Deep learning-based sink algae recognition and removal method, system, and medium

Abstract

The application relates to the technical field of control of artificial intelligence systems and discloses a method and system for recognizing and removing algae in a water tank based on deep learning and a medium. The real-time image of a secondary sedimentation tank water tank is preprocessed to obtain an initial water tank image, the initial water tank image is segmented to obtain a regional water tank image, target detection is performed, an algae area is recognized and predicted, and predicted algae feature data is obtained for processing to obtain a predicted algae measurement index corresponding to the regional water tank image. Through threshold comparison, it is determined whether the predicted algae area contains algae. If the predicted algae area contains algae, the number and position data of the algae pixels in the predicted algae area are further obtained, the data is combined with the regional water tank image for processing to obtain a water tank cleanliness evaluation index, the cleanliness state of the water tank is obtained through threshold comparison, and a removal scheme is generated according to the cleanliness state; thereby, intelligent recognition and removal of algae in the secondary sedimentation tank water tank are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of control of artificial intelligence systems, and more particularly, to a method and system for identifying and removing algae in a water tank based on deep learning, and a medium. BACKGROUND

[0002] In the algae cleaning work of the water tank in the sewage treatment plant, the traditional manual method has many disadvantages, such as low efficiency, high labor and time cost, and the manual operation is easily affected by subjective factors, which is prone to errors in judging the algae cleaning opportunity and cleaning degree, and it is difficult to guarantee the stability and accuracy of the cleaning effect. With the vigorous development of artificial intelligence technology, computer vision and deep learning technology have brought new opportunities to solve the problem of algae cleaning in the water tank. Related technologies have been explored in some fields, such as the algae cleaning method for the sedimentation tank in patent CN202411643386.8, but the intelligent identification and removal method of algae needs to be further optimized. SUMMARY

[0003] In view of the above problems, the purpose of the present application is to provide a method and system for identifying and removing algae in a water tank based on deep learning, and a medium.

[0004] To solve the above technical problems, the technical solution of the present application is as follows:

[0005] The first aspect of the present application provides a method for identifying and removing algae in a water tank based on deep learning, comprising the following steps:

[0006] Collecting real-time images of the water tank of the secondary sedimentation tank and performing preprocessing to obtain initial water tank images;

[0007] Segmenting the initial water tank images by a preset image segmentation algorithm to obtain regional water tank images;

[0008] Detecting the regional water tank images by a preset target detection algorithm to identify and predict the algae area, and obtaining the predicted algae feature data, and processing the predicted algae feature data to obtain the predicted algae measurement index corresponding to the regional water tank images;

[0009] Comparing the predicted algae measurement index with a preset algae measurement threshold;

[0010] If the predicted algae measurement index is less than the preset algae measurement threshold, it is determined that there is no algae in the predicted algae area;

[0011] If the predicted algae measurement index is greater than or equal to the preset algae measurement threshold, it is determined that there is algae in the predicted algae area, and the number and position data of the algae pixels in the predicted algae area are obtained;

[0012] According to the moss pixel quantity and position data, the region sink image is processed to obtain a sink cleanliness evaluation index;

[0013] The sink cleanliness evaluation index is compared with a preset sink cleanliness threshold to obtain a cleaning state of the sink, and a cleaning scheme is generated according to the cleaning state.

[0014] Optionally, in the deep learning-based sink moss identification and cleaning method, the target detection algorithm is used to detect the region sink image, identify a predicted moss region, and obtain predicted moss feature data, including color feature data and texture feature data.

[0015] The target detection algorithm is used to detect the region sink image, identify a predicted moss region, and obtain predicted moss feature data, including color feature data and texture feature data.

[0016] The color feature data includes a color histogram and color moments, and the color moments include first-order color moments, second-order color moments, and third-order color moments.

[0017] The texture feature data includes an LBP histogram and a GLCM feature matrix.

[0018] The color histogram, color moments, LBP histogram, and GLCM feature matrix are processed to obtain a predicted moss measurement index corresponding to the region sink image.

[0019] Optionally, in the deep learning-based sink moss identification and cleaning method, the color histogram, color moments, LBP histogram, and GLCM feature matrix are processed to obtain a predicted moss measurement index corresponding to the region sink image, including:

[0020] The color histogram, first-order color moments, second-order color moments, and third-order color moments are combined into a joint feature vector, which is input into a preset moss identification model for processing to obtain a moss color probability value.

