Sand filter working condition evaluation method based on Yolov8-Seg

Abstract

The invention discloses a Yolov8-Seg-based sand filter working condition evaluation method, relates to the technical field of industrial sewage treatment, and solves the technical problems that an existing sand filter working condition evaluation model is relatively large, the operation speed is low and the recognition efficiency is relatively low. Comprising the following steps: 1, acquiring an inspection image; 2, preprocessing the inspection image to obtain an original image; 3, reconstructing the original image based on the hexagonal pyramid model to obtain an input image; inputting the input image into the image recognition model to obtain a recognition result image; wherein the image recognition model is constructed based on a Yov8-seg model improved by a GSA module; 4, evaluating the working condition of the sand filter according to the identification result map; corresponding image defects can be recognized more easily after the inspection images collected under different illumination and different environment conditions are correspondingly processed, and the recognition precision of the network is improved; and thus, the accuracy of evaluating the working state of the sand filter is improved.

Description

Technical Field

[0001] The present application belongs to the field of industrial wastewater treatment technology, and relates to industrial wastewater treatment technology, and specifically is a sand filter condition evaluation method based on Yolov8-Seg. Background Art

[0002] At present, the domestic tap water treatment processes mainly include coagulation, sedimentation, filtration and disinfection. Among them, the sand filtration process is the key process link in the water production process. The suspended impurities in the water pass through the sand filter and adhere to the surface of the filter material particles and are retained. At the same time, the organic matter and bacteria in the water are removed as the turbidity of the water decreases. The operating process parameters of the sand filter play a decisive role in whether the water quality meets the standards. It is extremely important to evaluate the working conditions of the sand filter.

[0003] The working conditions of existing sand filters are mostly evaluated through image processing technology and training models to process the images collected by the sand filter. These models often have problems such as large model size, slow calculation speed, and low recognition efficiency, and are not suitable for embedded platforms with general processing capabilities. In addition, since sand filters are usually located outdoors, lighting conditions may change with weather and time. Strong lighting, shadows or reflections, as well as the presence of suspended matter, bubbles, sediments and other interferences in the sand filter will interfere with the camera's visual recognition, reducing the accuracy of the sand filter working status assessment. Therefore, a sand filter working condition evaluation method based on Yolov8-Seg is needed to improve the accuracy of working status assessment. Summary of the invention

[0004] The present application aims to solve at least one of the technical problems existing in the prior art; to this end, the present application proposes a sand filter working condition evaluation method based on Yolov8-Seg, which is used to solve the technical problems that the existing sand filter working condition evaluation model is large, the calculation speed is slow, and the recognition efficiency is low. The present application solves the above problems by introducing a global self-attention mechanism module into the Yolov8-seg model to process the inspection images.

[0005] To achieve the above objectives, the first aspect of the present application provides a sand filter condition evaluation method based on Yolov8-Seg, comprising the following steps:

[0006] Step 1: Obtain inspection images of the sand filter tank;

[0007] Step 2: preprocessing the inspection image to obtain an original image; wherein the preprocessing includes image denoising and image cropping;

[0008] Step 3: reconstruct the original image based on the hexagonal cone model to obtain an input image; input the input image into the constructed image recognition model to obtain a recognition result graph; wherein the image recognition model is constructed based on the Yolov8-seg model improved by the GSA module;

[0009] Step 4: Evaluate the working condition of the sand filter according to the identification result diagram.

[0010] The present application obtains inspection pictures, pre-processes the inspection pictures to obtain original images, then uses a hexagonal cone model to reconstruct the original images to obtain input images, and inputs the input images into an image recognition model to obtain a recognition result image, so that the inspection images collected under different lighting and different environmental conditions can be more easily identified with corresponding image defects after corresponding processing, thereby improving the recognition accuracy of the network; at the same time, reducing the influence of interferences such as suspended matter, bubbles, and sediments in the sand filter tank; and improving the accuracy of the assessment of the working status of the sand filter tank.

