Battery cell appearance inspection control system and method based on machine vision

Abstract

The invention discloses a battery cell appearance inspection control system and method based on machine vision, relates to the technical field of machine vision, and solves the problems that the quality of a product detection image is ignored in the prior art, so that the image definition cannot meet the requirement of accurately identifying defects due to the adoption of a light source with the same intensity in the detection image; the technical problem that the appearance inspection efficiency of the battery cell is relatively low is solved. The light source intensity is adaptively adjusted according to the product data and the environment data; inputting a detection image into the appearance recognition model to obtain a category label and a defect position and a defect size corresponding to the category label, calculating an image defect anomaly coefficient according to the category label, and then calculating an appearance defect anomaly coefficient; according to the invention, the alarm signal is generated according to the appearance defect abnormal coefficient, the cell products are classified according to the alarm signal, and the light source required by the detection image is adaptively adjusted, so that the quality of the detection image is higher, the abnormal condition of the product defect is considered from multiple angles, and the accuracy and efficiency of cell appearance abnormity identification are improved.

Description

Technical Field

[0001] This application belongs to the technical field of machine vision, and specifically relates to a control system and method for inspecting the appearance of battery cells based on machine vision. Background Art

[0002] As an important component of a battery, a battery cell has a wide range of application fields and important value. A battery cell refers to a single electrochemical cell containing positive and negative electrodes, which is an important part of a battery. Generally, it is not used directly, but is connected in series or parallel singly or in multiple numbers to form a battery. It is mainly divided into three types: aluminum shell battery cells, soft-pack battery cells, and cylindrical battery cells. With the continuous progress of technology, the performance of battery cells will continue to improve, making greater contributions to the development of human society.

[0003] Traditional inspection of the appearance of battery cells mainly relies on manual operation, which is not only inefficient, but also the detection results are easily affected by subjective factors, making it difficult to ensure consistency. When detecting product appearance defects through machine vision, the existing technology often ignores the quality of the product detection images. Since the light source intensity in different directions will affect the quality of the detection images, if the same intensity light source is used, the clarity of the collected images may not meet the requirements for accurately identifying defects, resulting in a relatively low efficiency of inspecting the appearance of battery cells. Therefore, the control system for inspecting the appearance of battery cells still needs further improvement. Summary of the Invention

[0004] This application aims to solve at least one of the technical problems existing in the prior art. For this purpose, this application proposes a control system and method for inspecting the appearance of battery cells based on machine vision, which is used to solve the technical problem that the existing technology ignores the quality of the product detection images, resulting in the image clarity not meeting the requirements for accurately identifying defects when using the same intensity light source for the detection images, and further leading to a relatively low efficiency of inspecting the appearance of battery cells.

[0005] To achieve the above object, the first aspect of this application provides a control system for inspecting the appearance of battery cells based on machine vision, including: a data acquisition module, a data analysis module, a control module, an early warning module, and a database;

[0006] The data acquisition module: obtains product data and environmental data of the product to be detected through data acquisition devices; obtains a number of detection images through the adjusted light source intensity setting value and image acquisition devices;

[0007] The data analysis module: adaptively adjusts the light source intensity according to product data and environmental data; inputs the detected image into the appearance recognition model to obtain the category label, the corresponding defect position and defect size, and the appearance recognition model is constructed through the YOLO network model; calculates the image defect anomaly coefficient and then calculates the appearance defect anomaly coefficient according to the category label, the corresponding defect position and defect size; generates an alarm signal according to the appearance defect anomaly coefficient;

[0008] The control module: assigns the adjusted light source intensity to the light source intensity set value; classifies the battery cell products according to the alarm signal;

[0009] The early warning module: makes a prompt according to the alarm signal and contacts the management personnel.

[0010] In this application, the light source intensity is adaptively adjusted according to product data and environmental data; the detected image is input into the appearance recognition model to obtain the category label, the corresponding defect position and defect size; the image defect anomaly coefficient is calculated and then the appearance defect anomaly coefficient is calculated according to the category label, the corresponding defect position and defect size; an alarm signal is generated according to the appearance defect anomaly coefficient, and the battery cell products are classified according to the alarm signal, and the light source required for the detected image is adaptively adjusted, so that the quality of the detected image is higher, and at the same time, the abnormal situation of product defects is considered from multiple angles, improving the accuracy and efficiency of the recognition of battery cell appearance anomalies.

