Battery cell assembly polarity detection system based on machine vision

Abstract

The invention relates to the technical field of image analysis, in particular to a battery cell assembly polarity detection system based on machine vision, which comprises a reference acquisition module, a brightness analysis module, a feature extraction module, a response identification module and a classification judgment module. According to the invention, the reference containing the reference vector direction angle and the initial gray average value is established, so that the direction and brightness deviation caused by illumination change in the current detection period can be quantified, and the compensation instruction is generated according to the direction and brightness deviation to dynamically adjust the exposure time length and the light source voltage, thereby realizing the real-time correction of the acquired image; meanwhile, the local direction entropy of the tab area is calculated in the corrected image, and weighting processing is carried out in combination with the neighborhood brightness mean value difference, so that the distinction degree of surface textures and gloss characteristics of materials with different polarities is enhanced, and the accuracy of the surface texture and gloss characteristics of the materials with different polarities is improved. And finally, performing comprehensive judgment according to multi-dimensional indexes such as a brightness reflectivity mean value and a local direction entropy strength ratio.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology, and in particular to a cell assembly polarity detection system based on machine vision. Background Technology

[0002] Image analysis technology is an application area of ​​computer vision. It mainly studies how to use image acquisition and processing technology to extract the morphological features, position, pose and spatial relationship information of objects from images in order to achieve the detection, recognition and localization of targets.

[0003] Among them, the cell assembly polarity detection system refers to the detection system used in the lithium battery manufacturing process to determine whether the positive and negative poles of the cell are correctly assembled. It usually uses photoelectric sensors, line scan cameras or color recognition components to analyze the shape, marking color or orientation difference of the cell tabs in order to determine the polarity direction of the cell assembly.

[0004] Existing technologies primarily rely on photoelectric sensors or cameras to identify the shape, color, or orientation of the battery cell tabs under preset parameters. This detection method is highly dependent on a stable and ideal lighting environment and precise mechanical positioning. In actual production, changes in ambient light, the aging and decay of the light source itself, or slight positional shifts caused by cell placement can all directly affect the quality of image or signal acquisition. This can lead to the system misjudging normal feature differences as defects or missing real polarity errors. Furthermore, its identification basis is limited to macroscopic physical features, such as outline and color. When encountering special cases where the positive and negative tabs are extremely similar in shape and have no obvious color markings due to batch differences or manufacturing tolerances, it is difficult to make a reliable distinction by analyzing the outline or orientation alone. This can easily lead to assembly errors and pose a potential threat to the consistency and safety of the final battery product. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a cell assembly polarity detection system based on machine vision.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a machine vision-based cell assembly polarity detection system includes: The benchmark acquisition module acquires image frames of the initial cycle lithium battery cell assembly area under illumination, determines the geometric center coordinates of the cell body area and the reflection center coordinates of the tab area, calculates the reference vector direction angle, and statistically analyzes the grayscale average of all image frames to form an illumination direction reference dataset. The brightness analysis module acquires image frames of the lithium battery cell assembly area in the current cycle, determines the reference vector direction angle and grayscale mean value of the current cycle, compares them with the reference vector direction angle and grayscale mean value of the initial cycle in the illumination direction reference data set, and outputs a brightness compensation adjustment command. The feature extraction module corrects the image frame of the current period through the brightness compensation adjustment command, calculates the orientation entropy weight of each pixel position in the battery cell tab region of the corrected image frame, and outputs the orientation entropy weight data. The response recognition module calculates the directional concentration of the cell tab region using the directional entropy weight data, and filters polarity difference response regions by combining the change amplitude of the average gray value between the initial period and the current period. The classification and determination module determines the positive or negative polarity classification code of the battery cell tab based on the polarity difference response region and outputs the polarity detection and determination result.

[0007] As a further aspect of the present invention, the illumination direction reference dataset includes the initial period reference vector direction angle, the initial period grayscale mean, the geometric center coordinates of the main body region of the battery cell, and the reflection center coordinates of the tab region. The brightness compensation adjustment command includes exposure time adjustment parameters, light source control voltage adjustment parameters, direction offset angle, and brightness change amplitude. The direction entropy weight data includes pixel gradient direction information, pixel gradient magnitude information, local direction entropy value, and direction entropy weight distribution. The polarity difference response region includes direction concentration parameters, brightness change amplitude parameters, response region coordinate range, and direction feature distribution. The polarity detection and determination result includes positive electrode classification code, negative electrode classification code, and polarity status identifier.

[0008] As a further aspect of the present invention, the benchmark acquisition module includes: The image frame region positioning submodule acquires image frames of the initial cycle lithium battery cell assembly area under illumination. It detects the brightness peak point of the reflected light from the metal surface of the tab area in the image frame as the reflection center coordinates, and calculates the geometric center coordinates of the main body area of ​​the cell based on the pixel range of the cell's outer boundary in the image frame. The two coordinates are integrated to establish a dual-center position coordinate system. The reference vector angle calculation submodule calls the dual-center position coordinates, sets the geometric center coordinates of the main body area of ​​the cell as the starting point of the reference vector, specifies the reflection center coordinates of the tab area as the endpoint of the reference vector, performs trigonometric function calculations based on the spatial coordinate difference between the two points, calculates the angle between the line connecting the two points and the horizontal direction, and generates the reference vector direction angle. The illumination direction data calibration submodule calculates the average pixel grayscale value of all image frames in the initial period, obtains the average grayscale value of the initial period, and combines the average grayscale value of the initial period with the reference vector direction angle in a structured way to establish an illumination direction reference dataset.

[0009] As a further aspect of the present invention, the brightness analysis module includes: The current period parameter extraction submodule acquires the image frames of the lithium battery cell assembly area in the current period, detects the reflection center coordinates of the cell tab area and calculates the reference vector direction angle in the current period, calculates the average pixel gray value of all image frames in the current period, obtains the gray value mean in the current period, integrates the reference vector direction angle in the current period and the gray value mean in the current period, and establishes the illumination features in the current period. The illumination state offset quantization submodule calls the illumination direction reference dataset and the current period illumination features, compares the reference vector direction angle of the current period with the reference vector direction angle of the initial period to calculate the direction offset angle, and then compares the gray value mean of the current period with the gray value mean of the initial period to calculate the brightness change amplitude and generate the comprehensive illumination deviation. The brightness compensation command generation submodule adjusts the light source control voltage and exposure time based on the directional offset angle and brightness change amplitude of the comprehensive illumination deviation, and combines the adjusted values ​​into a control command to output a brightness compensation adjustment command.