[0011] Preferably, reconstructing the original image based on the hexagonal pyramid model to obtain the original image comprises the following steps:

[0012] S2.1: Obtain the original image, convert the original image into the HSV model, and obtain the hue component H(i,j), the saturation component S(i,j) and the lightness component V(i,j);

[0013] S2.2: Perform inverse processing on the hue component H(i,j) to obtain the inverse hue component Calculated by the following formula:

[0014]

[0015] S2.3: For the reverse hue component The coefficient matrix H'(i,j) is obtained by normalization and calculated by the following formula:

[0016]

[0017] Where c is the inverse hue component The minimum value among them, g is the inverse hue component The maximum value in ;

[0018] S2.4: Add the saturation component S(i,j) to the brightness component V(i,j) and multiply it by the coefficient matrix H′(i,j) to obtain the output value F(i,j) calculated by the following formula:

[0019] F(i,j)=(2×S(i,j) + V(i,j))×H'(i,j)

[0020] S2.5: Reconstruct the grayscale of the output value F(i,j) to obtain the original image, where (i,j) is the coordinate of the corresponding pixel.

[0021] The present application uses a hexcone model (HM) model to reconstruct the sand filter tank image by recombining the hue component (Hue), saturation component (Saturation), and brightness component (Value), so as to facilitate processing of images with different brightness and ripple jitter speeds.

[0022] Preferably, the image recognition model is constructed based on the Yolov8-seg model improved by the GSA module, comprising:

[0023] S3.1: Get image samples;

[0024] S3.2: reconstructing the image sample through a hexagonal pyramid model to obtain a standard input image;

[0025] S3.3: Introduce the GSA module into the Yolov8-seg model, and set the loss function of the Yolov8-seg model to FocalLoss to obtain an improved Yolov8-seg model; where the GSA module is a global self-attention mechanism module;

[0026] S3.4: Setting learning parameters based on the stochastic gradient descent method, training the improved Yolov8-seg model through the standard input image to obtain an image recognition model; wherein the learning parameters include a learning rate, a momentum parameter, and a weight decay.

[0027] Preferably, the introduction of the GSA module into the Yolov8-seg model comprises:

[0028] After introducing the GSA module into the SPPF module in the Yolov8-seg model; the features output by the SPPF module Figure 1 Input GSA module to get features Figure 2 ; Characteristics of GSA module Figure 2 As a feature of the original SPPF module Figure 1 Input to the PANet network in the Yolov8-seg model.

[0029] Preferably, the feature output by the SPPF module Figure 1 Input GSA module to get features Figure 2 ,include:

[0030] S3.3.1: Obtaining features Figure 1 ;

[0031] S3.3.2: Define the Q matrix, K matrix, and V matrix;

[0032] S3.3.3: Use Q matrix, K matrix and V matrix to identify features Figure 1 The content attention is calculated by formula: F c =Q(ρ(K T )V) Calculate the content attention F c ; Among them, K T is the transpose of K, and ρ means applying softmax normalization to each row;

[0033] S3.3.4: Perform position attention calculation on feature image 1; the position attention calculation includes column-only position attention calculation and row-only attention calculation; wherein the column-only position attention calculation is performed by formula:

[0034]

[0035] Calculate the column position attention Among them, q ab Indicates query in pixels; The matrix representing the learnable relative position embedding corresponding to the L spatial offsets along the columns is given by Calculated; is a matrix consisting of L-length adjacent values ​​on the same column of pixel point (a, b), through the formula Calculated;

[0036] The row-only position attention is calculated as:

[0037]

[0038] Calculate the row position attention in, The matrix representing the learnable relative position embedding corresponding to the L spatial offsets along the columns is given by Calculated; is a matrix consisting of L-length adjacent values ​​on the same column of pixel point (a, b), through the formula Calculated; among them,

[0039] S3.3.5: Superimpose content attention and position attention to obtain features Figure 2 .

[0040] This application introduces the self-attention mechanism (SA) into the Yolov8-seg model, so that the attention weight can be adaptively allocated according to the importance of the input data; by learning the intrinsic relationship of the data, the global self-attention mechanism can adjust the allocation of attention at different positions or channels, thereby improving the model's attention to important features. GSA is introduced at the front end of the Yolov8-seg model to implement preprocessing of the sand filter pool scene. By superimposing the results of the content attention layer and the position attention layer as the final output of the module, different aggregation strategies, transformations, and enhancement functions are used to aggregate the same features from all positions to enhance the original features; to adapt to the reasoning of the backend.