[0011] Further, the adaptive adjustment of the light source intensity according to product data and environmental data includes:

[0012] Obtain the appearance material, reflectivity and color data in the product data and the environmental brightness in the environmental data, and integrate them into comprehensive data;

[0013] Input the comprehensive data into the light source intensity prediction model to obtain the light source intensity prediction value, and the light source intensity prediction model is constructed through an artificial intelligence model;

[0014] Use the light source intensity prediction value as the adjusted light source intensity.

[0015] In this application, the light source intensity for collecting detected images in each direction is adaptively adjusted according to the appearance material, reflectivity and color data in the product data and the environmental brightness in the environmental data, so that the detected images in each direction have higher quality and more detailed texture expression, improving the accuracy of appearance defect recognition.

[0016] Further, the construction of the light source intensity prediction model through an artificial intelligence model includes:

[0017] Obtain a number of historical appearance materials, reflectivity, color data, environmental brightness, and their corresponding light source intensities, and integrate the historical appearance materials, reflectivity, color data, and environmental brightness into historical comprehensive data;

[0018] Divide the historical comprehensive data and its corresponding light source intensities into training data, validation data, and test data;

[0019] Perform data preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set;

[0020] Use the training set to train the artificial intelligence model, adjust the parameters on the validation set, and perform model testing on the test data. Finally, obtain a light source intensity prediction model with the comprehensive data as the input and the predicted value of the light source intensity as the output.

[0021] Furthermore, the appearance recognition model is constructed through the YOLO network model, including:

[0022] Obtain a number of historical detection images and their corresponding class labels, defect positions, and defect sizes;

[0023] Divide the historical detection images and their corresponding class labels, defect positions, and defect sizes into training data, validation data, and test data;

[0024] Perform image preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set;

[0025] Use the training set to train the YOLO network model, adjust the parameters on the validation set, and perform model testing on the test data. Finally, obtain an appearance recognition model with the detection image as the input and the class label and its corresponding defect position and defect size as the output.

[0026] Furthermore, calculating the appearance defect anomaly coefficient after calculating the image defect anomaly coefficient according to the class label and its corresponding defect position and defect size includes:

[0027] Obtain the class label and its corresponding defect position and defect size;

[0028] Calculate the image defect anomaly coefficient TQYXk according to the class label and its corresponding defect position and defect size;

[0029] Calculate the appearance defect anomaly coefficient WQYX through the formula WQYX = ∑(TQYXk); where k represents the serial number of the appearance surface that needs to be detected on the cell appearance, k = 1, 2,..., K; K represents the total number of appearance surfaces that need to be detected on the cell.

[0030] After calculating the image defect anomaly coefficient based on the category label of the defect, its corresponding defect location, and defect size, this application calculates the appearance defect anomaly coefficient, comprehensively considering the defect anomaly of the product appearance from multiple aspects, and improving the accuracy of product appearance defect detection.

[0031] Further, calculating the image defect anomaly coefficient according to the category label, its corresponding defect location, and defect size includes:

[0032] Obtain the defect severity level QYDi corresponding to several category labels, the defect location influence level QWYDi corresponding to the defect location, and the defect size QDi;

[0033] Calculate the image defect anomaly coefficient TQYX through the formula TQYX = ∑(1 - exp(-QYDi × (α1 × QWYDi + α2 × QDi))); where i represents the number serial number corresponding to the category label in the image, i = 1, 2,..., N; N represents the total number of category labels; α1 and α2 are weight coefficients, and α1, α2 ∈ (0, 1).

[0034] Further, generating an alarm signal according to the appearance defect anomaly coefficient includes:

[0035] Obtain the appearance defect anomaly coefficient and the image defect anomaly coefficient;

[0036] Judge whether the appearance defect anomaly coefficient is greater than the appearance anomaly threshold; if yes, generate an alarm signal for the overall appearance defect of the product; if not, judge whether there is an image defect anomaly coefficient greater than the image anomaly threshold; if yes, generate an alarm signal for the single-sided defect of the product; if not, do nothing.

