wrinkle textures. Based on actual needs, a defect compilation constraint is configured, setting the number of defect textures compiled into the positive sample image each time to control the complexity and diversity of negative samples. Using image processing techniques, textures from the defect texture set are randomly compiled into the positive sample seal surface image, with each compilation adhering to the previously set defect compilation constraint. This compilation process is repeated several times to generate a sufficient number of negative sample seal surface images, which are annotated with defect texture labels to enable training to distinguish them from positive samples. A convolutional neural network (CNN) is constructed by using positive and negative sample sealing surface images as input data and defect texture labels and sealing area wrinkle texture labels as output supervision data. After multiple iterative training, a defect identifier is obtained, which can accurately identify defect textures in the sealing area of ​​biological products.

[0028] A texture consistency analysis is performed on the pseudo defect area to obtain a selected defect area.

[0029] After processing the seal area image information and extracting the false defect areas, texture consistency analysis is performed on these false defect areas to further screen out true defect areas. By comparing the texture features of different parts of the area, the consistency and differences between them are evaluated. If the majority of the texture features within the area are similar and conform to the texture pattern of a normal seal surface, the area may be a false defect that has been misidentified. Conversely, if there are significant texture changes or abnormal features within the area, it is more likely to be a true defect area.

[0030] Furthermore, if Figure 2 As shown, texture consistency analysis is performed on the pseudo defect area to obtain a selected defect area, including:

[0031] Performing a longitudinal texture consistency analysis on the pseudo defect area to obtain a longitudinal selected defect area; performing a transverse texture consistency analysis on the pseudo defect area to obtain a transverse selected defect area; and setting the union of the longitudinal selected defect area and the transverse selected defect area as the selected defect area.

[0032] Preferably, a texture consistency analysis is performed on each pseudo-defect region along its longitudinal direction (i.e., perpendicular to the sealing boundary line) to obtain a longitudinal selected defect region. A similar method is used to perform a texture consistency analysis on the pseudo-defect region in the transverse direction (i.e., parallel to the sealing boundary line) to obtain a transverse selected defect region. The union of the longitudinal selected defect region and the transverse selected defect region is defined as the final selected defect region. This process allows for a more comprehensive assessment of the texture consistency of pseudo-defect regions and more accurate identification of true defect regions.

[0033] Furthermore, performing longitudinal texture consistency analysis on the pseudo defect region to obtain a longitudinal selected defect region includes:

[0034] The pseudo defect area is placed into the positioning space to obtain the positioning coordinate information of the pseudo defect area; based on the positioning coordinate information of the pseudo defect area, the pseudo defect area is traversed to extract the longitudinal distribution length feature set of the pseudo defect; outlier analysis is performed according to the longitudinal distribution length feature set of the pseudo defect to obtain the first outlier pseudo defect; the first outlier pseudo defect is deleted from the positioning coordinate information of the pseudo defect area to obtain the updated positioning coordinate information of the pseudo defect area; according to the updated positioning coordinate information of the pseudo defect area, the adjacent pseudo defect areas in the longitudinal direction are traversed to extract the pseudo defect distribution spacing feature set, wherein the defect spacing is the longitudinal center distance of the defect; outlier analysis is performed according to the pseudo defect distribution spacing feature set to obtain the second outlier pseudo defect; the first outlier pseudo defect and the second outlier pseudo defect are added to the longitudinal selected defect area.