[0041] Preferably, evaluating the working condition of the sand filter according to the identification result graph includes:

[0042] Get the area ratio of several recognition boxes in the recognition result image;

[0043] Normalizing the proportions of the several regions;

[0044] The normalized values ​​are combined according to the set ratio to obtain a comprehensive score;

[0045] Determine whether the comprehensive score is greater than the set threshold;

[0046] If yes, the working condition of the sand filter tank is evaluated as unqualified; if no, the working condition of the sand filter tank is evaluated as qualified; wherein the set threshold is obtained by experience.

[0047] Compared with the prior art, the beneficial effects of this application are:

[0048] 1. The present application obtains inspection pictures, pre-processes the inspection pictures to obtain original images, then uses a hexagonal cone model to reconstruct the original images to obtain input images, and inputs the input images into an image recognition model to obtain a recognition result image, so that the inspection images collected under different lighting and different environmental conditions can be more easily identified with corresponding image defects after corresponding processing, thereby improving the recognition accuracy of the network; at the same time, reducing the influence of interferences such as suspended matter, bubbles, and sediments in the sand filter tank; and improving the accuracy of the assessment of the working status of the sand filter tank.

[0049] 2. The present application uses a hexagonal pyramid model to reconstruct the sand filter tank image by recombining the hue component, saturation component, and brightness component, thereby facilitating the processing of images with different brightness and ripple jitter speeds. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 A diagram of the method steps of this application;

[0052] Figure 2 This is the grayscale reconstruction image in this application

[0053] Figure 3 This is the internal framework diagram of the GSA module in this application;

[0054] Figure 4 The overall framework model of adding the GSA module to YOLOv8 in this application. DETAILED DESCRIPTION

[0055] The technical solution of the present application will be described clearly and completely in conjunction with the embodiments below. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0056] See also Figure 1 The first embodiment of the present application provides a sand filter condition evaluation method based on Yolov8-Seg, comprising the following steps:

[0057] Step 1: Obtain inspection images of the sand filter tank;

[0058] Step 2: preprocess the inspection image to obtain the original image; wherein the preprocessing includes image denoising and image cropping;

[0059] Step 3: reconstruct the original image based on the hexagonal cone model to obtain an input image; input the input image into the constructed image recognition model to obtain a recognition result graph; wherein the image recognition model is constructed based on the Yolov8-seg model improved by the GSA module;

[0060] Step 4: Evaluate the working condition of the sand filter according to the identification result diagram.

[0061] The present application obtains inspection pictures, pre-processes the inspection pictures to obtain original images, then uses a hexagonal cone model to reconstruct the original images to obtain input images, and inputs the input images into an image recognition model to obtain a recognition result image, so that the inspection images collected under different lighting and different environmental conditions can be more easily identified with corresponding image defects after corresponding processing, thereby improving the recognition accuracy of the network; at the same time, reducing the influence of interferences such as suspended matter, bubbles, and sediments in the sand filter tank; and improving the accuracy of the assessment of the working status of the sand filter tank.

[0062] Specifically, the preprocessing of inspection images includes: image rotation, image cropping and image enhancement; image rotation includes mirror rotation and left-right rotation; image enhancement includes sharpening, edge detection and pixel inversion, etc.; during training, image rotation, image cropping and image enhancement are used in random combination to achieve richer and more diverse image enhancement effects.

[0063] See also Figure 2 , reconstructing the original image based on the hexagonal cone model to obtain the original image, comprising the following steps:

[0064] S2.1: Obtain the original image, convert the original image into the HSV model, and obtain the hue component H(i,j), the saturation component S(i,j) and the lightness component V(i,j);

[0065] S2.2: Perform inverse processing on the hue component H(i,j) to obtain the inverse hue component Calculated by the following formula:

[0066]

[0067] S2.3: For the reverse hue component The coefficient matrix H'(i,j) is obtained by normalization and calculated by the following formula:

[0068]

[0069] Where c is the inverse hue component The minimum value among them, g is the inverse hue component The maximum value in ;

[0070] S2.4: Add the saturation component S(i,j) to the brightness component V(i,j) and multiply it by the coefficient matrix H′(i,j) to obtain the output value F(i,j) calculated by the following formula:

[0071] F(i,j)=(2×S(i,j) + V(i,j))×H'(i,j)

[0072] S2.5: Reconstruct the grayscale of the output value F(i,j) to obtain the original image, where (i,j) is the coordinate of the corresponding pixel.