[0021] The LBP histogram and GLCM feature matrix are respectively compared with a preset moss LBP histogram and a preset moss GLCM feature matrix for similarity comparison to obtain a local binary similarity value and a gray-level co-occurrence similarity value.

[0022] The moss color probability value, local binary similarity value, and gray-level co-occurrence similarity value are processed to obtain a predicted moss measurement index corresponding to the region sink image.

[0023] Optionally, in the deep learning-based sink moss identification and removal method described in the present application, the processing of the moss pixel number and position data in combination with the regional sink image to obtain a sink cleanliness evaluation index comprises:

[0024] extracting a regional pixel number of the regional sink image according to the regional sink image;

[0025] comparing the moss pixel number with the regional pixel number to obtain a regional moss coverage rate corresponding to the regional sink image, and then performing mean value processing to obtain a sink moss coverage rate;

[0026] obtaining a total number of regional sink images in a preset range, and determining a number of regional sink images with moss in the preset range according to the position data;

[0027] comparing the number of regional sink images with the total number to obtain a regional moss density corresponding to the preset range, and then performing mean value processing to obtain a sink moss density;

[0028] processing the regional moss coverage rate to obtain moss distribution uniformity data;

[0029] inputting the sink moss coverage rate, the sink moss density, and the moss distribution uniformity data into a preset sink cleanliness evaluation model for processing to obtain a sink cleanliness evaluation index.

[0030] Optionally, in the deep learning-based sink moss identification and removal method described in the present application, the threshold comparison of the sink cleanliness evaluation index with a preset sink cleanliness threshold to obtain a cleaning state of the sink, and the generation of a removal scheme according to the cleaning state comprises:

[0031] threshold comparison of the sink cleanliness evaluation index with a preset sink cleanliness threshold, wherein the preset sink cleanliness threshold comprises a first preset sink cleanliness threshold and a second preset sink cleanliness threshold, and the first preset sink cleanliness threshold is smaller than the second preset sink cleanliness threshold;

[0032] if the sink cleanliness evaluation index is smaller than or equal to the first preset sink cleanliness threshold, it is determined that the cleaning state is clean, and the removal scheme is no removal;

[0033] if the sink cleanliness evaluation index is greater than the first preset sink cleanliness threshold and smaller than or equal to the second preset sink cleanliness threshold, it is determined that the cleaning state is lightly contaminated, and the removal scheme is to remove by the linkage of a cleaning device;

[0034] If the sink cleanliness evaluation index is greater than a second preset sink cleanliness threshold, it is determined that the cleaning state is heavily contaminated, and a cleaning scheme is generated as linkage cleaning equipment and manual operation cleaning.

[0035] Optionally, in the deep learning-based sink moss identification and removal method described in the present application, further comprising:

[0036] Obtain the historical synchronous predicted moss measurement index mean corresponding to the real-time image acquisition time;

[0037] Compare the predicted moss measurement index with the predicted moss measurement index mean to obtain a moss measurement deviation rate;

[0038] Threshold comparison is performed between the moss measurement deviation rate and a preset moss measurement deviation rate threshold;

[0039] If the moss measurement deviation rate is less than or equal to the preset moss measurement deviation rate threshold, it is determined that the predicted moss measurement index is normal;

[0040] If the moss measurement deviation rate is greater than the preset moss measurement deviation rate threshold, it is determined that the predicted moss measurement index is abnormal.

[0041] Optionally, in the deep learning-based sink moss identification and removal method described in the present application, if the moss measurement deviation rate is greater than the preset moss measurement deviation rate threshold, it is determined that the predicted moss measurement index is abnormal, and then further comprising:

[0042] Obtain water quality evaluation data and environmental evaluation data of the secondary sedimentation tank sink at the real-time image acquisition time;

[0043] Obtain historical water quality evaluation data and historical environmental evaluation data of the secondary sedimentation tank at the real-time image acquisition time corresponding to the historical synchronous period;

[0044] According to the water quality evaluation data and environmental evaluation data combined with the historical water quality evaluation data and historical environmental evaluation data, a moss measurement correction coefficient is obtained;

[0045] According to the moss measurement correction coefficient, the predicted moss measurement index is corrected to obtain a moss measurement optimization index.

