Filter cartridge filtration test method and system based on visual detection

Abstract

The application discloses a filter core filterability test method and system based on visual detection, and belongs to the technical field of filter core testing. The test method comprises the following steps: arranging a plurality of cameras and a surface light source and performing calibration to generate a calibration parameter set; after immersing the filter core in a detection liquid, synchronously collecting images, correcting and splicing to generate a panoramic image, dividing the panoramic image into a plurality of detection partitions, applying a first gas pressure to the inside of the filter core, and extracting a first total bubble area value in each detection partition through image processing to determine a broken hole defect; applying a second gas pressure, and extracting a second total bubble area value in each detection partition by using a deep learning bubble segmentation model to determine a blockage defect or a defect of insufficient opening rate. The application further discloses a system, realizes automatic and quantitative detection of filter core defects, and improves precision, efficiency and safety.

Description

Technical Field

[0001] This invention relates to the field of filter cartridge testing technology. More specifically, this invention relates to a method and system for testing the filtration performance of filter cartridges based on visual inspection. Background Technology

[0002] As the core component of industrial filtration systems, the reliability and consistency of filter elements directly affect the quality of end products and even the long-term service life of major equipment. The traditional method of filter element inspection is to completely immerse the filter element in a transparent container containing a specific test liquid, introduce compressed gas into the filter element, and rely on the inspector to visually observe whether bubbles are generated on the outer surface of the filter element, as well as observe the shape, size and distribution of the bubbles, and then judge whether the filter element is qualified based on experience.

[0003] However, the immersion and aeration method widely used for filter element integrity testing currently has three main limitations: First, this method relies on operators to visually observe the shape and distribution of bubbles, resulting in low testing efficiency and difficulty in adapting to large-scale continuous production. Furthermore, due to the strong subjectivity of the judgment criteria, different inspectors often have different standards for judging "qualified" and "defective," lacking a unified and quantitative basis. Second, the test liquid often contains toxic and volatile components, and long-term close exposure to such an environment may pose health risks and occupational health and safety hazards to operators. Third, the entire filter element testing process lacks an automated data recording mechanism, making it impossible to objectively quantify bubble information or generate traceable test reports, thus failing to meet the requirements of modern quality control systems for data integrity and process traceability.

[0004] Therefore, there is an urgent need to propose a filter cartridge filtration performance testing method and system based on visual inspection to achieve automated detection of filter cartridge bubbles, replace manual operation, shorten the testing cycle, and improve the accuracy and consistency of bubble identification. Summary of the Invention

[0005] One objective of this invention is to provide a filter cartridge filtration performance testing method and system based on visual inspection, which enables automated detection of air bubbles in the filter cartridge, replaces manual operation, shortens the testing cycle, and improves the accuracy and consistency of air bubble identification.

[0006] According to one aspect of the present invention, a method for testing the filtration performance of a filter cartridge based on visual inspection is provided, comprising the following steps:

[0007] S1. Place the surface light source and multiple cameras on both sides of the transparent detection cylinder. Arrange the multiple cameras side by side along the preset filter length direction. Perform camera calibration to obtain a calibration parameter set. The calibration parameter set includes the internal parameters of each camera, lens distortion coefficient, and homography matrix between adjacent camera images.

[0008] S2. Completely immerse the filter element to be tested into the transparent testing liquid in the transparent testing cylinder;

[0009] S3. Control all cameras to synchronously acquire original images, perform distortion correction on the original images based on the calibration parameter set, and use the homography matrix to stitch the images to generate a panoramic image of the bubble generation area around the filter element to be tested, and vertically divide the panoramic image into multiple independent detection zones.

[0010] S4. Apply a first gas pressure lower than the nominal bubble point pressure of the filter element to be tested to the inside of the filter element, execute step S3 to obtain a first reference image, perform image processing on the image corresponding to each detection zone in the first reference image, count the single bubble areas in each detection zone whose pixel area exceeds the single bubble area threshold, summarize the pixel areas of the single bubble areas that meet the conditions to form a first total bubble area value, if the first total bubble area value is higher than the first area threshold, it is determined that the filter element to be tested has a pore defect in the detection zone.

[0011] S5. Apply a second gas pressure not lower than the nominal bubble point pressure of the filter element to be tested. Execute step S3 to obtain a second reference image. Use a bubble segmentation model trained based on deep learning to perform semantic segmentation on the images corresponding to each detection partition in the second reference image. Extract the total area of ​​pixels identified as bubbles in each detection partition as the second total bubble area value. If the second total bubble area value in any detection partition is lower than the second area threshold, it is determined that the filter element under test has a clogging defect or insufficient porosity in that detection partition.

[0012] Preferably, the step of performing camera calibration to obtain a calibration parameter set includes:

[0013] A1. Using a calibration board, acquire pose images in various poses within the field of view of each camera, and calculate the internal parameter matrix and lens distortion coefficients of each camera based on all pose images.

[0014] A2. Place a transparent calibration substrate that covers the field of view of all cameras between the surface light source and the multiple cameras. The calibration substrate has multiple sets of graphic and text labels for providing matching features. Place the calibration board in front of the calibration substrate. Each camera's field of view contains only one calibration board and has an overlapping area with the adjacent camera's field of view containing the same graphic and text labels.

[0015] A3. Control all cameras to synchronously acquire the original images of the calibration substrate. After distortion correction of the original images based on the internal parameter matrix of each camera and the lens distortion coefficient, the calibration images are obtained. The feature points in each calibration image are obtained by using a feature point extraction algorithm. After preliminary matching of the feature points of adjacent calibration images, the correct matching feature point pairs are selected by using a mismatch filtering algorithm. The homography matrix between each pair of adjacent calibration images is calculated based on the selected matching feature point pairs.

[0016] A4. Store the internal parameters and distortion coefficients of all cameras obtained in step A1, together with the homography matrix between all adjacent cameras obtained in step A3, as a calibration parameter set.

[0017] Preferably, the step of obtaining feature points in each calibration image includes the following steps:

[0018] B1. Calculate the grayscale change function when a local window in the calibration image moves in each direction. E(u,v) :

[0019]

[0020] in, (x, y) To calibrate the coordinates of pixels in an image, (you, you) This indicates the minute displacement of the window in the horizontal and vertical directions. W(i, j) For window functions, I(i, j) Indicates at point (i, j) grayscale value at that location w The window radius is 3 to 7 pixels.

[0021] B2. Construct the Harris matrix based on the grayscale change function. M :

[0022]

[0023] in, I x (i, j) and I y (i, j) At points respectively (i, j) The gradient values ​​at the location and along the x and y directions;

[0024] B3. According to the matrix M Calculate the corner response function R , ,in (M) For matrix M The determinant, trace(M) For matrix M traces, k Let be an empirical constant, RPoints whose values ​​exceed a set threshold are identified as feature points.