[0035] Specifically, each pseudo-defect area is mapped into the positioning space to obtain the positioning coordinate information of the pseudo-defect area; each pseudo-defect area is traversed to obtain the distribution length of the pseudo-defect area in the longitudinal direction (perpendicular to the sealing boundary line) to form a pseudo-defect longitudinal distribution length feature set; the longitudinal distribution length feature set is subjected to outlier analysis using statistical methods (such as box plots, Z-score, etc.) to identify those outliers that are significantly different from most data points, and the pseudo-defects corresponding to these points are regarded as first outlier pseudo-defects; the first outlier pseudo-defect is deleted from the pseudo-defect area positioning coordinate information to obtain updated coordinate information; based on the updated pseudo-defect area positioning coordinate information, adjacent pseudo-defect areas in the longitudinal direction are traversed to calculate the longitudinal center distance between adjacent defects as the defect spacing feature, and all defect spacing features are integrated to form a pseudo-defect distribution spacing feature set; the defect spacing feature set is subjected to outlier analysis to identify those outliers whose spacing is significantly different from other defects, and the pseudo-defects corresponding to these points are regarded as second outlier pseudo-defects; the first outlier pseudo-defect and the second outlier pseudo-defect are added to the longitudinal selected defect area. Through the above steps, the characteristics of the false defect area can be analyzed more finely, and outliers can be effectively identified and excluded, so as to more accurately determine the longitudinal selected defect area.

[0036] Furthermore, performing outlier analysis based on the pseudo-defect distribution spacing feature set to obtain a second outlier pseudo-defect includes:

[0037] An outlier analysis is performed based on the pseudo-defect distribution spacing feature set to obtain an outlier spacing feature set; when the first distribution spacing feature and the second distribution spacing feature of the pseudo-defect to be analyzed both belong to the outlier spacing feature set, the pseudo-defect to be analyzed is added to the second outlier pseudo-defect.

[0038] Preferably, the pseudo-defect distribution spacing feature set is analyzed for outliers using statistical methods (such as box plots, IQR, etc.) to identify points that are significantly different from most spacing features. The spacing values ​​corresponding to these points are regarded as outlier spacing features, and all outlier spacing features are integrated into a set, namely the outlier spacing feature set; for each pseudo-defect to be analyzed, it is necessary to calculate the first distribution spacing feature and the second distribution spacing feature of the pseudo-defect, the first distribution spacing feature is the longitudinal center distance between the defect and its previous adjacent defect, and the second distribution spacing feature is the longitudinal center distance between the defect and its next adjacent defect; the first distribution spacing feature and the second distribution spacing feature of the pseudo-defect to be analyzed are compared with the outlier spacing feature set. If both spacing features exist in the outlier spacing feature set, that is, they both significantly deviate from the normal spacing range, then it can be considered that the pseudo-defect to be analyzed has abnormal spacing performance and is likely to be a real defect. Therefore, the pseudo-defect to be analyzed is added to the second outlier pseudo-defect set.

[0039] When the size specification of the selected defect area meets the failure size specification, the biological product is marked as sealed failure.

[0040] After completing a detailed analysis and confirmation of the selected defect areas, it is possible to assess whether these defects have a real impact on the sealing performance of the biopharmaceutical. Specifically, a series of failure size specifications are set based on the characteristics and usage requirements of the biopharmaceutical. These specifications define which size and shape of defects will cause sealing failure. Based on the preset failure size specifications, the size specifications of the selected defect area are evaluated. If the size specifications of the selected defect area meet or exceed the preset failure size specifications, the biopharmaceutical will be immediately marked as sealing failure.

[0041] Furthermore, when the size specification of the selected defect area meets the failure size specification, the biological product is marked as sealed failure, including:

[0042] Extracting a set of color feature values ​​of the selected defect area, wherein the color feature value set has a set of color feature value distribution coordinates; fitting the color feature value set through a defect color calibration network to obtain a defect color feature value set; distributing the defect color feature value set according to the color feature value distribution coordinates to obtain a selected updated defect area; when the size specification of the selected updated defect area meets the failure size specification, marking the biological product as sealed and failed.