[0046] The second aspect of the present application provides a deep learning-based sink moss identification and removal system, which comprises a memory and a processor, the memory comprising a deep learning-based sink moss identification and removal method program, and the deep learning-based sink moss identification and removal method program is executed by the processor to realize the following steps:

[0047] Collecting real-time images of the water tank of the secondary sedimentation tank, and preprocessing to obtain initial water tank images;

[0048] Segmenting the initial water tank images by a preset image segmentation algorithm to obtain regional water tank images;

[0049] Detecting targets in the regional water tank images by a preset target detection algorithm to identify predicted moss regions and obtain predicted moss feature data, and processing the predicted moss feature data to obtain a predicted moss measurement index corresponding to the regional water tank images;

[0050] Comparing the predicted moss measurement index with a preset moss measurement threshold;

[0051] If the predicted moss measurement index is less than the preset moss measurement threshold, it is determined that the predicted moss region does not contain moss;

[0052] If the predicted moss measurement index is greater than or equal to the preset moss measurement threshold, it is determined that the predicted moss region contains moss, and the number and position data of moss pixels in the predicted moss region are obtained;

[0053] Processing the number and position data of moss pixels in combination with the regional water tank images to obtain a water tank cleanliness evaluation index;

[0054] Comparing the water tank cleanliness evaluation index with a preset water tank cleanliness threshold to obtain a cleaning state of the water tank, and generating a cleaning scheme according to the cleaning state.

[0055] Optionally, in the deep learning-based water tank moss identification and removal system described in the present application, the detecting targets in the regional water tank images by a preset target detection algorithm to identify predicted moss regions and obtain predicted moss feature data, processing the predicted moss feature data to obtain a predicted moss measurement index corresponding to the regional water tank images, comprises:

[0056] Detecting targets in the regional water tank images by a preset target detection algorithm to identify predicted moss regions and obtain predicted moss feature data, including color feature data and texture feature data;

[0057] The color feature data includes a color histogram and color moments, and the color moments include first-order color moments, second-order color moments and third-order color moments;

[0058] The texture feature data includes an LBP histogram and a GLCM feature matrix;

[0059] Processing the color histogram and color moments and the LBP histogram and GLCM feature matrix to obtain a predicted moss measurement index corresponding to the regional water tank images.

[0060] The third aspect of the present application provides a computer readable storage medium, wherein a deep learning-based sink algae identification and removal method program is stored, and when the deep learning-based sink algae identification and removal method program is executed by a processor, the steps of the deep learning-based sink algae identification and removal method according to any one of the preceding aspects are implemented.

[0061] The present application obtains an initial sink image by preprocessing a real-time image of a secondary sedimentation tank sink, and obtains a regional sink image by segmentation processing, then performs target detection to identify a predicted algae area, and obtains predicted algae feature data for processing to obtain a predicted algae measurement index corresponding to the regional sink image. By threshold comparison, it is determined whether there is algae in the predicted algae area. If there is, the number and position data of algae pixels in the predicted algae area are further obtained, combined with the regional sink image for processing to obtain a sink cleanliness evaluation index. Finally, by threshold comparison, the cleaning state of the sink is obtained, and a removal scheme is generated according to the cleaning state. Thus, intelligent identification and removal of algae in the secondary sedimentation tank sink are realized.

[0062] Compared with the prior art, the beneficial effects of the technical scheme of the present application are:

[0063] The cleaning efficiency is improved: compared with manual cleaning, the system can realize real-time monitoring, identification and automatic cleaning of the sink, thereby improving the cleaning efficiency.

[0064] The cleaning accuracy is improved: through the deep learning model, the algae can be accurately identified, and the distribution of the algae can be accurately evaluated, so that more targeted cleaning is realized.

[0065] The labor cost is reduced: the manual intervention is reduced, the removal targeting is improved, and the labor cost is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 The flowchart of the deep learning-based sink algae identification and removal method provided for the embodiments of the present application is provided.

[0067] Figure 2 The flowchart of obtaining a predicted algae measurement index corresponding to a regional sink image of the deep learning-based sink algae identification and removal method provided for the embodiments of the present application is provided.

[0068] Figure 3 The flowchart of obtaining a sink cleanliness evaluation index of the deep learning-based sink algae identification and removal method provided for the embodiments of the present application is provided. DETAILED DESCRIPTION

[0069] In order to enable a more clear understanding of the above-mentioned objects, features and advantages of the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0070] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, and therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below.