[0025] Preferably, the homography matrix between the two adjacent calibration images is implemented using a random sampling consensus algorithm, including the following steps:

[0026] C1. Randomly select four pairs of matching points from the initially matched feature point pairs as the minimum sample set to calculate the initial homography matrix model parameters;

[0027] C2. Calculate the error of all feature point pairs with the initial homography matrix model, count the number of inliers whose error is less than the set tolerance, repeat the above random sampling, model fitting and consistency test process, and retain the model with the most inliers as the optimal model.

[0028] C3. The calculation terminates when the maximum number of iterations N is reached. ,in p To achieve the expected success rate, ω This is an estimate of the proportion of interior points. n To obtain the minimum number of samples required to fit the model, the homography matrix is ​​finally obtained by refitting with all interior points to calculate the transformation relationship between adjacent camera images.

[0029] Preferably, step S4 includes the following steps:

[0030] S40. Apply a first gas pressure lower than the nominal bubble point pressure of the filter element to be tested, and execute step S3 to obtain the first reference image.

[0031] S41. The first partition image is obtained by smoothing the images corresponding to each detection partition in the first reference image using a mean filtering algorithm.

[0032] S42. The first partition image is processed using a dynamic threshold segmentation algorithm. The adaptive threshold of each pixel is calculated based on the local neighborhood. Each single bubble region is identified and segmented by a connected component analysis algorithm.

[0033] S43. Calculate the pixel area of ​​each single bubble region, count the single bubble regions in each first partition image whose pixel area is greater than the single bubble area threshold, and summarize the pixel areas of the single bubble regions that meet the conditions to form the first total bubble area value. If the first total bubble area value is higher than the first area threshold, it is determined that the filter element under test has a hole defect in the corresponding inspection partition. The single bubble area threshold and the first area threshold are both preset pixel area values.

[0034] Preferably, the method for setting the single bubble area threshold is to determine the critical defect pore size based on the quality standard of the filter element product to be tested. D 0Calculate the theoretical value of bubble volume ,in V B This is the theoretical value of the bubble volume. c 1 The dimensionless coefficient is related to factors such as the gas-liquid interface morphology. σ To detect the surface tension of the liquid, To detect the liquid density, Bubble density, g For gravitational acceleration, the corresponding theoretical projected area of ​​the bubble Based on the pixel size and optical magnification obtained from camera calibration, the theoretical projected area of ​​the bubble is converted into the corresponding pixel area, and the pixel area is used as the single bubble area threshold.

[0035] Preferably, step S5 includes the following steps:

[0036] S50. Apply a second gas pressure not lower than the nominal bubble point pressure of the filter element to be tested, and execute step S3 to obtain a second reference image.

[0037] S51. Using a multi-combination filter cartridge to collect a panoramic image under the second gas pressure and processing it in step S3, the panoramic image used for training is used as a training image. According to the size of the detection partition, the panoramic image used for training is cropped into several standard-sized training images. The bubble regions in the training images are manually labeled to generate corresponding binary mask labels.

[0038] S52. The U-Net convolutional neural network with encoder-decoder architecture is used as the semantic segmentation model. The training image is used as input and the corresponding binary mask label is used as the training target. The network parameters are optimized using a combination loss function and the bubble segmentation model is trained by backpropagation algorithm.

[0039] S53. After preprocessing the images corresponding to each detection partition in the second reference image, the second partition image is obtained. The second partition image is then input into the trained bubble segmentation model to obtain a pixel-level probability map.

[0040] S54. The pixel-level probability map is binarized and segmented to obtain the final bubble region segmentation mask. The total number of pixels identified as bubbles in the segmentation mask corresponding to each second partition image is counted to obtain the second total bubble area value of the detection partition. If the second total bubble area value in any detection partition is lower than the second area threshold, it is determined that the filter under test has a blockage defect or insufficient opening rate in the detection partition.

[0041] Preferably, the combined loss function L is composed of the cross-entropy loss function. L ceand Dice loss function L Dice The weighted summation is calculated using the following formula: ,in α、β These are the weighting coefficients, and α+β = 1, and the Adam optimizer is used to perform the backpropagation process. Based on the partial derivatives of the combined loss function L with respect to the network prediction results and the real labels, the weight parameters of the semantic segmentation model are iteratively optimized until the semantic segmentation model converges.

[0042] Preferably, the first gas pressure is less than 0.95 times the nominal bubble point pressure of the filter element, and the second gas pressure is set to more than 1.05 times the nominal bubble point pressure of the filter element. ,in D 1 The nominal accuracy of the filter element. θ Contact angle ,η This is the hole shape correction factor.

[0043] According to another aspect of the present invention, a visual inspection-based filter cartridge filtration performance testing system is also provided for implementing the above-described method, comprising:

[0044] A transparent cylinder is used to hold a transparent test liquid and accommodate the filter element to be tested.

[0045] A backlight is located on one side of the transparent cylinder to provide uniform backlight illumination to the surface of the filter element.

[0046] The air pressure control unit is connected to the filter element under test and is used to introduce gas into the filter element under test and precisely control the first gas pressure and the second gas pressure.

[0047] The image acquisition module includes several industrial cameras arranged side by side;

[0048] An image processing module, connected to the image acquisition module, includes:

[0049] The image preprocessing unit is used to perform distortion correction and image stitching on the images acquired by the camera based on the system's calibration parameter set, generate a panoramic image of the bubble generation area, divide the panoramic image into multiple independent detection zones, and output the image set of each detection zone.

[0050] The bubble detection unit is used to receive the first reference image and identify it to obtain single bubble regions with pixel areas greater than the single bubble area threshold, and output a line chart containing the first total bubble area value in each detection zone.

[0051] The uniformity detection unit is used to receive the second reference image, perform identification, calculate the second total bubble area value of each detection zone, and output a line chart containing the second total bubble area value of each detection zone.

[0052] The present invention has at least the following beneficial effects:

[0053] First, this invention employs multiple industrial cameras arranged in parallel along the filter element axis, forming a uniform illumination system with a high-brightness surface light source. Through precise optical calibration and image stitching algorithms, a panoramic image of the bubble generation area on the filter element surface is constructed, achieving full automation of the detection process. This transforms the process, which previously required long-term manual visual observation, into one that is automatically completed by the machine, significantly improving the detection speed and meeting the efficiency requirements of large-scale production. At the same time, by quantitatively analyzing the pixel area and number of bubbles in each detection zone, the system eliminates subjective judgment differences between different operators, ensuring the consistency and reliability of the detection results.