[0010] As a further aspect of the present invention, the process of adjusting the light source control voltage and exposure time based on the directional offset angle and brightness change amplitude of the comprehensive illumination deviation is specifically as follows: Set the directional offset adjustment threshold and the brightness change adjustment threshold; The directional offset angle is compared with the directional offset adjustment threshold. When the directional offset angle exceeds the directional offset adjustment threshold, the adjustment amount of the light source control voltage is linearly related to the difference between the directional offset angle exceeding the directional offset adjustment threshold. The brightness change amplitude is compared with the brightness change adjustment threshold. When the brightness change amplitude exceeds the brightness change adjustment threshold, the adjustment amount of the exposure time is linearly related to the difference between the brightness change amplitude and the brightness change adjustment threshold.

[0011] As a further aspect of the present invention, the feature extraction module includes: The image frame gradient feature extraction submodule calls the brightness compensation adjustment instruction to perform correction on the image frame of the current period, obtains the corrected image frame, and detects the gradient direction and gradient magnitude of each pixel in the battery tab area of ​​the corrected image frame. It integrates the gradient direction data and gradient magnitude data of all pixels to establish a gradient feature set. The local orientation entropy calculation submodule, based on the gradient orientation data in the gradient feature set, calculates the distribution of gradient orientation in the neighborhood of each pixel in the cell tab region, and calculates the local orientation entropy value of the pixel according to the dispersion of the orientation distribution. The local orientation entropy values ​​of all pixels are collected to generate a local orientation entropy set. The directional entropy weighted calculation submodule calculates the mean brightness difference of each pixel's neighborhood in the battery cell tab region for the corrected image frame, performs weighted calculation on the entropy value of each pixel in the local directional entropy set and the mean brightness difference of its corresponding neighborhood, and outputs directional entropy weight data.

[0012] As a further aspect of the present invention, the response recognition module includes: The orientation concentration calculation submodule calls the orientation entropy weight data and performs an aggregation operation on the weight values ​​of all pixels in the cell tab area to obtain the orientation concentration parameter. The feature change amplitude quantification submodule sorts the pixel coordinate sequence of the battery cell tab region in descending order according to the directional concentration parameter, maps the brightness change amplitude with the spatial distribution trend, and generates a comprehensive feature change amount. The polarity difference region filtering submodule calls the comprehensive feature change amount, compares the directional concentration parameter of each local region in the data with the preset directional concentration threshold, and compares the brightness change amplitude corresponding to each local region with the preset brightness change amplitude threshold to filter polarity difference response regions.

[0013] As a further aspect of the present invention, the classification determination module includes: The response region feature extraction submodule statistically analyzes the gradient direction of pixels within the polarity difference response region to determine the dominant distribution direction, and obtains the gray intensity change of the polarity difference response region as the brightness feature distribution to establish multidimensional features of the response region. The polarity feature index conversion submodule calculates the mean brightness reflectance of the brightness feature distribution data in the multidimensional features of the response region, and calculates the local directional entropy intensity ratio based on the dominant distribution direction information. The mean brightness reflectance and the local directional entropy intensity ratio are combined to generate a polarity classification quantification index. The electrode category coding generation submodule determines the positive or negative electrode category based on the polarity classification quantification index, assigns the corresponding classification code, and outputs the polarity detection judgment result.

[0014] As a further aspect of the present invention, the process of determining the positive or negative electrode classification based on the polarity classification quantification index is specifically as follows: Set the positive electrode brightness reflectance determination interval, the negative electrode brightness reflectance determination interval, the positive electrode directional entropy intensity determination interval, and the negative electrode directional entropy intensity determination interval; Determine whether the mean brightness reflectance in the polarity classification quantification index falls within the positive polarity brightness reflectance determination interval, and determine whether the local directional entropy intensity ratio in the polarity classification quantification index falls within the positive polarity directional entropy intensity determination interval. If the ratio of the average brightness reflectance to the local directional entropy intensity both satisfy the condition of falling into the corresponding judgment interval, then the classification is determined to be positive. Determine whether the mean brightness reflectance in the polarity classification quantification index falls within the negative polarity brightness reflectance determination interval, and determine whether the local directional entropy intensity ratio in the polarity classification quantification index falls within the negative polarity directional entropy intensity determination interval. If the ratio of the average brightness reflectance to the local directional entropy intensity both satisfy the condition of falling into the corresponding judgment interval, then the classification is determined to be negative.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by establishing a benchmark including the reference vector direction angle and the initial grayscale mean, the directional and brightness shifts caused by changes in illumination during the current detection cycle can be quantified. Based on this, compensation instructions are generated to dynamically adjust the exposure time and light source voltage, achieving real-time correction of the acquired image. This eliminates the interference of ambient light and equipment state fluctuations on subsequent feature analysis. Simultaneously, the local directional entropy of the tab region is calculated in the corrected image and weighted by combining it with the difference in the mean brightness of the neighborhood, thereby enhancing the distinguishability of the surface texture and gloss characteristics of materials with different polarities. Finally, a comprehensive judgment is made based on multi-dimensional indicators such as the mean brightness reflectance and the ratio of local directional entropy intensity. Compared with recognition methods that rely on a single shape or color feature, this adaptive correction to the illumination environment and in-depth mining of the essential characteristics of materials significantly improves the detection accuracy and system stability under interference such as complex illumination, minor installation displacement, and changes in the surface state of the battery cell. Attached Figure Description