[0043] Preferably, color analysis is performed on the pixels in the selected defect area to extract color feature values, which may include RGB values ​​and HSV values. These color feature values ​​are combined into a set, and the distribution coordinates of each color feature value in the defect area are recorded at the same time to form a color feature value distribution coordinate set; a pre-trained defect color calibration network (such as a convolutional neural network CNN based on a deep learning model) is used to fit the extracted color feature value set. The defect color calibration network can identify color feature patterns that are highly correlated with defects and filter out a more accurate defect color feature value set from the original color feature value set; using color calibration to obtain the color feature pattern, the color feature value set is used to obtain the color feature pattern. The color feature value distribution coordinate set is used to remap the filtered defect color feature values ​​back to the original image space to form a selected updated defect area, which more accurately reflects the location and shape of the actual defect. The dimensions of the selected updated defect area, including parameters such as length, width, and area, are measured and compared with the preset failure dimensions. If any dimension of the selected updated defect area exceeds the failure dimension threshold, the defect is considered to have a real impact on the sealing performance of the biopharmaceutical. If the dimension of the selected updated defect area meets the failure dimension, the biopharmaceutical is marked as sealed according to the previous method. Through the above process, color features can be used to further refine and update the selected defect area, thereby improving the accuracy and reliability of seal detection.

[0044] Furthermore, fitting the color feature value set through the defect color calibration network to obtain the defect color feature value set includes:

[0045] A set of color feature values ​​of non-sealing failure defects and a set of color feature values ​​of sealing failure defects are collected to construct a classifier input data set; and the defect color calibration network is configured according to the classifier input data set.

[0046] From samples of biological products with non-sealing failures, regions containing various non-sealing failure defects (such as scratches, stains, and color differences) are selected. Color feature values ​​are extracted from these regions. These color feature values ​​and their corresponding labels (non-sealing failure) form a non-sealing failure defect color feature value set, which serves as part of the classifier input dataset. From samples of biological products with sealing failures, regions containing sealing failure defects (such as holes, cracks, and bubbles) are selected. Similarly, color feature values ​​are extracted from these regions. These color feature values ​​and their corresponding labels (sealing failure) form a sealing failure defect color feature value set, which serves as another part of the classifier input dataset. The non-sealing failure defect color feature value set and the sealing failure defect color feature value set are merged to form a complete classifier input dataset. The constructed classifier input dataset is used to train a defect color calibration network. The defect color calibration network is built based on a convolutional neural network and can output a defect type (non-sealing failure or sealing failure) based on color feature value input. Through the above steps, a defect color calibration network based on color feature values ​​can be constructed to identify sealing failure defects in biological products.

[0047] In summary, the embodiments of the present application have at least the following technical effects:

[0048] First, the biological product is sealed by a first flexible sealing material and a second flexible sealing material and transmitted to a first detection platform. The biological product has a sealing area positioning coordinate on the first detection platform, wherein the sealing area is the sealing boundary area of ​​the first flexible sealing material and the second flexible sealing material. Next, the camera is activated to collect image information of the sealing area of ​​the biological product based on the sealing area positioning coordinates. Then, the sealing area image information is processed by a defect identifier to extract the false defect area. Finally, a texture consistency analysis is performed on the false defect area to obtain a selected defect area; when the size specification of the selected defect area meets the failure size specification, the biological product is marked as sealed. The technical problem of poor accuracy in the detection of sealing failure of biological products in the prior art is solved, and the technical effect of improving detection efficiency and accuracy is achieved through image recognition technology.

[0049] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0050] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

[0051] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.