[0073] The present application uses a hexcone model (HM) model to reconstruct the sand filter tank image by recombining the hue component (Hue), saturation component (Saturation), and brightness component (Value), so as to facilitate processing of images with different brightness and ripple jitter speeds.

[0074] The image recognition model is constructed based on the Yolov8-seg model improved by the GSA module, including:

[0075] S3.1: Get image samples;

[0076] S3.2: reconstructing the image sample through a hexagonal pyramid model to obtain a standard input image;

[0077] S3.3: Introduce the GSA module into the Yolov8-seg model, and set the loss function of the Yolov8-seg model to FocalLoss to obtain an improved Yolov8-seg model; where the GSA module is a global self-attention mechanism module;

[0078] In the sand filter scene detection, the number of samples in different categories is often unbalanced. Here, the FocalLoss loss function is used to calculate the classification loss. By reducing the weight of easy-to-classify samples, the model will pay more attention to difficult-to-classify samples, thereby improving the model's recognition ability for minority categories.

[0079] The specific FocalLoss loss function is:

[0080]

[0081] Among them, p is the predicted probability, α is the weight factor used to balance the importance of positive and negative samples, γ is the modulation factor used to adjust the rate at which the sample weight is reduced, y is the category of the true value, and the predicted probability is p when y=1.

[0082] S3.4: setting learning parameters based on the stochastic gradient descent method, training the improved Yolov8-seg model through the standard input image, and obtaining an image recognition model; wherein the learning parameters include a learning rate, a momentum parameter, and a weight decay;

[0083] For example, set the learning rate to 0.01, the momentum parameter to 0.9, and the weight decay to 0.0005; train the model;

[0084] It is worth noting that in the Yolov8-seg model, adaptive pooling is used to dynamically pool according to the size and features of the input data; the size of the pooling area is adaptively adjusted according to the size of the input data, so as to better capture the important features of the input data; the accuracy and robustness of the algorithm under the conditions of the joint use of local features and global features are effectively solved, so as to achieve the key indicator target identification of sand filter pools under a small number of samples and complex environments. For example, adaptive average pooling and adaptive maximum pooling can adaptively adjust the size of the pooling area according to the size of the input data, so as to better capture the important features of the input data; avoid traditional pooling operations, such as maximum pooling and average pooling, which use the same pooling in each pooling area, resulting in missing features and unclear problems.

[0085] See also Figure 3 and Figure 4 , introduce the GSA module into the Yolov8-seg model, including:

[0086] After introducing the GSA module into the SPPF module in the Yolov8-seg model; the features output by the SPPF module Figure 1 Input GSA module to get features Figure 2 ; Characteristics of GSA module Figure 2 As a feature of the original SPPF module Figure 1 Input to the PANet network in the Yolov8-seg model.

[0087] The features output by the SPPF module Figure 1 Input GSA module to get features Figure 2 ,include:

[0088] S3.3.1: Obtaining features Figure 1 ;

[0089] S3.3.2: Define the Q matrix, K matrix, and V matrix;

[0090] S3.3.3: Use Q matrix, K matrix and V matrix to identify features Figure 1 The content attention is calculated by formula: F c =Q(ρ(K T )V) Calculate the content attention F c ; Among them, K T is the transpose of K, and ρ means applying softmax normalization to each row;

[0091] S3.3.4: Perform position attention calculation on feature image 1; the position attention calculation includes column-only position attention calculation and row-only attention calculation; wherein the column-only position attention calculation is performed by formula:

[0092]

[0093] Calculate the column position attention Among them, q ab Indicates query in pixels; The matrix representing the learnable relative position embedding corresponding to the L spatial offsets along the columns is given by Calculated; is a matrix consisting of L-length adjacent values ​​on the same column of pixel point (a, b), through the formula Calculated;

[0094] The row-only position attention is calculated as:

[0095]

[0096] Calculate the row position attention in, The matrix representing the learnable relative position embedding corresponding to the L spatial offsets along the columns is given by Calculated; is a matrix consisting of L-length adjacent values ​​on the same column of pixel point (a, b), through the formula Calculated; among them,

[0097] S3.3.5: Superimpose content attention and position attention to obtain features Figure 2 .