[0016] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a flowchart of the benchmark acquisition module of the present invention; Figure 3 This is a flowchart of the brightness analysis module of the present invention; Figure 4 This is a flowchart of the feature extraction module of the present invention; Figure 5 This is a flowchart of the response recognition module of the present invention; Figure 6 This is a flowchart of the classification and determination module of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Please see Figure 1 The machine vision-based cell assembly polarity detection system includes: The benchmark acquisition module acquires image frames of the initial cycle lithium battery cell assembly area under illumination, determines the geometric center coordinates of the cell body area and the reflection center coordinates of the tab area, calculates the reference vector direction angle, and statistically analyzes the grayscale average of all image frames to form an illumination direction reference dataset. The brightness analysis module acquires image frames of the lithium battery cell assembly area in the current cycle, determines the reference vector direction angle and grayscale mean of the current cycle, compares them with the reference vector direction angle and grayscale mean of the initial cycle in the illumination direction reference data, and outputs brightness compensation adjustment instructions. The feature extraction module corrects the image frame of the current period through brightness compensation adjustment instructions, calculates the orientation entropy weight of each pixel position in the battery cell tab region of the corrected image frame, and outputs the orientation entropy weight data. The response recognition module calculates the directional concentration of the cell tab area using directional entropy weight data, and filters polarity difference response areas by combining the change amplitude of the average gray value between the initial period and the current period. The classification and determination module determines the positive or negative polarity classification code of the battery cell tabs based on the polarity difference response area and outputs the polarity detection and determination result. The illumination direction reference dataset includes the initial period reference vector direction angle, the initial period grayscale mean, the geometric center coordinates of the main body area of ​​the battery cell, and the reflection center coordinates of the tab area. The brightness compensation adjustment instructions include exposure time adjustment parameters, light source control voltage adjustment parameters, direction offset angle, and brightness change amplitude. The direction entropy weight data includes pixel gradient direction information, pixel gradient magnitude information, local direction entropy value, and direction entropy weight distribution. The polarity difference response region includes direction concentration parameters, brightness change amplitude parameters, response region coordinate range, and direction feature distribution. The polarity detection and determination results include positive electrode classification code, negative electrode classification code, and polarity status identifier.

[0019] Please see Figure 2 The benchmark acquisition module includes: The image frame region positioning submodule acquires image frames of the initial cycle lithium battery cell assembly area under illumination. It detects the brightness peak point of the reflected light from the metal surface of the tab area in the image frame as the reflection center coordinates, and calculates the geometric center coordinates of the main body area of ​​the cell based on the pixel range of the cell's outer boundary in the image frame. The two coordinates are integrated to establish a dual-center position coordinate system. The illumination device is activated to illuminate the first lithium-ion battery cell in the initial batch placed at the inspection station, and the image sensor is driven to capture image frames of the cell assembly area. These image frames are stored at a resolution of 1920 pixels by 1080 pixels. Subsequently, within the acquired image frame data, for a pre-defined tab area, such as a rectangular region located at the top of the image frame with coordinates ranging from (800, 200) to (1200, 400), the brightness value of each pixel is detected. By traversing all pixels within this region, the highest brightness value formed by reflected light on the metal surface is identified, and the coordinates of this pixel, for example (1015, 258), are recorded as the reflection center coordinates. Simultaneously, edge detection processing is performed to identify the complete outline of the battery cell body within the entire image frame. Based on the set of contour pixels, the smallest rectangular boundary that can completely cover the main body area of ​​the battery cell is calculated. For example, the coordinates of its four vertices are (400, 100), (1520, 100), (400, 980), and (1520, 980). Based on the coordinates of the four vertices of this rectangular boundary, the geometric center coordinates of the main body area of ​​the battery cell are calculated as (960, 540) by taking the average of the horizontal and vertical coordinates. Finally, the reflection center coordinates (1015, 258) of the tab area and the geometric center coordinates (960, 540) of the main body area of ​​the battery cell are structurally integrated to form a data pair containing two coordinate points, establishing a dual-center position coordinate system.

[0020] The reference vector angle calculation submodule calls the dual-center position coordinates, sets the geometric center coordinates of the main body area of ​​the cell as the starting point of the reference vector, specifies the reflection center coordinates of the tab area as the endpoint of the reference vector, performs trigonometric function calculations based on the spatial coordinate difference between the two points, calculates the angle between the line connecting the two points and the horizontal direction, and generates the reference vector direction angle. The generated coordinates of the two centers are retrieved, namely the geometric center coordinates (960, 540) and the reflection center coordinates (1015, 258). Based on this, the geometric center coordinates (960, 540) of the main body area of ​​the battery cell are set as the starting point of the spatial vector, and the reflection center coordinates (1015, 258) of the tab area are designated as the ending point of the spatial vector. To calculate the angle between the line connecting these two coordinate points and the horizontal direction, the coordinate difference between the two points in the horizontal direction is first calculated, resulting in 55. Next, the coordinate difference between the two points in the vertical direction is calculated, resulting in -282. After completing the coordinate difference calculation, trigonometric function operations are performed, specifically calculating the ratio of the vertical coordinate difference (-282) to the horizontal coordinate difference (55), resulting in -5.127. Then, the arctangent value of this ratio is obtained, and the angle between the line connecting the two points and the horizontal direction is calculated to be approximately -78.96 degrees. This value of -78.96 degrees was determined as the reference vector direction angle for the initial period.

[0021] The illumination direction data calibration submodule calculates the average pixel gray value of all image frames in the initial period, obtains the average gray value of the initial period, and combines the average gray value of the initial period with the reference vector direction angle in a structured way to establish an illumination direction reference dataset. Processing is performed on all image frames of the initial production cycle, which is set to include the first 500 battery cells in a production batch. For each of these 500 cells, the process of acquiring image frames is repeated, resulting in 500 independent image frames. For each image frame, all pixels within its entire resolution range are traversed, and the grayscale value of each pixel is read. The grayscale values ​​of all pixels within a single frame are summed, and then divided by the total number of pixels to calculate the average pixel grayscale value of that frame. For example, the average pixel grayscale value of the first image frame is 145.8, the second frame is 146.3, and so on, until the average pixel grayscale value of the 500th image frame is 145.5. After acquiring the average pixel grayscale values ​​of all 500 image frames, these 500 grayscale averages are summed and then divided by the total number of image frames, 500, to obtain the final grayscale average value of the initial cycle, for example, 146.0. The initial periodic grayscale mean of 146.0 and the calculated reference vector direction angle of -78.96 degrees are combined in a structured manner to form a data structure containing two data items: direction angle and grayscale mean, which is then established as the illumination direction reference dataset.