[0054] Secondly, this invention has a high degree of automation and is fully compatible with non-contact automated testing. By using automated equipment such as robotic arms and robots to replace the filter element, operators only need to complete the work of filling the robot's hopper and obtaining the separated workpieces. Then, the robot fills the workpieces, and the subsequent application of gas pressure, image acquisition, and processing analysis can all be completed automatically by the system. This keeps operators away from the volatile gas environment of the test liquid, effectively avoiding the health risks that may be caused by long-term contact with industrial test liquids and other test liquids containing toxic volatile components, and significantly improving the safety level of the testing operation.

[0055] Third, a comprehensive data acquisition and recording mechanism has been established, which can automatically save panoramic images of bubble formation on the filter element surface and quantitative data of each testing zone, including key parameters such as the first and second total bubble area values. This data is automatically linked with metadata such as product number, testing time, and pressure parameters to generate structured testing reports, achieving comprehensive data recording of the testing process and providing reliable data support for product quality traceability, process optimization, and quality system certification.

[0056] Fourth, addressing the technical challenges of filter element defect detection, the system innovatively employs differentiated image processing strategies. For pore defects, dynamic threshold segmentation and connected component analysis algorithms are used to effectively identify and count the number of large air bubbles; for blockage defects, a deep learning-based bubble segmentation model is used to accurately identify and calculate the area of ​​tiny air bubbles. Multi-camera calibration technology solves the problem of curved surface imaging distortion, and a bubble threshold is scientifically set based on a fluid dynamics model, forming a comprehensive solution capable of simultaneously and accurately detecting multiple defect types, thus improving the scientific rigor and adaptability of the detection system.

[0057] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of a system in one technical solution of the present invention;

[0059] Figure 2 This is a schematic diagram of camera calibration in one technical solution of the present invention, wherein (a) is a schematic diagram of calibration image and (b) is a schematic diagram of panoramic image;

[0060] Figure 3 This is an image stitching effect diagram in an embodiment of the present invention, wherein (a) is the image before stitching and (b) is the panoramic image after stitching;

[0061] Figure 4 This is a graph showing the test results of the filter element under test under the first gas pressure in an embodiment of the present invention;

[0062] Figure 5 This is an enlarged view of part A in an embodiment of the present invention;

[0063] Figure 6 This is a graph showing the test results of the filter element under test under the second gas pressure in an embodiment of the present invention;

[0064] Figure 7 This is an enlarged view of part B in an embodiment of the present invention. Detailed Implementation

[0065] The present invention will now be described in further detail so that those skilled in the art can implement it based on the description.

[0066] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0067] like Figure 1 As shown, the present invention provides a method for testing the filtration performance of filter cartridges based on visual inspection, characterized by comprising the following steps:

[0068] S1. Place the surface light source and multiple cameras on both sides of the transparent detection cylinder. The multiple cameras are arranged side by side along the preset filter length direction. Perform camera calibration to obtain a calibration parameter set. The calibration parameter set includes the internal parameters of each camera, lens distortion coefficient, and homography matrix between adjacent camera images. Since the camera lens inevitably has distortion, the relative positions between multiple cameras are not strictly aligned. If precise calibration is not performed and these geometric errors are not corrected, the subsequent panoramic image will have serious distortion and misalignment, making it impossible to accurately calculate the area and position of the bubble, ultimately affecting the accuracy of defect judgment.

[0069] Specifically, firstly, a high-brightness surface light source is fixed to one side of the transparent detection cylinder, and multiple industrial cameras are arranged side-by-side on the other side of the cylinder, ensuring that the optical axes of all cameras are approximately perpendicular to the axis of the filter element. Then, camera calibration is performed. Camera calibration can employ a multi-view geometry-based calibration method, using a known, precisely sized checkerboard or dot array calibration board. Multiple posture images are captured by moving the camera within its field of view in various poses. Existing algorithms are used to process these posture images to determine the internal parameters and lens distortion coefficients of each camera. Next, by placing calibration boards or special markers within the common field of view of all cameras, a suitable existing algorithm is used to calculate the homography matrix between adjacent camera images. Finally, all parameters are summarized and stored as a calibration parameter set for subsequent image processing.

[0070] S2. Immerse the filter element to be tested completely in the transparent testing liquid of the transparent testing cylinder. Specifically, the transparent testing liquid is a clear and transparent liquid with stable physical properties such as surface tension. Deionized water or an alcohol aqueous solution of a specific concentration can be selected with reference to the relevant technical specifications.

[0071] S3. Control all cameras to synchronously acquire original images. Based on the calibration parameter set, perform distortion correction on the original images and use the homography matrix to stitch the images together to generate a panoramic image of the bubble generation area around the filter element under test. Divide this panoramic image vertically into multiple independent detection zones. Specifically, trigger all cameras to perform synchronous exposure and acquire a complete set of original images covering the filter element under test and its upper area. Then, call the calibration parameter set obtained in S1, first perform distortion correction on each original image to eliminate barrel or pincushion distortion of the lens, and then use the homography matrix in the calibration parameter set to project all corrected images onto a unified coordinate system. Use an image fusion algorithm to smooth the pixels in the overlapping areas, and finally generate a clear and seamless panoramic image of the bubble generation area around the filter element. In order to accurately locate the specific section where the defect occurs and realize a complete data acquisition and recording mechanism, the panoramic image is vertically divided into multiple independent monitoring zones.

[0072] S4. Apply a first gas pressure lower than the nominal bubble point pressure of the filter element to be tested. Execute step S3 to obtain a first reference image. Perform image processing on the images corresponding to each detection zone in the first reference image. Count the single bubble areas in each detection zone whose pixel area exceeds the single bubble area threshold. Summarize the pixel areas of the single bubble areas that meet the conditions to form a first total bubble area value. If the first total bubble area value is higher than the first area threshold, it is determined that the filter element under test has a pore defect in that detection zone. The first gas pressure refers to a test pressure lower than the nominal bubble point pressure of the filter element. Under this pressure, the gas at the intact filter membrane cannot break through the liquid membrane to form bubbles, while the pore defect of the filter element will continuously generate large bubbles due to the large pore size. The single bubble area threshold is a preset pixel area value used to distinguish between large bubbles generated by pores and possible noise or small bubbles.