[0071] Embodiment 1

[0072] As Figure 1 shown, the present embodiment discloses a deep learning-based sink moss identification and removal method, comprising the following steps:

[0073] S11, collecting real-time images of the secondary sedimentation tank sink and performing preprocessing to obtain initial sink images;

[0074] S12, performing segmentation processing on the initial sink images by a pre-set image segmentation algorithm to obtain regional sink images;

[0075] S13, performing target detection on the regional sink images by a pre-set target detection algorithm, identifying a predicted moss region, and obtaining predicted moss feature data, processing according to the predicted moss feature data to obtain a predicted moss measurement index corresponding to the regional sink images;

[0076] S14, comparing the predicted moss measurement index with a pre-set moss measurement threshold value;

[0077] S15, if the predicted moss measurement index is less than the pre-set moss measurement threshold value, it is determined that the predicted moss region does not exist moss;

[0078] S16, if the predicted moss measurement index is greater than or equal to the pre-set moss measurement threshold value, it is determined that the predicted moss region exists moss, and the number and position data of moss pixels in the predicted moss region are obtained;

[0079] S17, processing according to the number and position data of moss pixels in combination with the regional sink images to obtain a sink cleanliness evaluation index;

[0080] S18, comparing the sink cleanliness evaluation index with a pre-set sink cleanliness threshold value to obtain a cleaning state of the sink, and generating a removal scheme according to the cleaning state.

[0081] It should be noted that, in order to accurately identify the moss growth condition of the water tank of the secondary sedimentation tank and determine the cleanliness of the water tank, first, a camera deployed in the area of the secondary sedimentation tank takes pictures at intervals of a preset time as input data, the collected pictures are preprocessed including denoising, contrast enhancement and color space conversion, then a preset image segmentation algorithm is used for segmentation processing to obtain a regional water tank image, in this embodiment, U-Net is used as a basic model, and the model is optimized and improved through attention mechanism and multi-scale feature fusion, so that the model can focus on the key areas in the image, improve the segmentation accuracy, and through the jump connection, the features of different scales are fused, so that the model can capture multi-scale information and improve the segmentation details; after the segmentation is completed, a target detection algorithm is used for target detection to identify and predict the moss area, that is, the image of the area where the moss may exist, and the predicted moss feature data is obtained for processing to obtain the predicted moss measurement index corresponding to the regional water tank image, in this embodiment, a target detection algorithm based on CNN is used for target detection; then, whether the predicted moss area exists moss is determined through threshold comparison, in this embodiment, the preset moss measurement threshold is set to (0, 0.85), [0.85, 1], corresponding to no moss and moss respectively, if it is judged that there is moss, the number and position data of the moss pixels in the predicted moss area are further obtained and processed in parallel to obtain the water tank cleanliness evaluation index, finally, the threshold is compared with the preset water tank cleanliness threshold to obtain the cleaning state of the water tank, and a cleaning scheme is generated according to the cleaning state.

[0082] Embodiment 2

[0083] As shown in Figure 2 , the embodiment discloses a flowchart for obtaining a predicted moss measurement index corresponding to a regional water tank image based on a deep learning-based water tank moss identification and removal method. According to the embodiment of the present application, the target detection algorithm is used for target detection on the regional water tank image, the predicted moss area is identified, the predicted moss feature data is obtained, the predicted moss feature data is processed, and the predicted moss measurement index corresponding to the regional water tank image is obtained, including:

[0084] S21, a target detection algorithm is used for target detection on the regional water tank image, the predicted moss area is identified, and the predicted moss feature data is obtained, including color feature data and texture feature data;

[0085] S22, the color feature data includes a color histogram and a color moment, and the color moment includes a first-order color moment, a second-order color moment and a third-order color moment;

[0086] S23, the texture feature data includes an LBP histogram and a GLCM feature matrix;

[0087] S24, processing according to the color histogram and color moment and the LBP histogram and GLCM feature matrix, obtaining the predicted moss measurement index corresponding to the region gutter image.