[0022] Please see Figure 3 The brightness analysis module includes: The current period parameter extraction submodule acquires the image frames of the lithium battery cell assembly area in the current period, detects the reflection center coordinates of the cell tab area and calculates the reference vector direction angle in the current period, calculates the average pixel gray value of all image frames in the current period, obtains the gray value mean in the current period, integrates the reference vector direction angle in the current period and the gray value mean in the current period, and establishes the illumination features in the current period. When the production process enters a new inspection cycle, such as when inspecting the 501st cell, a new image frame of the current lithium battery cell assembly area is acquired. The brightness of all pixels within the preset tab area of ​​the current image frame is detected, and the position of the brightness peak point is found and recorded as the reflection center coordinates for the current cycle, for example, (1020, 265). Simultaneously, through the same edge detection and geometric center calculation process, the geometric center coordinates of the current cell's main body area are obtained, for example, (962, 543). Based on these two new coordinates, the geometric center (962, 543) is set as the starting point, and the reflection center (1020, 265) as the endpoint, and the reference vector direction angle for the current cycle is calculated, for example, -78.19 degrees. At the same time, for all image frames acquired in the current cycle, assuming 100 cells are processed in the current cycle, 100 image frames are acquired, the average pixel grayscale value of each frame is calculated, and then these 100 average grayscale values ​​are arithmetically averaged to obtain the grayscale mean for the current cycle, for example, 152.5. By integrating the current period reference vector direction angle of -78.19 degrees with the current period grayscale average of 152.5, a dataset describing the current illumination state is established, namely the current period illumination feature.

[0023] The illumination state offset quantization submodule calls the illumination direction reference dataset and the illumination features of the current period, compares the reference vector direction angle of the current period with the reference vector direction angle of the initial period to calculate the direction offset angle, and then compares the gray value of the current period with the gray value of the initial period to calculate the brightness change amplitude and generate the comprehensive illumination deviation. The illumination direction reference dataset (containing a reference vector direction angle of -78.96 degrees and an initial period's average grayscale value of 146.0) and the current period's illumination features (containing a current period's reference vector direction angle of -78.19 degrees and a current period's average grayscale value of 152.5) ​​are retrieved. The reference vector direction angle of the current period (-78.19 degrees) is subtracted from the reference vector direction angle of the initial period (-78.96 degrees), yielding 0.77 degrees, which is the direction offset angle. Next, the average grayscale value of the current period (152.5) ​​is subtracted from the average grayscale value of the initial period (146.0), yielding 6.5, which quantifies the magnitude of brightness change. The calculated direction offset angle of 0.77 degrees and the brightness change magnitude of 6.5 are combined to generate a data pair containing these two offset values, serving as the overall illumination deviation.

[0024] The brightness compensation command generation submodule adjusts the light source control voltage and exposure time based on the directional offset angle and brightness change amplitude of the overall illumination deviation, and combines the adjusted values ​​into a control command to output the brightness compensation adjustment command. The process of adjusting the light source control voltage and exposure time based on the directional offset angle and brightness change amplitude of the overall illumination deviation is as follows: Set the directional offset adjustment threshold and the brightness change adjustment threshold; The directional offset angle is compared with the directional offset adjustment threshold. When the directional offset angle exceeds the directional offset adjustment threshold, the adjustment amount of the light source control voltage is linearly related to the difference between the directional offset angle exceeding the directional offset adjustment threshold. The brightness change amplitude is compared with the brightness change adjustment threshold. When the brightness change amplitude exceeds the brightness change adjustment threshold, the adjustment amount of the exposure time is linearly related to the difference between the brightness change amplitude and the brightness change adjustment threshold. First, we set the orientation offset adjustment threshold and the brightness change adjustment threshold. The orientation offset adjustment threshold was set based on a statistical analysis of the reference vector orientation angles of over 10,000 qualified battery cells from historical production data. It was found that under stable lighting and standard installation conditions, the orientation angle fluctuation range of 99.5% of the samples was within ±0.5 degrees; therefore, the orientation offset adjustment threshold was set to 0.5 degrees. The brightness change adjustment threshold was also set based on the analysis of the image grayscale mean of the above samples. It was found that under stable conditions, the grayscale mean fluctuation range was within ±5 grayscale units; therefore, the brightness change adjustment threshold was set to 5.0. The calculated orientation offset angle of 0.77 degrees in the overall lighting deviation was compared with the orientation offset adjustment threshold of 0.5 degrees. Since 0.77 degrees exceeded 0.5 degrees, the difference was calculated to be 0.27 degrees. The adjustment amount of the light source control voltage is linearly related to this difference. Experimentally, it was calibrated that for every 0.1 degree exceeding this difference, the light source control voltage decreases by 0.02 volts. Therefore, the voltage adjustment is 0.27 divided by 0.1 and then multiplied by 0.02, resulting in 0.054 volts. The brightness change amplitude of 6.5 is then compared with the brightness change adjustment threshold of 5.0. Since 6.5 exceeds 5.0, the difference of 1.5 is calculated. The exposure time adjustment is also linearly related to this difference; experimentally calibrated, it is determined that for every 1.0 gray unit exceeding the threshold, the exposure time decreases by 10 microseconds. Therefore, the exposure time adjustment is 1.5 divided by 1.0 and then multiplied by 10, resulting in 15 microseconds. The calculated light source control voltage adjustment of -0.054 volts and the exposure time adjustment of -15 microseconds are combined into a control command, outputting a brightness compensation adjustment instruction.