[0073] Specifically, the pressure applied to the filter element under test is precisely adjusted to a first gas pressure and kept stable by the air pressure control unit. Then, the image acquisition and processing flow in S3 is executed once to obtain a first reference image under the current pressure. The first reference image is a stitched and partitioned image. Subsequently, the first partition image corresponding to each detection partition is processed. Image filtering algorithms can be used to smooth the image to suppress noise. Then, image segmentation algorithms are used to separate the bubble foreground and background in the image. Each independent bubble region is identified through connected component analysis, and its pixel area is calculated. All single bubble regions in the detection partition with pixel areas greater than the single bubble area threshold are counted. The total pixel area value of all single bubble regions that meet the condition is counted as the first total bubble area value and compared with the preset first area threshold. If the first total bubble area value of a certain detection partition exceeds the first area threshold, it is determined that there is a pore defect in the filter element part corresponding to the detection partition. Using a lower pressure and focusing on detecting large bubbles can capture pore defects very specifically, while effectively avoiding misjudging normally generated small bubbles as defects, ensuring high specificity and low false alarm rate in this detection process.

[0074] S5. Apply a second gas pressure not lower than the nominal bubble point pressure of the filter element to be tested. Execute step S3 to obtain a second reference image. Use a bubble segmentation model trained based on deep learning to perform semantic segmentation on the images corresponding to each detection zone in the second reference image. Extract the total area of ​​pixels identified as bubbles in each detection zone as the second total bubble area value. If the second total bubble area value in any detection zone is lower than the second area threshold, it is determined that the filter element under test has a blockage defect or insufficient porosity in that detection zone. The second gas pressure is a test pressure not lower than the nominal bubble point pressure of the filter element. Under this pressure, gas will break through on the filter membrane in the intact area of ​​the filter element under test to form a large number of tiny bubbles. The second area threshold is a preset pixel area value, representing the minimum total bubble area that a detection zone should produce under normal conditions. The physical manifestations of blockage defects and pore defects are completely opposite. It causes the number of bubbles generated at the blockage point of the filter element to decrease or disappear completely. Therefore, sufficient bubbles must be generated in the normal area under pressure above the bubble point to serve as a background contrast. Since the tiny bubbles are intertwined and difficult to distinguish at this point, traditional image processing algorithms struggle to handle the number of bubbles. Thus, a deep learning semantic segmentation model is needed to process the image.

[0075] Specifically, the pressure applied inside the filter element is precisely adjusted to the second gas pressure and kept stable by the air pressure control unit. Then, the image acquisition and processing flow in S3 is executed again to obtain the second reference image. A bubble segmentation model pre-trained using deep learning technology is called. The bubble segmentation model is a semantic segmentation model that can perform pixel-level classification of the input image, classifying each pixel as either "bubble" or "background". The second partition image corresponding to each detection partition is input into the model to obtain a binary segmentation mask image of the bubble region. The total number of pixels in the segmentation mask image that are determined to be "bubbles" is calculated to obtain the second total bubble area value of the partition. The second total bubble area value is compared with a second area threshold. If the second total bubble area value of a certain detection partition is lower than the threshold, it is determined that there is a blockage defect in the filter element part corresponding to the detection partition.

[0076] In this technical solution, the units of the first area threshold, the second area threshold, the single bubble area threshold, the first total bubble area value, and the second total bubble area value are unified. Optionally, when the camera pixel size and optical magnification are unified, the number of pixels can be used to represent the area.

[0077] The following are specific on-site examples:

[0078] Specific implementation examples are as follows:

[0079] A petrochemical company's filter regeneration workshop uses a transparent acrylic testing cylinder measuring 2 meters long, 0.8 meters wide, and 0.5 meters high. The cylinder contains industrial ethanol testing solution that meets industry standards. A 20,000 lux LED surface light source is installed on one side of the cylinder to provide uniform backlighting to the testing area. On the other side of the cylinder, six 20-megapixel industrial cameras are installed side-by-side along its length. The cameras are fixed with brackets to ensure their optical axes are perpendicular to the axis of the filter element under test. The six cameras can completely cover the entire circumferential surface of the filter element. A high-precision air pressure control unit is connected to the filter element's interface via pipelines.

[0080] Before the system was officially put into operation, a one-time calibration process was first executed to calculate and store the internal parameters of each camera, the lens distortion coefficients, and the homography matrix between adjacent camera images. These parameters together constitute the calibration parameter set of the system. The calibration parameter set can be directly called in subsequent detection without repeated calibration.

[0081] The operator fully immerses a 1.5-meter-long, DN25-diameter, 0.5μm-precision sintered metal powder filter element to be tested into the test liquid in the test cylinder, and ensures that the interface of the filter element to be tested is reliably connected to the air pressure pipeline.

[0082] Six cameras are simultaneously triggered to acquire six original images covering the filter to be tested. A calibration parameter set is then called to correct distortion in the six images. Finally, a pre-stored homography matrix is ​​used to stitch the images together to generate a seamless panoramic image, as shown below. Figure 4 As shown, the panoramic image was then divided into 80 consecutive detection zones at equal intervals along its length.

[0083] The air pressure control unit applies a gas pressure of 0.05 MPa to the inside of the filter element under test. After the pressure stabilizes, six cameras are simultaneously triggered within microseconds to acquire six raw images covering the surface of the filter element. These images are then stitched together to form the first reference image. A dynamic threshold segmentation and connected component analysis algorithm is used to quickly identify and calculate the pixel area of ​​each independent bubble in each detection zone. The single bubble area threshold is set to 500 pixels; bubbles with an area greater than 500 pixels are classified as large bubbles. The first area threshold for each detection zone is set to 25,000 pixels (50 bubbles). A line graph showing the number of all detection zones is generated. The detection results are as follows: Figure 4 , Figure 5 As shown, where Figure 4 The vertical axis of the line graph represents the number of bubbles exceeding 500 pixels, in units of individual bubbles. As can be seen from the graph, the number of large bubbles with a pixel area exceeding the single bubble area threshold in detection zone number 72 is 57. The first total bubble area value exceeds the first area threshold, indicating that the filter element has a pore defect in detection zone number 72.

[0084] The air pressure control unit applies a gas pressure of 0.2 MPa to the inside of the filter element under test, triggering all cameras again to form a second baseline image. The pre-trained U-Net deep learning semantic segmentation model is then used to process the image of each detection region, generating a segmentation mask for the bubble region. The total area of ​​pixels identified as bubbles within each detection region is calculated. The second area threshold for each detection region is set to 500,000 pixels. The detection results are as follows: Figure 6 , Figure 7 As shown, where Figure 6 The vertical axis of the line graph represents the total area of ​​pixels identified as bubbles, in thousands of pixels. It was found that the second total bubble area value of all detection zones was higher than the second area threshold, indicating that the filter element under test did not have defects such as blockage or insufficient porosity.