[0037] Another aspect of the present invention provides a control method for the appearance inspection of battery cells based on machine vision, including:

[0038] S0: Obtain the product data and environmental data of the product to be detected; adaptively adjust the light source intensity according to the product data and environmental data, and assign the adjusted light source intensity to the light source intensity setting value;

[0039] S1: Obtain several detection images through the adjusted light source intensity setting value; input the detection images into the appearance recognition model to obtain the category label, its corresponding defect location, and defect size;

[0040] S2: Calculate the image defect anomaly coefficient according to the category label, its corresponding defect location, and defect size, and then calculate the appearance defect anomaly coefficient;

[0041] S3: Generate an alarm signal according to the appearance defect anomaly coefficient, and classify and process the battery cell products according to the alarm signal;

[0042] S4: Make a prompt according to the alarm signal and contact the management personnel.

[0043] Compared with the prior art, the beneficial effects of this application are as follows:

[0044] 1. This application adaptively adjusts the light source intensity according to product data and environmental data; inputs the detected image into the appearance recognition model to obtain the category label and its corresponding defect position and defect size; calculates the image defect anomaly coefficient and then the appearance defect anomaly coefficient according to the category label and its corresponding defect position and defect size; generates an alarm signal according to the appearance defect anomaly coefficient, classifies the battery cell products according to the alarm signal, and adaptively adjusts the light source required for the detected image, making the quality of the detected image higher. At the same time, considering the abnormal situation of product defects from multiple angles, the accuracy and efficiency of the abnormal recognition of the battery cell appearance are improved.

[0045] 2. This application adaptively adjusts the light source intensity when collecting detected images in each direction according to the appearance material, reflectivity, and color data in the product data and the environmental brightness in the environmental data, making the detected images in each direction have higher quality and more detailed texture expression, and improving the accuracy of appearance defect recognition.

[0046] 3. This application calculates the image defect anomaly coefficient and then the appearance defect anomaly coefficient according to the category label of the defect and its corresponding defect position and defect size, comprehensively considering the abnormal situation of product appearance defects in multiple aspects, and improving the accuracy of product appearance defect detection. Description of the Drawings

[0047] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0048] Figure 1 It is a schematic diagram of the principle of a battery cell appearance inspection control system based on machine vision for this application;

[0049] Figure 2 It is a flowchart of a battery cell appearance inspection control method based on machine vision for this application. Detailed Embodiments

[0050] The technical solution of the present application will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.

[0051] Please refer to Figure 1 , an embodiment of the first aspect of the present application provides a control system for inspecting the appearance of battery cells based on machine vision, including: a data acquisition module, a data analysis module, a control module, an early warning module, and a database;

[0052] Data acquisition module: Obtain product data and environmental data of the product to be detected through data acquisition devices; obtain a number of detection images through the adjusted light source intensity setting value and image acquisition devices; the data acquisition devices include high-precision cameras and various sensors, etc.; the product data includes appearance materials, reflectivity, and color data, etc., and the environmental data includes environmental brightness, etc.;

[0053] Data analysis module: Adaptively adjust the light source intensity according to the product data and environmental data; input the detection images into the appearance recognition model to obtain the category label and its corresponding defect position and defect size, and the appearance recognition model is constructed through the YOLO network model; calculate the image defect anomaly coefficient and then calculate the appearance defect anomaly coefficient according to the category label and its corresponding defect position and defect size. The image defect anomaly coefficient refers to the abnormal degree of the product appearance in the detection image, and the appearance defect anomaly coefficient refers to the abnormal degree of the entire product appearance; generate an alarm signal according to the appearance defect anomaly coefficient;

[0054] Control module: Assign the adjusted light source intensity to the light source intensity setting value, where the light source intensity setting value refers to the setting value of each light source; classify the battery cell products according to the alarm signal;

[0055] Early warning module: Make a prompt according to the alarm signal and contact the management personnel. The alarm signals include product overall appearance defect alarm signals and product single-sided defect alarm signals, etc.

[0056] In this embodiment, the light source intensity is adaptively adjusted according to the product data and environmental data; the detection images are input into the appearance recognition model to obtain the category label and its corresponding defect position and defect size; the image defect anomaly coefficient is calculated and then the appearance defect anomaly coefficient is calculated according to the category label and its corresponding defect position and defect size; an alarm signal is generated according to the appearance defect anomaly coefficient, and the battery cell products are classified according to the alarm signal. The light source required for the detection images is adaptively adjusted, so that the quality of the detection images is higher. At the same time, the abnormal conditions of product defects are considered from multiple angles, improving the accuracy and efficiency of the recognition of battery cell appearance anomalies.