[0025] Please see Figure 4 The feature extraction module includes: The image frame gradient feature extraction submodule calls the brightness compensation adjustment command to perform correction on the image frame of the current period, obtains the corrected image frame, and detects the gradient direction and gradient magnitude of each pixel in the battery tab area of ​​the corrected image frame. It integrates the gradient direction data and gradient magnitude data of all pixels to establish a gradient feature set. A brightness compensation adjustment command is retrieved and sent to the illumination controller and image sensor. The illumination controller reduces the current light source control voltage by 0.054 volts based on the voltage adjustment amount in the command; the image sensor reduces the exposure time for the next acquisition by 15 microseconds based on the exposure time adjustment amount in the command. Under these adjusted parameters, a new image frame is captured for the current cycle of the lithium battery cell; this image is the corrected image frame. After acquiring the corrected image frame, gradient information is extracted for each pixel within the cell tab region of that image frame. For any pixel, its grayscale value is read, along with the grayscale values ​​of its horizontal and vertical adjacent pixels. The gradient magnitude is calculated as the square root of the sum of the squares of the horizontal and vertical grayscale differences. The gradient direction is calculated as the arctangent of the ratio of the vertical to the horizontal grayscale difference. This process iterates through all pixels within the tab region, integrating the gradient direction and gradient magnitude data of all pixels into a single dataset to establish a gradient feature set.

[0026] The local orientation entropy calculation submodule, based on the gradient orientation data in the gradient feature set, calculates the distribution of gradient orientation in the neighborhood of each pixel in the battery cell tab region, and calculates the local orientation entropy value of the pixel according to the dispersion of the orientation distribution. The local orientation entropy values ​​of all pixels are collected to generate a local orientation entropy set. The gradient direction data in the gradient feature set is processed. For each pixel within the battery cell tab region, a 5×5 pixel neighborhood centered on that pixel is defined. The local orientation entropy value of the pixel is calculated according to the following expression: .in, Representing pixels The local entropy value is a dimensionless parameter used to quantify the degree of disorder in the gradient directions within the neighborhood of a pixel. A higher entropy value indicates a more complex texture or more inconsistent directions within the neighborhood, such as scratches or foreign objects on an electrode surface; a lower entropy value indicates more consistent directions, such as a smooth electrode surface. It is the specified pixel currently located in the image of the battery cell tab area. This is the directional dispersion sensitivity coefficient, a dimensionless parameter. Its function is to adjust the sensitivity of the calculation results to directional disorder. When the value is greater than 1, it amplifies the entropy value, making the difference between regions with slight directional changes and completely disordered regions more pronounced. Its setting is based on maximizing the contrast that distinguishes different texture features. Experimental tests were conducted on 1000 sample images containing both sharp edges and complex textures. The value is set at 1.1. This represents the total number of gradient direction intervals. In this embodiment, the direction from 0 to 360 degrees is divided into 8 intervals (e.g., 0-45 degrees, 45-90 degrees, etc.). Set to 8. It is the index of the gradient direction interval, from 1 to... . In pixels Within the neighborhood of , the gradient direction falls into the . The proportion (probability) of the number of pixels in a given interval to the total number of pixels in its neighborhood is a dimensionless parameter. The base of the logarithm is set to 2, which is the standard for calculating information entropy, allowing the unit of entropy value to be understood as "bit", which facilitates the measurement of information content.

[0027] For a certain pixel The counts of the 25 pixels within its 5×5 neighborhood in the 8 directional intervals are as follows. Then the corresponding probabilities are... The expression is [0.4, 0.2, 0.08, 0.08, 0.04, 0.04, 0.08, 0.08]. Substitute... and Perform the calculation: The summation term is: The final This calculation process is applied to all pixels within the epipolar region, generating a set of local directional entropy.

[0028] The directional entropy weighted calculation submodule calculates the mean difference in brightness of each pixel neighborhood in the battery cell tab region for the corrected image frame, performs weighted calculation on the entropy value of each pixel in the local directional entropy set and the mean difference in brightness of the corresponding neighborhood, and outputs directional entropy weight data. First, for the corrected image frame, the mean brightness difference of each pixel's neighborhood in the battery cell tab region is calculated. For each pixel, a 5×5 neighborhood is used, and the neighborhood is divided into two halves based on the pixel's principal gradient direction. The average gray value of all pixels in each of these two halves is calculated, and then the absolute value of the difference between these two average gray values ​​is calculated as the mean brightness difference of the pixel's neighborhood. Subsequently, a weighted calculation is performed according to the following expression: .in, It is a pixel The final directional entropy weight data is a dimensionless parameter that combines the texture complexity and brightness contrast of the pixel neighborhood. The larger the value, the more significant the features of the region where the pixel is located. It is the local orientation entropy value of the pixel, a dimensionless parameter whose value is calculated from the previous step. This is the average brightness difference in the neighborhood of a pixel, expressed in grayscale values. It reflects the degree of brightness variation within the neighborhood; a higher value indicates stronger contrast. This is a reference brightness value used for normalization, and its unit is also grayscale value. It is set to the maximum image grayscale value of 255, so that the ratio term... Becoming a dimensionless parameter ensures the consistency of dimensions throughout the expression. This is the brightness / contrast enhancement factor, a dimensionless exponent. Its function is to adjust the impact of brightness / contrast on the final weighting. When... When the value is greater than 1, it will non-linearly amplify the effect of brightness contrast, making the high contrast area ( The weights of larger (higher) values ​​are disproportionately amplified to highlight edge or defect features. This is based on the principle of enhancing the salience of defect features. Through comparative experiments on defect samples with different contrasts, [the following is observed / implemented]. Set to 1.5.

[0029] Substitute the calculated The value is 2.46, assuming the mean brightness difference of the pixel's neighborhood. It is set to 20. It is 255. The value is 1.5. The calculation process is as follows: Calculate the ratio term: Calculate the ratio term Power: The final Perform this operation on all pixels, outputting directional entropy weight data corresponding to the pixels in the polar region.