[0098] This application introduces the self-attention mechanism (SA) into the Yolov8-seg model, so that attention weights can be adaptively allocated according to the importance of the input data; by learning the intrinsic relationship of the data, the global self-attention mechanism can adjust the allocation of attention at different positions or channels, thereby improving the model's attention to important features. GSA is introduced at the front end of the Yolov8-seg model to implement preprocessing of the sand filter pool scene, by superimposing the results of the content attention layer and the position attention layer as the final output of the module, and using different aggregation strategies, transformations, and enhancement functions to aggregate the same features from all positions to enhance the original features; to adapt to the reasoning of the backend;

[0099] Evaluating the working condition of the sand filter tank according to the recognition result map includes: obtaining area proportions of several recognition frames in the recognition result map;

[0100] Normalizing the proportions of the several regions;

[0101] The normalized values ​​are combined according to the set ratio to obtain a comprehensive score;

[0102] Determine whether the comprehensive score is greater than the set threshold;

[0103] If yes, the working condition of the sand filter tank is evaluated as unqualified; if no, the working condition of the sand filter tank is evaluated as qualified; wherein the set threshold is obtained by experience.

[0104] Specifically, in this embodiment, the backwash bubble identification frame and the white foam floating object identification frame in the sand filter tank are used as examples to calculate the comprehensive score; the area of ​​each bubble identification frame in the identification result graph is obtained, marked as PMi, and the total number of bubble identification frames is marked as n; the area of ​​each floating object identification frame in the identification result graph is obtained, marked as FMj, and the total number of floating object identification frames is marked as m;

[0105] The comprehensive score PF of the recognition result is calculated by the formula PF = α1×∑(PMi / ZM)+α2×∑(FMj / ZM), where ZM is the area of ​​the inspection image; α1 and α2 are weight coefficients set according to experience; when the number of items considered is large, the number of corresponding weight coefficients is the same as the number of items.

[0106] This application conducts a unified quantitative evaluation of the working conditions of the sand filter based on the results obtained from the sand filter target and regional segmentation, such as the number and proportion of bubbles in the sand filter backwash, the proportion of the area occupied by white foam floating objects, the proportion of cement platform moss and other objects, etc., and obtains numerical values ​​after unified normalization. The numerical values ​​are integrated as the comprehensive score of the target sand filter, and various factors that affect the working condition evaluation of the sand filter are comprehensively considered to facilitate a more accurate working condition evaluation.

[0107] It is worth noting that, considering that the target detection in the model is dynamic video detection, the upper and lower frame recognition situations can be compared to verify the recognition result graph, and then the sand filter tank working condition can be evaluated based on the recognition result graph; specifically, the frame images in the dynamic video detection are captured multiple times in succession as inspection images, and the inspection images are integrated into an inspection image group in the order of capture time; the comprehensive evaluation value of each inspection image in the image group is obtained, and the variance of the comprehensive evaluation value is calculated. When the corresponding variance is within the variance threshold, the comprehensive score of the last inspection image in the image group is taken as the comprehensive score of this working condition evaluation; otherwise, the next group of inspection images is captured; wherein the variance threshold is set by experience, and the time period for capturing the next group of inspection images should not overlap with the time period for capturing the previous image group; the capture time period is the time period between the time of the first inspection image and the time of the last inspection image in the image group.

[0108] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0109] How this application works:

[0110] The present application obtains inspection pictures, pre-processes the inspection pictures to obtain original images, then uses a hexagonal cone model to reconstruct the original images to obtain input images, and inputs the input images into an image recognition model to obtain a recognition result image, so that the inspection images collected under different lighting and different environmental conditions can be more easily identified with corresponding image defects after corresponding processing, thereby improving the recognition accuracy of the network; at the same time, reducing the influence of interferences such as suspended matter, bubbles, and sediments in the sand filter tank; and improving the accuracy of the assessment of the working status of the sand filter tank.