[0057] In this embodiment, adaptively adjusting the light source intensity according to product data and environmental data includes:

[0058] Obtaining the appearance material, reflectivity, and color data in the product data and the environmental brightness in the environmental data, and integrating them into comprehensive data;

[0059] Inputting the comprehensive data into a light source intensity prediction model to obtain a predicted value of the light source intensity. The light source intensity prediction model is constructed through an artificial intelligence model;

[0060] Using the predicted value of the light source intensity to adjust the light source intensity.

[0061] In this embodiment, by adaptively adjusting the light source intensity when collecting detection images in each direction according to the appearance material, reflectivity, and color data in the product data and the environmental brightness in the environmental data, the detection images in each direction have higher quality and more detailed texture expression, improving the accuracy of appearance defect recognition.

[0062] The light source intensity prediction model in this embodiment is constructed through an artificial intelligence model, including:

[0063] Obtaining a number of historical appearance materials, reflectivities, color data, and environmental brightnesses and their corresponding light source intensities, and integrating the historical appearance materials, reflectivities, color data, and environmental brightnesses into historical comprehensive data;

[0064] Dividing the historical comprehensive data and its corresponding light source intensity into training data, validation data, and test data;

[0065] Performing data preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set; the ratio among the training set, the test set, and the validation set is 7:2:1;

[0066] Using the training set to train the artificial intelligence model, adjusting parameters on the validation set, and performing model testing on the test data, finally obtaining a light source intensity prediction model with the comprehensive data as the input and the predicted value of the light source intensity as the output; the artificial intelligence model includes a BP neural network model and an RBF neural network model.

[0067] The appearance recognition model in this embodiment is constructed through a YOLO network model, including:

[0068] Obtaining a number of historical detection images and their corresponding class labels, defect positions, and defect sizes;

[0069] Dividing the historical detection images and their corresponding class labels, defect positions, and defect sizes into training data, validation data, and test data;

[0070] Perform image preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set; the ratio among the training set, the test set, and the validation set is 7:2:1;

[0071] Use the training set to train the YOLO network model, adjust the parameters on the validation set, and conduct model testing on the test data. Finally, obtain an appearance recognition model with the input being the detection image and the output being the class label and its corresponding defect location and defect size.

[0072] In this embodiment, after calculating the image defect abnormality coefficient based on the class label and its corresponding defect location and defect size, calculate the appearance defect abnormality coefficient, including:

[0073] Obtain the class label and its corresponding defect location and defect size;

[0074] Calculate the image defect abnormality coefficient TQYXk based on the class label and its corresponding defect location and defect size;

[0075] Calculate the appearance defect abnormality coefficient WQYX through the formula WQYX = ∑(TQYXk); where k represents the serial number of the appearance surface that needs to be detected on the cell appearance, k = 1, 2,..., K; K represents the total number of appearance surfaces that need to be detected on the cell; since the cell has various three-dimensional shapes and a single detection image cannot cover the entire cell surface, defect abnormality detection is performed on multiple detection images simultaneously; therefore, the appearance defect abnormality coefficient increases as the image defect abnormality coefficient increases.

[0076] In this embodiment, after calculating the image defect abnormality coefficient based on the class label of the defect and its corresponding defect location and defect size, calculate the appearance defect abnormality coefficient, comprehensively considering the defect abnormality of the product appearance from multiple aspects, and improving the accuracy of product appearance defect detection.

[0077] In this embodiment, calculating the image defect abnormality coefficient based on the class label and its corresponding defect location and defect size includes:

[0078] Obtain the defect severity level QYDi corresponding to several class labels, the defect location influence level QWYDi corresponding to the defect location, and the defect size QDi; the defect severity level refers to the degree of adverse impact on the product caused by different defect categories, and the defect location influence level refers to the degree of adverse impact on the product caused by defects in different locations;

[0079] Calculate the image defect anomaly coefficient TQYX through the formula TQYX = ∑(1 - exp(-QYDi × (α1 × QWYDi + α2 × QDi))); where i represents the numbering corresponding to the category label in the image, i = 1, 2, …, N; N represents the total number corresponding to the category label; α1 and α2 are weight coefficients, α1 and α2 ∈ (0, 1), and the specific values are set according to experience; defects of different categories have the same impact, and at the same time, defects of different positions and sizes also have different impacts; therefore, the image defect anomaly coefficient increases with the increase in the number of defects, the severity level of defects, the impact level of defect positions, and the size of defects.