[0030] Please see Figure 5 The response recognition module includes: The orientation concentration calculation submodule calls the orientation entropy weight data and performs an aggregation operation on the weight values ​​of all pixels in the cell tab area to obtain the orientation concentration parameter. The directional entropy weight data map is invoked. This map is a two-dimensional data matrix of the same size as the electrode region of the battery cell. To obtain the directional concentration parameter, an aggregation operation needs to be performed on the weight values ​​of all pixels within the electrode region; that is, the values ​​of all elements in this two-dimensional data matrix are summed. For example, if the electrode region is 200×200 pixels, totaling 40,000 pixels, then all 40,000 directional entropy weight data values ​​are summed. Assuming that the summation operation yields a total of 5,834,120, this value represents the directional concentration parameter for the entire electrode region.

[0031] The feature change amplitude quantification submodule sorts the pixel coordinate sequence of the battery cell tab region in descending order according to the orientation concentration parameter, maps the brightness change amplitude with the spatial distribution trend, and generates a comprehensive feature change amount. Based on the directional concentration parameter and the specific directional entropy weight data of each pixel, the pixel coordinate sequence of the battery cell tab region is sorted in descending order. The sorting is based on the directional entropy weight data value of each pixel, with pixels having higher values ​​appearing first. This results in a pixel spatial coordinate sequence from high weight to low weight. Next, the initial periodic grayscale average value of 146.0 and the current periodic grayscale average value of 152.5 are obtained. The brightness change amplitude between these two is calculated to be 6.5. Subsequently, the brightness change amplitude of 6.5 is mapped to the pixel spatial distribution trend. The mapping process is as follows: the sorted pixel coordinate sequence is analyzed to determine whether high-weight pixels are clustered or dispersed. If high-weight pixels are highly clustered in a small area and the brightness change amplitude is positive, they are marked as "strongly positively correlated clustering" in the comprehensive feature change data; if the pixels are dispersed, they are marked as "weakly positively correlated dispersion". Based on the positive change amplitude of 6.5 and the pixel distribution, the corresponding comprehensive feature change is generated.

[0032] The polarity difference region filtering submodule calls the comprehensive feature change amount, compares the direction concentration parameter of each local region in the data with the preset direction concentration threshold, and compares the brightness change amplitude of each local region with the preset brightness change amplitude threshold to filter the polarity difference response region. The process involves invoking the comprehensive feature change. This process presets a directional concentration threshold and a brightness change amplitude threshold. The directional concentration threshold is set by analyzing local regions (e.g., 10×10 pixel areas) of 500 known positive electrode samples and 500 known negative electrode samples. It was found that regions clearly distinguishing polarity typically have a local directional concentration parameter (the sum of directional entropy weights within the local region) above 8000, while background or non-feature areas have a value below 3000. Therefore, the preset directional concentration threshold is set to 5000. The brightness change amplitude threshold is set by comparing the changes in positive and negative electrodes relative to a baseline under different lighting conditions. It was found that the absolute value of the brightness change amplitude in the effective response area is typically greater than 4.0 grayscale units. Therefore, the preset brightness change amplitude threshold is set to 4.0. The screening process involves dividing the cell tab area into several non-overlapping 10×10 pixel local regions. For each local region, the sum of the directional entropy weights of all pixels within it is calculated as the directional concentration parameter for that local region. Simultaneously, the difference between the average gray value of the local area and the average gray value of the corresponding area in the initial period is calculated as its brightness change amplitude. Only when the directional concentration parameter of a local area is greater than 5000 and the absolute value of its brightness change amplitude is also greater than 4.0, the local area is selected and marked as a polarity difference response area.

[0033] Please see Figure 6 The classification and determination module includes: The response region feature extraction submodule statistically analyzes the gradient direction of pixels within the polarity difference response region to determine the dominant distribution direction, and obtains the gray intensity change of the polarity difference response region as the brightness feature distribution to establish multidimensional features of the response region. All selected polarity difference response regions are analyzed. All labeled response regions are merged into a single analysis region. Within this region, the gradient direction data of all pixels are statistically analyzed. The directional range from 0 to 180 degrees is divided into 18 intervals of 10 degrees each, and the number of pixels falling into each interval is counted. The direction represented by the interval with the most pixels, such as the 40-50 degree interval, is determined as the dominant distribution direction. Simultaneously, the grayscale values ​​of all pixels within these response regions are acquired, and the change in grayscale value relative to the corresponding region in the initial period is calculated. This set of change data is used as the brightness feature distribution. For example, the average grayscale value of the response region increases from the initial 160 to 185. The determined dominant distribution direction (40-50 degrees) is combined with the brightness feature distribution (average grayscale change + 25) to establish a multidimensional feature describing the characteristics of this response region.

[0034] The polarity feature index conversion submodule calculates the mean brightness reflectance from the brightness feature distribution data in the multidimensional features of the response region, and calculates the local directional entropy intensity ratio based on the dominant distribution direction information. The mean brightness reflectance and the local directional entropy intensity ratio are combined to generate a polarity classification quantification index. The mean luminance reflectance is calculated from the luminance feature distribution data in the multidimensional features. This process directly takes the average grayscale value of all pixels within the response region in the corrected image frame. For example, the calculated average grayscale value of the response region is 185.3. Next, the local directional entropy intensity ratio is calculated based on the dominant distribution direction information. The specific calculation method is as follows: Within the response region, all pixels whose gradient direction is consistent with or close to the dominant distribution direction (40-50 degrees) (e.g., within ±10 degrees) are selected, and the local directional entropy values ​​of these pixels are summed to obtain the "main direction entropy sum". Then, all pixels whose gradient direction is perpendicular to the dominant distribution direction (e.g., within the 130-150 degree range) are selected, and the local directional entropy values ​​of these pixels are summed to obtain the "vertical direction entropy sum". Finally, the main direction entropy sum is divided by the vertical direction entropy sum to obtain the local directional entropy intensity ratio. For example, if the calculated main direction entropy sum is 2450.7 and the vertical direction entropy sum is 980.2, then the local directional entropy intensity ratio is 2.5. The average brightness reflectance of 185.3 is combined with the ratio of local directional entropy intensity of 2.5 to generate a polarity classification quantification index.