[0085] The entire testing process is completed within 35 seconds, automatically generating a test report containing the filter element number, testing time, pressure curve, defect location information with panoramic annotations, and the final judgment result, which is then stored in the database. Operators can query and access historical test data at any time based on the filter element number, successfully replacing the original manual visual inspection position. The efficiency of single filter element testing is increased by more than 10 times, and the safety risks of operators coming into contact with toxic ethanol liquid are completely eliminated. Filter element quality is determined through objective and quantitative image data, with 100% consistency in results, solving the problem of strong subjectivity in manual judgment. All test data is digitized, perfectly meeting the petrochemical company's needs for strict traceability and control of filter element quality.

[0086] In another technical solution, the steps of performing camera calibration to obtain a calibration parameter set include:

[0087] A1. Using a calibration board, acquire pose images in various poses within the field of view of each camera. Calculate the internal parameter matrix and lens distortion coefficients of each camera based on all pose images.

[0088] Specifically, during implementation, the operator holds a high-precision calibration board with a checkerboard pattern and moves it sequentially within the field of view of each camera at various angles and distances. Each change in posture triggers the camera to capture an image, ensuring that the calibration board appears in different positions and orientations in the image. 15-25 posture images with different postures are needed for each camera. Then, existing algorithms based on planar templates, such as the Zhang Zhengyou algorithm, are used to process these posture images. The algorithm identifies the checkerboard corner points in each posture image and uses the correspondence between the image coordinates of these corner points and their known world coordinates. Through mathematical methods such as least squares optimization, the high-precision internal parameter matrix and lens distortion coefficient of the camera are finally solved.

[0089] A2. A transparent calibration substrate covering the field of view of all cameras is placed between the surface light source and the multiple cameras. This calibration substrate has multiple sets of graphic symbols for providing matching features. The calibration board is placed in front of the calibration substrate, with each camera's field of view containing only one calibration board and overlapping areas with adjacent camera fields of view containing the same graphic symbols. The calibration substrate is a material with good light transmittance and flatness. The graphic symbols are patterns printed on the calibration substrate that provide rich and stable features for the feature extraction algorithm, such as... Figure 2 As shown, specifically, a calibration substrate is placed between the installed surface light source and the camera array to ensure that the field of view of any two adjacent cameras must contain at least one set of identical graphic symbols. At the same time, a standard calibration plate is placed on this transparent substrate to ensure that in the independent field of view of each camera, there is one and only one complete calibration plate located in the central area of ​​the field of view.

[0090] A3. Control all cameras to synchronously acquire the original images of the calibration substrate. After distortion correction of the original images based on the internal parameter matrix of each camera and the lens distortion coefficient, the calibration images are obtained. Feature point extraction algorithm is used to obtain feature points in each calibration image. After preliminary matching of feature points of adjacent calibration images, the wrong match filtering algorithm is used to filter out the correct matching feature point pairs. Based on the filtered matching feature point pairs, the homography matrix between each pair of adjacent calibration images is calculated. Feature points are pixels in the calibration images that have significant characteristics and are easy to be repeatedly located in different images. The wrong match filtering algorithm is a robust algorithm used to remove wrong matching pairs from the preliminary matching results. The homography matrix is ​​a 3x3 matrix that shows the mapping relationship between the image pixel coordinates when two different viewpoints image the same plane.

[0091] Specifically, all cameras are controlled to capture images simultaneously, acquiring raw images of a substrate with calibration plates and graphic markings in a single step. The internal parameter matrices and lens distortion coefficients of each camera obtained in step A1 are used to perform distortion correction on each raw image, resulting in a geometrically accurate calibration image. Then, a feature point extraction algorithm, such as Harris corner detection, SIFT algorithm, or ORB algorithm, is selected to extract a large number of feature points from each calibration image. A mismatch filtering algorithm is used to refine the preliminary matching results, automatically selecting correct, high-confidence matching feature point pairs. Finally, using these correct matching feature points, an optimal homography matrix is ​​calculated through algorithms such as direct linear transformation. This matrix can accurately map points in one image to the coordinate system of another image.

[0092] A4. Store the intrinsic parameters and distortion coefficients of all cameras obtained in step A1, together with the homography matrix between all adjacent cameras obtained in step A3, as a calibration parameter set. Pack the intrinsic parameter matrix and lens distortion coefficient of each camera obtained in step A1, together with the homography matrix between every two adjacent images obtained in step A3, into a structured calibration parameter set.

[0093] In another technical solution, obtaining feature points in each calibration image includes the following steps:

[0094] B1. Calculate the grayscale change function when a local window in the calibration image moves in each direction. E(u,v) :

[0095]

[0096] in, (x, y) To calibrate the coordinates of pixels in an image, (you, you) This indicates the minute displacement of the window in the horizontal and vertical directions. W(i, j) For window functions, I(i, j) Indicates at point (i, j) grayscale value at that location w The window radius is 3 to 7 pixels. Specifically, it iterates through every pixel in the calibration image, and for each point to be judged... (x,y) A local window is extracted centered on this point, and the grayscale change function is calculated after the window undergoes minute displacements in different directions. E(u, v) To assess whether the point might be a corner point, where W(x,y) Pixels in the center of the window are given higher weights, while pixels further away from the center are given lower weights. This makes the calculation of grayscale changes focus more on the center of the window.

[0097] B2. Construct the Harris matrix based on the grayscale change function. M :

[0098]

[0099] in, I x (i, j) and I y (i, j) At points respectively (i, j) place, along x and y The gradient value in the direction.

[0100] B3. According to the matrix M Calculate the corner response function R , ,in (M) For matrix M The determinant, trace(M) For matrix M traces, k Let be an empirical constant, R Points whose values ​​exceed a set threshold are identified as feature points. Specifically, by calculating the corner response function, corner regions in an image can be effectively distinguished from flat or edge regions. When the corner response function value exceeds the set threshold, the point is identified as a feature point, thereby enabling the extraction of stable feature points in the image. The setting of the preset threshold needs to take into account both the sensitivity and specificity of feature point detection. If the threshold is too high, it may lead to the missed detection of real feature points, while if the threshold is too low, it may introduce too many noise points.

[0101] In another technical solution, the homography matrix between the two adjacent calibration images is achieved through a random sampling consensus algorithm, including the following steps:

[0102] C1. Randomly select four pairs of matching points from the initially matched feature point pairs as the minimum sample set to calculate the initial homography matrix model parameters. For example, the homography matrix H is a 3×3 matrix that describes the projection transformation relationship between two planes, with 8 degrees of freedom. Therefore, at least four pairs of matching feature points are needed to solve for the model parameters. Solve the system of equations using the direct linear transformation algorithm:

[0103]

[0104] in and These are the coordinates of a pair of matched feature points in the two calibration images. Random sampling is used to avoid interference from incorrect matching points in the model calculation, providing initial parameter estimates for subsequent model optimization.