[0088] It should be noted that, in order to accurately obtain the predicted moss measurement index of the predicted moss region, in order to judge whether the region exists moss, first obtain the color feature data including color histogram and color moment, wherein the color moment includes first-order color moment, second-order color moment and third-order color moment, the first-order color moment refers to the mean, the second-order color moment refers to the variance, and the third-order color moment refers to the skewness; the texture feature data including LBP histogram and GLCM feature matrix, wherein the LBP histogram refers to the histogram of local binary feature map, and the GLCM feature matrix refers to the gray level co-occurrence matrix, and further analysis and processing are performed to obtain the predicted moss measurement index corresponding to the region gutter image.

[0089] According to the examples of the present application, the processing according to the color histogram and color moment and the LBP histogram and GLCM feature matrix, obtaining the predicted moss measurement index corresponding to the region gutter image, comprises:

[0090] The color histogram, the first-order color moment, the second-order color moment and the third-order color moment are combined into a joint feature vector, and input into a preset moss recognition model for processing to obtain a moss color probability value;

[0091] The LBP histogram and GLCM feature matrix are compared with the preset moss LBP histogram and the preset moss GLCM feature matrix respectively to obtain a local binary similarity value and a gray level co-occurrence similarity value;

[0092] According to the moss color probability value, the local binary similarity value and the gray level co-occurrence similarity value, processing is performed to obtain the predicted moss measurement index corresponding to the region gutter image.

[0093] It should be noted that, in order to improve the accuracy of judging whether there is moss based on color and texture features, first combine the obtained color histogram and color moment into a joint feature vector, for example, the color histogram is a one-dimensional vector H=(h1, h2, h3, …, h512) with a length of 512, and the first-order color moment of the red channel R, the green channel G and the yellow channel B is respectively μR, μG, μB, the second-order color moment of the red channel R, the green channel G and the yellow channel B is respectively σR, σG, σB, and the third-order color moment of the red channel R, the green channel G and the yellow channel B is respectively sR, sG, sB. 512 R G B R G B ​​​​​​​Color moments are typically one-dimensional vectors, and the color moment vector is represented as... The joint eigenvector is F = (H, M);

[0094] Right now The data is then input into a preset moss recognition model for processing to obtain moss color probability values. This preset moss recognition model is trained using joint feature vectors and corresponding moss color probability values ​​from a large number of historical samples. The LBP histogram and GLCM feature matrix are then compared with the preset moss LBP histogram and GLCM feature matrix to obtain local binary similarity and gray-level co-occurrence similarity values. The preset moss LBP histogram and GLCM feature matrix are obtained through analysis of a large number of known moss images. Finally, the obtained moss color probability values, local binary similarity values, and gray-level co-occurrence similarity values ​​are processed to obtain the predicted moss measurement index corresponding to the regional water tank image.

[0095] The formula for calculating the predicted moss measurement index is as follows:

[0096] y q =α1c g +β1(b l +g l );

[0097] Among them, y q To predict the moss measurement index, c g b l g l These are the probability value of moss color, the local binary similarity value, and the gray-level co-occurrence similarity value, respectively. α1 and β1 are preset feature coefficients (the feature coefficients are obtained by querying the preset water tank moss recognition and removal control platform). All the water tank images of the multiple regions obtained by segmentation are used to complete target detection.

[0098] Example 3

[0099] like Figure 2 As shown, this embodiment discloses a flowchart of a deep learning-based method for identifying and removing algae in a sink to obtain a sink cleanliness evaluation index. This embodiment discloses a flowchart of a deep learning-based method for identifying and removing algae in a sink to obtain a sink cleanliness evaluation index. According to this embodiment, the step of processing the algae pixel quantity and location data in conjunction with the regional sink image to obtain the sink cleanliness evaluation index includes:

[0100] S31. Extract the number of pixels in the region of the water tank image based on the region water tank image;

[0101] S32, compare the number of moss pixels with the number of region pixels to obtain a region moss coverage corresponding to the region gutter image, and then perform mean value processing to obtain a gutter moss coverage;

[0102] S33, obtain a total number of region gutter images in a preset range, and determine the number of region gutter images with moss in the preset range according to the position data;

[0103] S34, compare the number of region gutter images with the total number to obtain a region moss density corresponding to the preset range, and then perform mean value processing to obtain a gutter moss density;

[0104] S35, process according to the region moss coverage to obtain moss distribution uniformity data;

[0105] S36, input the gutter moss coverage, gutter moss density and moss distribution uniformity data into a preset gutter cleanliness evaluation model for processing to obtain a gutter cleanliness evaluation index.