Claims

1. A method for detecting sealing failure of biological products based on image recognition, characterized in that: The method comprises: The biological product is sealed by a first flexible sealing material and a second flexible sealing material and transmitted to a first detection platform, wherein the biological product has a sealing area positioning coordinate on the first detection platform, wherein the sealing area is a sealing interface area between the first flexible sealing material and the second flexible sealing material; activating a camera to collect image information of the sealing area of ​​the biological product based on the sealing area positioning coordinates; Processing the image information of the sealing area by a defect identifier to extract a false defect area; The pseudo-defect area refers to an area that is significantly different from a normal area; Performing texture consistency analysis on the pseudo defect area to obtain a selected defect area; When the size specification of the selected defect area meets the failure size specification, marking the biological product as sealed failure; The image information of the sealing area is processed by a defect identifier to extract the false defect area, including: Get defect texture set; Obtaining a positive sample sealing surface image of the biological product, wherein the positive sample sealing surface image has a sealing area wrinkle texture label; Configure defect compilation quantity constraints; According to the defect compilation quantity constraint, randomly compile the defect texture set to the positive sample sealed surface image, repeat several times, and obtain a negative sample sealed surface image, wherein the negative sample sealed surface image has a defect texture label; According to the positive sample sealing surface image and the negative sample sealing surface image as input data, the defect texture label and the sealing area wrinkle texture label are used as convolutional network output supervision data to configure the defect identifier; The pseudo defect area is subjected to texture consistency analysis to obtain a selected defect area, including: Performing longitudinal texture consistency analysis on the pseudo defect area to obtain a longitudinal selected defect area; Performing a transverse texture consistency analysis on the pseudo defect area to obtain a transverse selected defect area; The union of the longitudinal selected defect region and the transverse selected defect region is set as the selected defect region; The pseudo defect area is subjected to longitudinal texture consistency analysis to obtain a longitudinal selected defect area, including: The pseudo defect area is placed into the positioning space to obtain the positioning coordinate information of the pseudo defect area; Based on the positioning coordinate information of the pseudo defect area, traverse the pseudo defect area to extract a pseudo defect longitudinal distribution length feature set; Performing outlier analysis based on the pseudo-defect longitudinal distribution length feature set to obtain a first outlier pseudo-defect; Deleting the first outlier false defect from the false defect area positioning coordinate information to obtain false defect area positioning updated coordinate information; According to the positioning and updating coordinate information of the pseudo defect area, traversing the adjacent pseudo defect areas in the longitudinal direction, and extracting the pseudo defect distribution spacing feature set, wherein the defect spacing is the longitudinal center distance of the defect; Performing outlier analysis according to the pseudo-defect distribution spacing feature set to obtain a second outlier pseudo-defect; Adding the first outlier false defect and the second outlier false defect into the longitudinal selected defect area; The pseudo defect distribution spacing feature set is subjected to outlier analysis to obtain a second outlier pseudo defect, including: Performing outlier analysis based on the pseudo-defect distribution spacing feature set to obtain an outlier spacing feature set; When the first distribution spacing feature and the second distribution spacing feature of the pseudo defect to be analyzed both belong to the outlier spacing feature set, adding the pseudo defect to be analyzed into the second outlier pseudo defect; The first distribution spacing feature refers to the longitudinal center distance between the pseudo defect to be analyzed and its previous adjacent defect; The second distribution spacing feature refers to the longitudinal center distance between the pseudo defect to be analyzed and its subsequent adjacent defect.

2. The method for detecting sealing failure of biological products based on image recognition according to claim 1, characterized in that: The biological product is sealed by a first flexible sealing material and a second flexible sealing material and transmitted to a first detection platform, wherein the biological product has a sealing area positioning coordinate on the first detection platform, comprising: Obtaining sealing control parameters of the first flexible sealing material and the second flexible sealing material for the biological product, wherein the sealing control parameters have coordinate feature identifiers; According to the sealing control parameters, performing regional clustering analysis on the coordinate feature identifiers to obtain regional clustering results, wherein the same category of the regional clustering results has the same sealing control parameters; The regional clustering results are added to the sealing area positioning coordinates.

3. The method for detecting sealing failure of biological products based on image recognition according to claim 1, characterized in that: When the size specification of the selected defect area meets the failure size specification, the biological product is marked as sealed failure, including: Extracting a color feature value set of the selected defect area, wherein the color feature value set has a color feature value distribution coordinate set; Fitting the color feature value set through a defect color calibration network to obtain a defect color feature value set; Distributing the defect color feature value set according to the color feature value distribution coordinate set to obtain a selected updated defect area; When the size specification of the selected updated defect area meets the failure size specification, the biological product is marked as sealed failure.

4. The method for detecting sealing failure of biological products based on image recognition according to claim 3, characterized in that: The color feature value set is fitted by a defect color calibration network to obtain a defect color feature value set, including: Collecting a set of color feature values ​​of non-sealing failure defects and a set of color feature values ​​of sealing failure defects to construct a classifier input data set; The defect color calibration network is configured according to the classifier input data set.

Images