[0105] C2. Calculate the error between all feature point pairs and the initial homography matrix model. Statistically count the number of inliers whose error is less than a set tolerance. Repeat the above random sampling, model fitting, and consistency check process. Retain the model with the largest number of inliers as the optimal model. For example, calculate the projection error between all feature point pairs and the initial homography matrix model. For the ... i For the matching point, the formula for calculating its projection error is:

[0106]

[0107] The `project` function represents the transformation from homogeneous coordinates to Cartesian coordinates, with the calculation error less than a set tolerance. τ (Select 1-3 pixels) of interior points. Repeat the random sampling, model fitting, and consistency check process, and retain the model with the most interior points as the current optimal model.

[0108] C3. The calculation terminates when the maximum number of iterations N is reached. ,in p To achieve the expected success rate, ω This is an estimate of the proportion of interior points. n To obtain the minimum number of samples required to fit the model, the homography matrix is ​​finally refitted using all inliers to calculate the transformation relationship between adjacent camera images. For example, where p We set the value to 0.99 and finally refitted the homography matrix using the least squares method with all interior points.

[0109] In another technical solution, step S4 includes the following steps:

[0110] S40. Apply a first gas pressure lower than the nominal bubble point pressure of the filter element to be tested, and execute step S3 to obtain the first reference image.

[0111] S41. The first partition image is obtained by smoothing the images corresponding to each detection partition in the first reference image using a mean filtering algorithm. For example, the mean filtering uses a sliding window of size m×n, and the arithmetic mean of the gray values ​​of all pixels in the sliding window is taken as the new gray value of the center pixel.

[0112] S42. The first partition image is processed using a dynamic threshold segmentation algorithm. An adaptive threshold is calculated for each pixel based on its local neighborhood. Each single bubble region is identified and segmented using a connected component analysis algorithm. For example, the adaptive threshold is calculated based on the local neighborhood of each pixel. T(x,y) The calculation formula is as follows: ， in μ (x,y) and σ(x,y) These represent the mean and standard deviation of the pixel grayscale values ​​within the local neighborhood, respectively. C The constant adjustment factor is used to binarize the image by comparing the gray value of each pixel with the corresponding adaptive threshold. Then, a connected component analysis algorithm is used to identify and segment each independent single bubble region through region growing and labeling techniques, accurately separating bubbles that are in contact or overlapping.

[0113] S43. Calculate the pixel area of ​​each single bubble region, count the single bubble regions in each first partition image whose pixel area is greater than the single bubble area threshold, and summarize the pixel areas of the single bubble regions that meet the conditions to form the first total bubble area value. If the first total bubble area value is higher than the first area threshold, it is determined that the filter element under test has a hole defect in the corresponding inspection partition. The single bubble area threshold and the first area threshold are both preset pixel area values.

[0114] In another technical solution, the method for setting the single bubble area threshold is to determine the critical defect pore size according to the quality standard of the filter element product to be tested. D 0 Calculate the theoretical value of bubble volume ,in V B This is the theoretical value of the bubble volume. c 1 The dimensionless coefficient is related to factors such as the gas-liquid interface morphology. σ To detect the surface tension of the liquid, To detect the liquid density, Bubble density, g For gravitational acceleration, the corresponding theoretical projected area of ​​the bubble Based on the pixel size and optical magnification obtained from camera calibration, the theoretical projected area of ​​the bubble is converted into the corresponding pixel area. This pixel area is then used as the single bubble area threshold. Specifically, a calculation method based on a physical model is employed to set the single bubble area threshold. c 1 An empirical value of 0.93 can be used. After obtaining the theoretical projected area of ​​the bubble, the actual physical area corresponding to a single pixel is first calculated using the pixel size and optical magnification parameters obtained through camera calibration. Then, the number of pixels corresponding to the theoretical projected area of ​​the bubble, i.e., the single bubble area threshold, can be obtained.

[0115] In another technical solution, step S5 includes the following steps:

[0116] S50. Apply a second gas pressure not lower than the nominal bubble point pressure of the filter element to be tested, and execute step S3 to obtain a second reference image.

[0117] S51. Using panoramic images obtained by collecting multiple sets of filter cartridges under the second gas pressure and processing them in step S3, the panoramic images are used as training images. According to the size of the detection partition, the panoramic images are cropped into several standard-sized training images. The bubble regions in the training images are manually labeled to generate corresponding binary mask labels. For example, firstly, panoramic images of no less than 100 sets of filter cartridge samples collected and processed under the second gas pressure are collected as training panoramic images. Then, according to the actual physical size of the detection partition, a bicubic interpolation algorithm is used to uniformly crop all the training panoramic images into standard-sized training images of 512×512 pixels. During the labeling process, LabelMe is used to outline the bubble regions in each training image to generate PNG format binary mask label files, where the pixel value of the bubble region is 255 and the pixel value of the background region is 0.

[0118] S52. A U-Net convolutional neural network with an encoder-decoder architecture is used as the semantic segmentation model. The training image is used as input and the corresponding binary mask label is used as the training target. The network parameters are optimized using a combined loss function. The bubble segmentation model is trained by backpropagation algorithm. For example, the encoder part uses a VGG16 backbone as the feature extractor, which contains 13 convolutional layers and 4 max pooling layers. It extracts image features step by step while halving the size at each level. The decoder part upsamples through 4 transposed convolutions. Each level is skip-connected to the image of the corresponding level of the encoder to fuse low-level detail features and high-level semantic features. The batch size of the model training is set to 16, the initial learning rate is 0.001, and the Adam optimizer is used to optimize the parameters. The loss function is a linear combination of cross-entropy loss and Dice loss. The training process lasts for 100 rounds. Every 20 rounds, the learning rate is reduced to half of the original value. The best-performing model parameters are saved on the validation set.

[0119] S53. After preprocessing the images corresponding to each detection partition in the second reference image, a second partition image is obtained. The second partition image is input into the trained bubble segmentation model to obtain a pixel-level probability map. For example, the images corresponding to each detection partition in the second reference image to be detected are standardized, including adjusting the size of the second partition image to 512×512 pixels and normalizing the pixel values ​​to the range of [0,1]. The processed second partition image is input into the trained bubble segmentation model. The model output is a probability map of the same size as the input image. The value of each pixel represents the probability value of the point belonging to the bubble category.