[0080] Generating an alarm signal according to the appearance defect anomaly coefficient in this embodiment includes:

[0081] Obtain the appearance defect anomaly coefficient and the image defect anomaly coefficient;

[0082] Judge whether the appearance defect anomaly coefficient is greater than the appearance anomaly threshold, and the appearance anomaly threshold is set according to experience; if yes, generate an alarm signal for the overall appearance defect of the product; if not, judge whether there is an image defect anomaly coefficient greater than the image anomaly threshold, and the image anomaly threshold is set according to experience; if yes, generate an alarm signal for the single-sided defect of the product; if not, do nothing.

[0083] Please refer to Figure 2 , another embodiment of the present application provides a method for controlling the appearance inspection of battery cells based on machine vision, including:

[0084] S0: Obtain the product data and environmental data of the product to be detected; adaptively adjust the light source intensity according to the product data and environmental data, and assign the adjusted light source intensity to the light source intensity setting value;

[0085] S1: Obtain a number of detection images through the adjusted light source intensity setting value; input the detection images into the appearance recognition model to obtain the category label and its corresponding defect position and defect size;

[0086] S2: Calculate the image defect anomaly coefficient according to the category label and its corresponding defect position and defect size, and then calculate the appearance defect anomaly coefficient;

[0087] S3: Generate an alarm signal according to the appearance defect anomaly coefficient, and classify the battery cell products according to the alarm signal;

[0088] S4: Make a prompt according to the alarm signal and contact the management personnel.

[0089] Some of the data in the above formula is the numerical value obtained by removing the dimension. The formula is the one that is closest to the actual situation obtained through software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0090] The working principle of this application: Obtain the product data and environmental data of the product to be detected through a data acquisition device; adaptively adjust the light source intensity according to the product data and environmental data, and assign the adjusted light source intensity to the light source intensity setting value; obtain a number of detection images through the adjusted light source intensity setting value and an image acquisition device; input the detection images into an appearance recognition model to obtain the category label and its corresponding defect position and defect size; calculate the image defect anomaly coefficient and then calculate the appearance defect anomaly coefficient according to the category label and its corresponding defect position and defect size; generate an alarm signal according to the appearance defect anomaly coefficient; classify the battery cell products according to the alarm signal; make a prompt according to the alarm signal, contact the management personnel, and adaptively adjust the light source required for the detection images, so that the quality of the detection images is higher. At the same time, considering the abnormal situation of product defects from multiple angles, the accuracy and efficiency of the battery cell appearance abnormality recognition are improved, and the problem that the prior art ignores the quality of product detection images, resulting in the image clarity not meeting the requirements for accurate defect recognition when using the same intensity light source for detection images, and further leading to low efficiency of battery cell appearance inspection is avoided.

[0091] The above embodiments are only used to illustrate the technical method of this application and not to limit it. Although this application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of this application can be modified or equivalently replaced without departing from the spirit and scope of the technical method of this application.

Claims

1. A battery cell appearance inspection control system based on machine vision, characterized in that: include: Data acquisition module, data analysis module, control module, early warning module and database; The data acquisition module: acquires product data and environmental data of the inspected product through a data acquisition device; acquires a number of inspection images through an adjusted light source intensity setting value and an image acquisition device; The data analysis module: adaptively adjusts the light source intensity according to the product data and the environmental data; inputs the detected image into the appearance recognition model to obtain the category label and its corresponding defect position and defect size, and the appearance recognition model is constructed by the YOLO network model; calculates the image defect anomaly coefficient according to the category label and its corresponding defect position and defect size, and then calculates the appearance defect anomaly coefficient; Generate an alarm signal based on the appearance defect abnormality coefficient; The control module: assigns the adjusted light source intensity to the light source intensity setting value; and classifies the battery products according to the alarm signal.