[0035] The electrode category coding generation submodule determines the positive or negative electrode category based on the polarity classification quantification index, assigns the corresponding classification code, and outputs the polarity detection judgment result. The process of determining the positive or negative polarity classification based on polarity classification quantification indicators is as follows: Set the positive electrode brightness reflectance determination interval, the negative electrode brightness reflectance determination interval, the positive electrode directional entropy intensity determination interval, and the negative electrode directional entropy intensity determination interval; Determine whether the mean value of brightness reflectance in the polarity classification quantification index falls within the positive polarity brightness reflectance determination interval, and determine whether the ratio of local directional entropy intensity in the polarity classification quantification index falls within the positive polarity directional entropy intensity determination interval. If the ratio of the mean brightness reflectance to the local directional entropy intensity both meet the condition of falling into the corresponding judgment interval, then the classification is determined to be positive. Determine whether the mean value of brightness reflectance in the polarity classification quantification index falls within the negative polarity brightness reflectance determination interval, and determine whether the ratio of local directional entropy intensity in the polarity classification quantification index falls within the negative polarity directional entropy intensity determination interval. If the ratio of the mean brightness reflectance to the local directional entropy intensity both meet the condition of falling into the corresponding judgment interval, then the classification is determined to be negative. First, we set the determination intervals for positive electrode brightness reflectivity, negative electrode brightness reflectivity, positive electrode directional entropy intensity, and negative electrode directional entropy intensity. The values ​​for these intervals were determined by statistically analyzing the distribution patterns of polarity classification quantification indicators after performing the aforementioned full-process analysis on a large number of battery cell samples clearly labeled as positive (aluminum) and negative (copper) electrodes. For example, we set the positive electrode brightness reflectivity determination interval to [2.2, 3.8], the negative electrode brightness reflectivity determination interval to [0.7, 1.6], the positive electrode directional entropy intensity determination interval to [0.7, 1.6].

[0036] Subsequently, the polarity classification quantification indicators of the current battery cell (average brightness reflectance 185.3, local directional entropy intensity ratio 2.5) are judged. It is determined whether the average brightness reflectance of 185.3 falls within the positive polarity brightness reflectance determination interval, and whether the local directional entropy intensity ratio of 2.5 falls within the positive polarity directional entropy intensity determination interval [2.2, 3.8]. If both conditions are met, the cell is classified as positive. In this example, 185.3 falls within the positive polarity, and 2.5 falls within [2.2, 3.8], therefore it is classified as positive, and the corresponding classification code "P-TYPE-01" is assigned, outputting the polarity detection result. Simultaneously, it is determined whether the average brightness reflectance falls within the negative polarity brightness reflectance determination interval, and whether the local directional entropy intensity ratio falls within the negative polarity directional entropy intensity determination interval [0.7, 1.6]. If both conditions are met, the polarity is determined to be negative, and a corresponding classification code is assigned, outputting the polarity detection result. In this example, 185.3 does not fall into the negative brightness range, and 2.5 does not fall into the negative intensity range, so it is not determined to be negative.

[0037] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A cell assembly polarity detection system based on machine vision, characterized in that, The system includes: The benchmark acquisition module acquires image frames of the initial cycle lithium battery cell assembly area under illumination, determines the geometric center coordinates of the cell body area and the reflection center coordinates of the tab area, calculates the reference vector direction angle, and statistically analyzes the grayscale average of all image frames to form an illumination direction reference dataset. The brightness analysis module acquires image frames of the lithium battery cell assembly area in the current cycle, determines the reference vector direction angle and grayscale mean value of the current cycle, compares them with the reference vector direction angle and grayscale mean value of the initial cycle in the illumination direction reference data set, and outputs a brightness compensation adjustment command. The feature extraction module corrects the image frame of the current period through the brightness compensation adjustment command, calculates the orientation entropy weight of each pixel position in the battery cell tab region of the corrected image frame, and outputs the orientation entropy weight data. The response recognition module calculates the directional concentration of the cell tab region using the directional entropy weight data, and filters polarity difference response regions by combining the change amplitude of the average gray value between the initial period and the current period. The classification and determination module determines the positive or negative polarity classification code of the battery cell tab based on the polarity difference response region and outputs the polarity detection and determination result.

2. The cell assembly polarity detection system based on machine vision according to claim 1, characterized in that, The illumination direction reference dataset includes the initial period reference vector direction angle, the initial period grayscale mean, the geometric center coordinates of the main body region of the battery cell, and the reflection center coordinates of the tab region. The brightness compensation adjustment command includes exposure time adjustment parameters, light source control voltage adjustment parameters, direction offset angle, and brightness change amplitude. The direction entropy weight data includes pixel gradient direction information, pixel gradient magnitude information, local direction entropy value, and direction entropy weight distribution. The polarity difference response region includes direction concentration parameters, brightness change amplitude parameters, response region coordinate range, and direction feature distribution. The polarity detection and determination result includes positive electrode classification code, negative electrode classification code, and polarity status identifier.

3. The cell assembly polarity detection system based on machine vision according to claim 1, characterized in that, The benchmark acquisition module includes: The image frame region positioning submodule acquires image frames of the initial cycle lithium battery cell assembly area under illumination. It detects the brightness peak point of the reflected light from the metal surface of the tab area in the image frame as the reflection center coordinates, and calculates the geometric center coordinates of the main body area of ​​the cell based on the pixel range of the cell's outer boundary in the image frame. The two coordinates are integrated to establish a dual-center position coordinate system. The reference vector angle calculation submodule calls the dual-center position coordinates, sets the geometric center coordinates of the main body area of ​​the cell as the starting point of the reference vector, specifies the reflection center coordinates of the tab area as the endpoint of the reference vector, performs trigonometric function calculations based on the spatial coordinate difference between the two points, calculates the angle between the line connecting the two points and the horizontal direction, and generates the reference vector direction angle. The illumination direction data calibration submodule calculates the average pixel grayscale value of all image frames in the initial period, obtains the average grayscale value of the initial period, and combines the average grayscale value of the initial period with the reference vector direction angle in a structured way to establish an illumination direction reference dataset.