[0120] S54. The pixel-level probability map is binarized and segmented to obtain the final bubble region segmentation mask. The total number of pixels identified as bubbles in the segmentation mask corresponding to each second partition image is counted to obtain the second total bubble area value of the detection partition. If the second total bubble area value in any detection partition is lower than the second area threshold, it is determined that the filter under test has a blockage defect or insufficient opening rate in the detection partition. For example, the probability map is converted into a binary segmentation mask, and then 8-connected component analysis is used to post-process the segmentation result to remove noise points with an area of ​​less than 10 pixels and fill the possible holes inside the bubble region. Finally, the total number of all pixels with a value of 1 in the segmentation mask corresponding to each second partition image is counted to obtain the second total bubble area value of the partition.

[0121] In another technical solution, the combined loss function L is derived from the cross-entropy loss function. L ce and Dice loss function L Dice The weighted summation is calculated using the following formula: ,in α、β These are the weighting coefficients, and α+β = 1, and the Adam optimizer is used to perform the backpropagation process. Based on the partial derivatives of the combined loss function L with respect to the network prediction results and the real labels, the weight parameters of the semantic segmentation model are iteratively optimized until the model converges.

[0122] In another technical solution, the first gas pressure is less than 0.95 times the nominal bubble point pressure of the filter element, and the second gas pressure is set to more than 1.05 times the nominal bubble point pressure of the filter element. ,in D 1 The nominal accuracy of the filter element. θ Contact angle ,η This is the hole shape correction factor. the A value ranging from 0.5 to 1.0 is used to correct for the geometric differences between ideal cylindrical channels and actual irregular channels. θ Based on experience, a value of 0 can be chosen.

[0123] In another technical solution, a filter cartridge filtration performance testing system based on visual inspection is also provided for implementing the above method, including:

[0124] A transparent cylinder is used to hold a transparent test liquid and accommodate the filter element to be tested.

[0125] A backlight is located on one side of the transparent cylinder to provide uniform backlight illumination to the surface of the filter element.

[0126] The air pressure control unit is connected to the filter element under test and is used to introduce gas into the filter element under test and precisely control the first gas pressure and the second gas pressure.

[0127] The image acquisition module includes several industrial cameras arranged side by side;

[0128] An image processing module, connected to the image acquisition module, includes:

[0129] The image preprocessing unit is used to perform distortion correction and image stitching on the images acquired by the camera based on the system's calibration parameter set, generate a panoramic image of the bubble generation area, divide the panoramic image into multiple independent detection zones, and output the image set of each detection zone.

[0130] The bubble detection unit is used to receive the first reference image and identify it to obtain single bubble regions with pixel areas greater than the single bubble area threshold, and output a line chart containing the first total bubble area value in each detection zone.

[0131] The uniformity detection unit is used to receive the second reference image, perform identification, calculate the second total bubble area value of each detection zone, and output a line chart containing the second total bubble area value of each detection zone.

[0132] It should be noted that although the steps are described in a specific order above, this does not mean that they must be performed in that order. In fact, some of these steps can be executed concurrently, or even in a different order, as long as the required functionality is achieved. The number of devices and processing scale described herein are for simplification of the invention; applications, modifications, and variations of this invention will be readily apparent to those skilled in the art.

[0133] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A filter cartridge filtration performance testing method based on visual inspection, characterized in that, Includes the following steps: S1. Place the surface light source and multiple cameras on both sides of the transparent detection cylinder. Arrange the multiple cameras side by side along the preset filter length direction. Perform camera calibration to obtain a calibration parameter set. The calibration parameter set includes the internal parameters of each camera, lens distortion coefficient, and homography matrix between adjacent camera images. S2. Completely immerse the filter element to be tested into the transparent testing liquid in the transparent testing cylinder; S3. Control all cameras to synchronously acquire original images, perform distortion correction on the original images based on the calibration parameter set, and use the homography matrix to stitch the images to generate a panoramic image of the bubble generation area around the filter element to be tested, and vertically divide the panoramic image into multiple independent detection zones. S4. Apply a first gas pressure lower than the nominal bubble point pressure of the filter element to be tested to the inside of the filter element, execute step S3 to obtain a first reference image, perform image processing on the image corresponding to each detection zone in the first reference image, count the single bubble areas in each detection zone whose pixel area exceeds the single bubble area threshold, summarize the pixel areas of the single bubble areas that meet the conditions to form a first total bubble area value, if the first total bubble area value is higher than the first area threshold, it is determined that the filter element to be tested has a pore defect in the detection zone. S5. Apply a second gas pressure not lower than the nominal bubble point pressure of the filter element to be tested. Execute step S3 to obtain a second reference image. Use a bubble segmentation model trained based on deep learning to perform semantic segmentation on the images corresponding to each detection partition in the second reference image. Extract the total area of ​​pixels identified as bubbles in each detection partition as the second total bubble area value. If the second total bubble area value in any detection partition is lower than the second area threshold, it is determined that the filter element under test has a clogging defect or insufficient porosity in that detection partition.

2. The filter cartridge filtration performance testing method based on visual inspection as described in claim 1, characterized in that, The steps for performing camera calibration to obtain a calibration parameter set include: A1. Using a calibration board, acquire pose images in various poses within the field of view of each camera, and calculate the internal parameter matrix and lens distortion coefficients of each camera based on all pose images. A2. Place a transparent calibration substrate that covers the field of view of all cameras between the surface light source and the multiple cameras. The calibration substrate has multiple sets of graphic and text labels for providing matching features. Place the calibration board in front of the calibration substrate. Each camera's field of view contains only one calibration board and has an overlapping area with the adjacent camera's field of view containing the same graphic and text labels. A3. Control all cameras to synchronously acquire the original images of the calibration substrate. After distortion correction of the original images based on the internal parameter matrix of each camera and the lens distortion coefficient, the calibration images are obtained. The feature points in each calibration image are obtained by using a feature point extraction algorithm. After preliminary matching of the feature points of adjacent calibration images, the correct matching feature point pairs are selected by using a mismatch filtering algorithm. The homography matrix between each pair of adjacent calibration images is calculated based on the selected matching feature point pairs. A4. Store the internal parameters and distortion coefficients of all cameras obtained in step A1, together with the homography matrix between all adjacent cameras obtained in step A3, as a calibration parameter set.