2. The battery cell appearance inspection control system based on machine vision according to claim 1, characterized in that: The adaptively adjusting the light source intensity according to the product data and the environment data includes: Obtain appearance material, reflectivity and color data from product data and ambient brightness from environmental data, and integrate them into comprehensive data; Inputting the comprehensive data into a light source intensity estimation model to obtain a light source intensity estimation value, wherein the light source intensity estimation model is constructed by an artificial intelligence model; Use the estimated light intensity as the basis for adjusting the light intensity.

3. The battery appearance inspection control system based on machine vision according to claim 2, characterized in that: The light source intensity estimation model is constructed by an artificial intelligence model, including: Obtaining a number of historical appearance materials, reflectivity, color data, and ambient brightness as well as their corresponding light source intensities, and integrating the historical appearance materials, reflectivity, color data, and ambient brightness into historical comprehensive data; Divide the historical comprehensive data and its corresponding light source intensity into training data, verification data and test data; Perform data preprocessing on training data, verification data and test data to obtain training set, verification set and test set; The artificial intelligence model is trained using the training set, the parameters are adjusted on the validation set, and the model is tested on the test data. Finally, a light source intensity estimation model is obtained, whose input is comprehensive data and output is the estimated value of the light source intensity.

4. The battery cell appearance inspection control system based on machine vision according to claim 1, characterized in that: The appearance recognition model is constructed through the YOLO network model, including: Obtain several historical inspection images and their corresponding category labels, defect locations and defect sizes; Divide historical inspection images and their corresponding category labels, defect locations, and defect sizes into training data, verification data, and test data; Perform image preprocessing on training data, verification data and test data to obtain training set, verification set and test set; The YOLO network model is trained using the training set, the parameters are adjusted on the validation set, and the model is tested on the test data. Finally, an appearance recognition model is obtained whose input is the detection image and whose output is the category label and its corresponding defect location and defect size.

5. The battery cell appearance inspection control system based on machine vision according to claim 1, characterized in that: The method of calculating the image defect anomaly coefficient according to the category label and its corresponding defect position and defect size and then calculating the appearance defect anomaly coefficient includes: Get the category label and its corresponding defect location and defect size; Calculate the image defect anomaly coefficient TQYXk according to the category label and its corresponding defect position and defect size; The appearance defect anomaly coefficient WQYX is calculated by summing up all the image defect anomaly coefficients TQYXk; wherein k represents the appearance surface number of the battery cell that needs to be inspected, k=1, 2, ..., K; and K represents the total number of appearance surfaces of the battery cell that needs to be inspected.

6. The battery cell appearance inspection control system based on machine vision according to claim 5, characterized in that: The method of calculating the image defect anomaly coefficient according to the category label and its corresponding defect position and defect size includes: Obtain defect severity levels QYDi corresponding to several category labels and defect location impact levels QWYDi and defect sizes QDi corresponding to their defect locations; The image defect anomaly coefficient TQYX is calculated by the formula TQYX=∑(1-exp(-QYDi×(α1×QWYDi+α2×QDi))); wherein i represents the number of category labels in the image, i=1, 2,…, N; N represents the total number of category labels; α1 and α2 are weight coefficients, α1 and α2∈(0, 1).

7. The battery appearance inspection control system based on machine vision according to claim 1, characterized in that: The generating of an alarm signal according to the appearance defect abnormality coefficient comprises: Obtain appearance defect abnormality coefficient and image defect abnormality coefficient; Determine whether the appearance defect abnormality coefficient is greater than the appearance abnormality threshold; Yes, generate an alarm signal for overall product appearance defects; If no, determine whether there is an image defect abnormality coefficient greater than the image abnormality threshold; if yes, generate a product single-side defect alarm signal; if no, do nothing.

8. A method for controlling the appearance inspection of a battery cell based on machine vision, applied to a control system for the appearance inspection of a battery cell based on machine vision according to any one of claims 1 to 7, characterized in that: include: S0: Acquire product data and environmental data of the product under inspection; adaptively adjust the light source intensity according to the product data and environmental data, and assign the adjusted light source intensity to the light source intensity setting value; S1: Obtain several test images by adjusting the light source intensity setting value; input the test images into the appearance recognition model to obtain the category label and its corresponding defect position and defect size; S2: Calculate the image defect anomaly coefficient based on the category label and its corresponding defect position and defect size, and then calculate the appearance defect anomaly coefficient; S3: Generate an alarm signal according to the appearance defect abnormality coefficient, and classify and process the battery products according to the alarm signal.

Images