[0106] It should be noted that, in order to determine the cleanliness of the gutter, the region pixel number of the region gutter image with moss is extracted, the number of moss pixels obtained is compared with the number of region pixels to obtain a region moss coverage, for example, the number of moss pixels is N m , the number of region pixels is N t , and N m / N t is the region moss coverage, and the mean value of the obtained multiple region moss coverages is obtained to obtain a gutter moss coverage; the entire initial gutter image is divided into N preset ranges, the preset range includes N1 region gutter images, the number of region gutter images with moss in the preset range is N2 according to the obtained position data, the region moss density corresponding to the preset range is N2 / N1, and then the mean value is obtained to obtain a gutter moss density; the initial gutter image is divided into multiple sub-regions, for example, the image is uniformly divided into axb small squares, and the moss coverage in each sub-region is calculated respectively, in this embodiment, the standard deviation of the sub-region moss coverage is calculated to obtain moss distribution uniformity data, which measures the uniformity of moss distribution, the smaller the standard deviation, the more uniform the moss distribution, and the larger the standard deviation, the more uneven the moss distribution; the obtained gutter moss coverage, gutter moss density and moss distribution uniformity data are input into a preset gutter cleanliness evaluation model for processing to obtain a gutter cleanliness evaluation index, wherein the preset gutter cleanliness evaluation model is trained by obtaining a large number of historical sample gutter moss coverages, gutter moss densities and moss distribution uniformity data and corresponding gutter cleanliness evaluation indexes.

[0107] According to the example of the present application, the sink cleanliness evaluation index is compared with the preset sink cleanliness threshold value to obtain the cleaning state of the sink, and a cleaning scheme is generated according to the cleaning state, including:

[0108] The sink cleanliness evaluation index is compared with the preset sink cleanliness threshold value, the preset sink cleanliness threshold value includes a first preset sink cleanliness threshold value and a second preset sink cleanliness threshold value, and the first preset sink cleanliness threshold value is less than the second preset sink cleanliness threshold value;

[0109] If the sink cleanliness evaluation index is less than or equal to the first preset sink cleanliness threshold value, it is determined that the cleaning state is clean, and the cleaning scheme is not cleaned;

[0110] If the sink cleanliness evaluation index is greater than the first preset sink cleanliness threshold value and less than or equal to the second preset sink cleanliness threshold value, it is determined that the cleaning state is lightly contaminated, and the cleaning scheme is to clean the sink by the linkage cleaning equipment;

[0111] If the sink cleanliness evaluation index is greater than the second preset sink cleanliness threshold value, it is determined that the cleaning state is heavily contaminated, and the cleaning scheme is to clean the sink by the linkage cleaning equipment and manual operation.

[0112] It should be noted that the obtained sink cleanliness evaluation index is compared with the preset sink cleanliness threshold value, and in this embodiment, the preset sink cleanliness threshold value is set to (0, 0.4], (0.4, 0.75], (0.75, 1], which respectively corresponds to clean, lightly contaminated, and heavily contaminated. For example, if the obtained sink cleanliness evaluation index is 0.3, which is less than the first preset sink cleanliness threshold value, it means that the amount of moss is small, and it is determined that the cleaning state is clean, and the cleaning scheme is not cleaned. If the obtained sink cleanliness evaluation index is 0.55, which is greater than the first preset sink cleanliness threshold value and less than the second preset sink cleanliness threshold value, it means that the amount of moss is moderate, and it is determined that the cleaning state is lightly contaminated, and the cleaning scheme is to clean the sink by the linkage cleaning equipment. If the obtained sink cleanliness evaluation index is 0.85, which is greater than the second preset sink cleanliness threshold value, it means that the amount of moss is too much, and it is determined that the cleaning state is heavily contaminated. Only the equipment cleaning may not be timely or complete, so the cleaning scheme is to clean the sink by the linkage cleaning equipment and manual operation.

[0113] According to the example of the present application, it further includes:

[0114] Obtaining the average of the predicted moss amount index of the historical same period corresponding to the real-time image acquisition time;

[0115] Comparing the predicted moss amount index with the average of the predicted moss amount index to obtain the moss amount index deviation rate;

[0116] threshold comparison is performed between the moss growth deviation rate and a preset moss growth deviation rate threshold value;

[0117] If the moss growth deviation rate is less than or equal to the preset moss growth deviation rate threshold value, it is determined that the predicted moss growth index is normal.