4. The cell assembly polarity detection system based on machine vision according to claim 3, characterized in that, The brightness analysis module includes: The current period parameter extraction submodule acquires the image frames of the lithium battery cell assembly area in the current period, detects the reflection center coordinates of the cell tab area and calculates the reference vector direction angle in the current period, calculates the average pixel gray value of all image frames in the current period, obtains the gray value mean in the current period, integrates the reference vector direction angle in the current period and the gray value mean in the current period, and establishes the illumination features in the current period. The illumination state offset quantization submodule calls the illumination direction reference dataset and the current period illumination features, compares the reference vector direction angle of the current period with the reference vector direction angle of the initial period to calculate the direction offset angle, and then compares the gray value mean of the current period with the gray value mean of the initial period to calculate the brightness change amplitude and generate the comprehensive illumination deviation. The brightness compensation command generation submodule adjusts the light source control voltage and exposure time based on the directional offset angle and brightness change amplitude of the comprehensive illumination deviation, and combines the adjusted values ​​into a control command to output a brightness compensation adjustment command.

5. The cell assembly polarity detection system based on machine vision according to claim 4, characterized in that, The process of adjusting the light source control voltage and exposure time based on the directional offset angle and brightness change amplitude of the overall illumination deviation is as follows: Set the directional offset adjustment threshold and the brightness change adjustment threshold; The directional offset angle is compared with the directional offset adjustment threshold. When the directional offset angle exceeds the directional offset adjustment threshold, the adjustment amount of the light source control voltage is linearly related to the difference between the directional offset angle exceeding the directional offset adjustment threshold. The brightness change amplitude is compared with the brightness change adjustment threshold. When the brightness change amplitude exceeds the brightness change adjustment threshold, the adjustment amount of the exposure time is linearly related to the difference between the brightness change amplitude and the brightness change adjustment threshold.

6. The cell assembly polarity detection system based on machine vision according to claim 4, characterized in that, The feature extraction module includes: The image frame gradient feature extraction submodule calls the brightness compensation adjustment instruction to perform correction on the image frame of the current period, obtains the corrected image frame, and detects the gradient direction and gradient magnitude of each pixel in the battery tab area of ​​the corrected image frame. It integrates the gradient direction data and gradient magnitude data of all pixels to establish a gradient feature set. The local orientation entropy calculation submodule, based on the gradient orientation data in the gradient feature set, calculates the distribution of gradient orientation in the neighborhood of each pixel in the cell tab region, and calculates the local orientation entropy value of the pixel according to the dispersion of the orientation distribution. The local orientation entropy values ​​of all pixels are collected to generate a local orientation entropy set. The directional entropy weighted calculation submodule calculates the mean brightness difference of each pixel's neighborhood in the battery cell tab region for the corrected image frame, performs weighted calculation on the entropy value of each pixel in the local directional entropy set and the mean brightness difference of its corresponding neighborhood, and outputs directional entropy weight data.

7. The cell assembly polarity detection system based on machine vision according to claim 6, characterized in that, To calculate the local orientation entropy value of a pixel, the formula is: ； in, Representing pixels The local directional entropy value, It is the specified pixel currently located in the image of the battery cell tab area. It is the sensitivity coefficient for directional dispersion. It represents the total number of gradient direction intervals. It is the index of the gradient direction interval. In pixels Within the neighborhood of , the gradient direction falls into the . The proportion of pixels in each interval to the total number of pixels in the neighborhood. The base of the logarithm is set to 2.

8. The cell assembly polarity detection system based on machine vision according to claim 6, characterized in that, The response recognition module includes: The orientation concentration calculation submodule calls the orientation entropy weight data and performs an aggregation operation on the weight values ​​of all pixels in the cell tab area to obtain the orientation concentration parameter. The feature change amplitude quantification submodule sorts the pixel coordinate sequence of the battery cell tab region in descending order according to the directional concentration parameter, maps the brightness change amplitude with the spatial distribution trend, and generates a comprehensive feature change amount. The polarity difference region filtering submodule calls the comprehensive feature change amount, compares the directional concentration parameter of each local region in the data with the preset directional concentration threshold, and compares the brightness change amplitude corresponding to each local region with the preset brightness change amplitude threshold to filter polarity difference response regions.

9. The cell assembly polarity detection system based on machine vision according to claim 8, characterized in that, The classification determination module includes: The response region feature extraction submodule statistically analyzes the gradient direction of pixels within the polarity difference response region to determine the dominant distribution direction, and obtains the gray intensity change of the polarity difference response region as the brightness feature distribution to establish multidimensional features of the response region. The polarity feature index conversion submodule calculates the mean brightness reflectance of the brightness feature distribution data in the multidimensional features of the response region, and calculates the local directional entropy intensity ratio based on the dominant distribution direction information. The mean brightness reflectance and the local directional entropy intensity ratio are combined to generate a polarity classification quantification index. The electrode category coding generation submodule determines the positive or negative electrode category based on the polarity classification quantification index, assigns the corresponding classification code, and outputs the polarity detection judgment result.

10. The cell assembly polarity detection system based on machine vision according to claim 9, characterized in that, The process of determining the positive or negative polarity classification based on the aforementioned polarity classification quantification index is as follows: Set the positive electrode brightness reflectance determination interval, the negative electrode brightness reflectance determination interval, the positive electrode directional entropy intensity determination interval, and the negative electrode directional entropy intensity determination interval; Determine whether the mean brightness reflectance in the polarity classification quantification index falls within the positive polarity brightness reflectance determination interval, and determine whether the local directional entropy intensity ratio in the polarity classification quantification index falls within the positive polarity directional entropy intensity determination interval. If the ratio of the average brightness reflectance to the local directional entropy intensity both satisfy the condition of falling into the corresponding judgment interval, then the classification is determined to be positive. Determine whether the mean brightness reflectance in the polarity classification quantification index falls within the negative polarity brightness reflectance determination interval, and determine whether the local directional entropy intensity ratio in the polarity classification quantification index falls within the negative polarity directional entropy intensity determination interval. If the ratio of the average brightness reflectance to the local directional entropy intensity both satisfy the condition of falling into the corresponding judgment interval, then the classification is determined to be negative.

Images