[0111] The above embodiments are only used to illustrate the technical method of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, a person of ordinary skill in the art should understand that the technical method of the present application may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present application.

Claims

1. A sand filter condition evaluation method based on Yolov8-Seg, characterized in that: The following steps are involved: Step 1: Obtain inspection images of the sand filter tank; Step 2: preprocessing the inspection image to obtain an original image; wherein the preprocessing includes image denoising and image cropping; Step 3: reconstruct the original image based on the hexagonal cone model to obtain an input image; input the input image into the constructed image recognition model to obtain a recognition result graph; wherein the image recognition model is constructed based on the Yolov8-seg model improved by the GSA module; Step 4: Evaluate the working condition of the sand filter according to the identification result diagram.

2. A sand filter operating condition evaluation method based on Yolov8-Seg according to claim 1, characterized in that: Reconstructing the original image based on the hexagonal cone model to obtain the original image includes the following steps: S2.1: Obtain the original image, convert the original image into the HSV model, and obtain the hue component H(i,j), the saturation component S(i,j) and the lightness component V(i,j); S2.2: Perform inverse image processing on the hue component H(i,j) to obtain the inverse hue component S2.3: For the reverse hue component Perform normalization to obtain the coefficient matrix H'(i,j); S2.4: Add the saturation component S(i,j) to the brightness component V(i,j) and multiply it by the coefficient matrix H′(i,j) to obtain the output value F(i,j); S2.5: Reconstruct the grayscale of the output value F(i,j) to obtain the original image, where (i,j) is the coordinate of the corresponding pixel.

3. A sand filter operating condition evaluation method based on Yolov8-Seg according to claim 1, characterized in that: The image recognition model is constructed based on the Yolov8-seg model improved by the GSA module, including: S3.1: Get image samples; S3.2: reconstructing the image sample through a hexagonal pyramid model to obtain a standard input image; S3.3: Introduce the GSA module into the Yolov8-seg model, and set the loss function of the Yolov8-seg model to FocalLoss to obtain an improved Yolov8-seg model; where the GSA module is a global self-attention mechanism module; S3.4: Setting learning parameters based on the stochastic gradient descent method, training the improved Yolov8-seg model through the standard input image to obtain an image recognition model; wherein the learning parameters include a learning rate, a momentum parameter, and a weight decay.

4. A sand filter operating condition evaluation method based on Yolov8-Seg according to claim 3, characterized in that: The introduction of the GSA module into the Yolov8-seg model includes: After introducing the GSA module into the SPPF module in the Yolov8-seg model; the feature map 1 output by the SPPF module is input into the GSA module to obtain the feature map 2; the feature map 2 of the GSA module is used as the feature map 1 of the original SPPF module and input into the PANet network in the Yolov8-seg model.

5. A sand filter operating condition evaluation method based on Yolov8-Seg according to claim 4, characterized in that: The step of inputting the feature graph 1 output by the SPPF module into the GSA module to obtain the feature graph 2 includes: S3.3.1: Get feature map 1; S3.3.2: Define the Q matrix, K matrix, and V matrix; S3.3.3: Use Q matrix, K matrix and V matrix to calculate the content attention of feature map 1; S3.3.4: Perform position attention calculation on feature image 1; the position attention calculation includes column-only position attention calculation and row-only position attention calculation; S3.3.5: Superimpose content attention and position attention to obtain feature map 2.

6. A sand filter operating condition evaluation method based on Yolov8-Seg according to claim 1, characterized in that: Evaluating the sand filter working condition according to the identification result diagram includes: Get the area proportions of several recognition boxes in the recognition result graph; Normalizing the proportions of the several regions; The normalized values ​​are combined according to the set ratio to obtain a comprehensive score; Determine whether the comprehensive score is greater than the set threshold; If yes, the working condition of the sand filter tank is evaluated as unqualified; if no, the working condition of the sand filter tank is evaluated as qualified.

Images