3. The filter cartridge filtration performance testing method based on visual inspection as described in claim 2, characterized in that, The process of obtaining feature points in each calibration image includes the following steps: B1. Calculate the grayscale change function when a local window in the calibration image moves in each direction. E(u,v) : in, (x, y) To calibrate the coordinates of pixels in an image, (u, v) This indicates the minute displacement of the window in the horizontal and vertical directions. W(i, j) For window functions, I(i, j) Indicates at point (i, j) grayscale value at that location w The window radius is 3 to 7 pixels. B2. Construct the Harris matrix based on the grayscale change function. M : in, I x (i, j) and I y (i, j) At points respectively (i, j) The gradient values ​​at the location and along the x and y directions; B3. According to the matrix M Calculate the corner response function R , ,in det(M) For matrix M The determinant, trace(M) For matrix M traces, k Let be an empirical constant, R Points whose values ​​exceed a set threshold are identified as feature points.

4. The filter cartridge filtration performance testing method based on visual inspection as described in claim 2, characterized in that, The homography matrix between two adjacent calibration images is achieved using a random sampling consensus algorithm, including the following steps: C1. Randomly select four pairs of matching points from the initially matched feature point pairs as the minimum sample set to calculate the initial homography matrix model parameters; C2. Calculate the error of all feature point pairs with the initial homography matrix model, count the number of inliers whose error is less than the set tolerance, repeat the above random sampling, model fitting and consistency test process, and retain the model with the most inliers as the optimal model. C3. The calculation terminates when the maximum number of iterations N is reached. ,in p To achieve the expected success rate, ω This is an estimate of the proportion of interior points. n To obtain the minimum number of samples required to fit the model, the homography matrix is ​​finally obtained by refitting with all interior points to calculate the transformation relationship between adjacent camera images.

5. The filter cartridge filtration performance testing method based on visual inspection as described in claim 1, characterized in that, Step S4 includes the following steps: S40. Apply a first gas pressure lower than the nominal bubble point pressure of the filter element to be tested, and execute step S3 to obtain the first reference image. S41. The first partition image is obtained by smoothing the images corresponding to each detection partition in the first reference image using a mean filtering algorithm. S42. The first partition image is processed using a dynamic threshold segmentation algorithm. The adaptive threshold of each pixel is calculated based on the local neighborhood. Each single bubble region is identified and segmented by a connected component analysis algorithm. S43. Calculate the pixel area of ​​each single bubble region, count the single bubble regions in each first partition image whose pixel area is greater than the single bubble area threshold, and summarize the pixel areas of the single bubble regions that meet the conditions to form the first total bubble area value. If the first total bubble area value is higher than the first area threshold, it is determined that the filter element under test has a hole defect in the corresponding inspection partition. The single bubble area threshold and the first area threshold are both preset pixel area values.

6. The filter cartridge filtration performance testing method based on visual inspection as described in claim 1, characterized in that, The method for setting the single bubble area threshold is to determine the critical defect pore size based on the quality standard of the filter element product to be tested. D 0 Calculate the theoretical value of bubble volume ,in V B This is the theoretical value of the bubble volume. c 1 This refers to a dimensionless coefficient related to factors such as the gas-liquid interface morphology. σ To detect the surface tension of the liquid, To detect the liquid density, Bubble density, g For gravitational acceleration, the corresponding theoretical projected area of ​​the bubble Based on the pixel size and optical magnification obtained from camera calibration, the theoretical projected area of ​​the bubble is converted into the corresponding pixel area as the single bubble area threshold.

7. The filter cartridge filtration performance testing method based on visual inspection as described in claim 1, characterized in that, Step S5 includes the following steps: S50. Apply a second gas pressure not lower than the nominal bubble point pressure of the filter element to be tested, and execute step S3 to obtain a second reference image. S51. Using a multi-combination filter cartridge to collect a panoramic image under the second gas pressure and processing it in step S3, the panoramic image used for training is used as a training image. According to the size of the detection partition, the panoramic image used for training is cropped into several standard-sized training images. The bubble regions in the training images are manually labeled to generate corresponding binary mask labels. S52. The U-Net convolutional neural network with encoder-decoder architecture is used as the semantic segmentation model. The training image is used as input and the corresponding binary mask label is used as the training target. The network parameters are optimized using a combination loss function and the bubble segmentation model is trained by backpropagation algorithm. S53. After preprocessing the images corresponding to each detection partition in the second reference image, the second partition image is obtained. The second partition image is then input into the trained bubble segmentation model to obtain a pixel-level probability map. S54. The pixel-level probability map is binarized and segmented to obtain the final bubble region segmentation mask. The total number of pixels identified as bubbles in the segmentation mask corresponding to each second partition image is counted to obtain the second total bubble area value of the detection partition. If the second total bubble area value in any detection partition is lower than the second area threshold, it is determined that the filter under test has a blockage defect or insufficient opening rate in the detection partition.

8. The filter cartridge filtration performance testing method based on visual inspection as described in claim 7, characterized in that, The combined loss function L is composed of the cross-entropy loss function. L ce and Dice loss function L Dice The weighted summation is calculated using the following formula: ,in α、β These are the weighting coefficients, and α+β = 1, and the Adam optimizer is used to perform the backpropagation process. Based on the partial derivatives of the combined loss function L with respect to the network prediction results and the real labels, the weight parameters of the semantic segmentation model are iteratively optimized until the semantic segmentation model converges.

9. The filter cartridge filtration performance testing method based on visual inspection as described in claim 1, characterized in that, The first gas pressure is less than 0.95 times the nominal bubble point pressure of the filter element, and the second gas pressure is set to more than 1.05 times the nominal bubble point pressure of the filter element. ,in D 1 The nominal accuracy of the filter element. θ Contact angle , σ To detect the surface tension of the liquid, η This is the hole shape correction factor.

10. A filter cartridge filtration performance testing system based on visual inspection, used to implement the method according to any one of claims 1 to 9, characterized in that, include: A transparent cylinder is used to hold a transparent test liquid and accommodate the filter element to be tested. A backlight is located on one side of the transparent cylinder to provide uniform backlight illumination to the surface of the filter element. The air pressure control unit is connected to the filter element under test and is used to introduce gas into the filter element under test and precisely control the first gas pressure and the second gas pressure. The image acquisition module includes several industrial cameras arranged side by side; An image processing module, connected to the image acquisition module, includes: The image preprocessing unit is used to perform distortion correction and image stitching on the images acquired by the camera based on the system's calibration parameter set, generate a panoramic image of the bubble generation area, divide the panoramic image into multiple independent detection zones, and output the image set of each detection zone. The bubble detection unit is used to receive the first reference image and identify it to obtain single bubble regions with pixel areas greater than the single bubble area threshold, and output a line chart containing the first total bubble area value in each detection zone. The uniformity detection unit is used to receive the second reference image, perform identification, calculate the second total bubble area value of each detection zone, and output a line chart containing the second total bubble area value of each detection zone.

Images