[0118] If the moss growth deviation rate is greater than the preset moss growth deviation rate threshold value, it is determined that the predicted moss growth index is abnormal.

[0119] It should be noted that the growth of moss in the secondary sedimentation tank has certain rules, and the growth is periodic, which is usually related to seasons, water quality and light. Therefore, after obtaining the predicted moss growth index, it should be further judged whether it is accurate and reasonable. First, the mean value of the predicted moss growth index corresponding to the historical same period of the real-time image acquisition time is obtained, and the predicted moss growth index is compared to obtain the moss growth deviation rate. The moss growth deviation rate refers to the ratio of the absolute value of the difference between the predicted moss growth index and the mean value of the predicted moss growth index to the mean value of the predicted moss growth index. Then, threshold comparison is performed between the moss growth deviation rate and a preset moss growth deviation rate threshold value. The preset moss growth deviation rate threshold value is obtained by querying the preset tank moss identification and removal control platform. The threshold comparison determines whether the obtained predicted moss growth index is normal or abnormal.

[0120] According to the example of the present application, if the moss growth deviation rate is greater than the preset moss growth deviation rate threshold value, it is determined that the predicted moss growth index is abnormal. Then, the following steps are further included:

[0121] Obtain water quality evaluation data and environmental evaluation data of the secondary sedimentation tank at the real-time image acquisition time;

[0122] Obtain historical water quality evaluation data and historical environmental evaluation data of the secondary sedimentation tank corresponding to the historical same period of the real-time image acquisition time;

[0123] According to the water quality evaluation data and the environmental evaluation data, the historical water quality evaluation data and the historical environmental evaluation data are processed to obtain a moss growth correction coefficient;

[0124] According to the moss growth correction coefficient, the predicted moss growth index is corrected to obtain a moss growth optimization index.

[0125] It should be noted that if the obtained predicted moss growth index is determined to be abnormal, the moss growth correction coefficient is obtained by comparing and processing the changes of the secondary sedimentation tank water quality and the environment. The obtained predicted moss growth index is corrected to obtain a moss growth optimization index. Then, threshold comparison is performed again. If it is still abnormal, an early warning response is output.

[0126] The moss growth optimization index calculation formula is:

[0127] y xq =ζ×c f ×y q ;

[0128] Among them, y xq To optimize the moss measurement index, c f y q These are the moss measurement correction coefficient and the predicted moss measurement index, respectively, and ζ is the preset characteristic coefficient (the characteristic coefficient is obtained by querying the preset water tank moss identification and removal control platform).

[0129] It is worth mentioning that, according to an example of the present invention, the step of processing the water quality assessment data and environmental assessment data in combination with the historical water quality assessment data and historical environmental assessment data to obtain the algae measurement correction coefficient includes:

[0130] The water quality assessment data includes real-time nitrogen content and real-time phosphorus content;

[0131] The environmental assessment data includes real-time light intensity, real-time light duration, and real-time water temperature.

[0132] The historical water quality assessment data includes the historical average nitrogen content and the historical average phosphorus content;

[0133] The historical environmental assessment data includes the historical average light intensity, historical average light duration, and historical average water temperature.

[0134] The algae measurement correction coefficient is obtained by processing the real-time nitrogen and phosphorus content in combination with the historical average nitrogen and phosphorus content, as well as the real-time light intensity, real-time light duration, and real-time water temperature in combination with the historical average light intensity, historical average light duration, and historical average water temperature.

[0135] It should be noted that the formula for calculating the moss measurement correction factor is as follows:

[0136]

[0137] Among them, c f n is the correction factor for the measurement of moss. r p r n h p h These represent real-time nitrogen content, real-time phosphorus content, historical average nitrogen content, and historical average phosphorus content, respectively, in g. r t r g tr g h t h g thReal-time light intensity, real-time light duration, real-time water temperature, historical light intensity average, historical light duration average, and historical water temperature average, respectively, ε and χ are preset characteristic coefficients (characteristic coefficients are obtained by querying a preset water tank moss identification and removal control platform).

[0138] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed components can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0139] The units described above as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units; they can be located in one place or distributed on multiple network units; part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0140] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can be a unit alone, or two or more units can be integrated into one unit; the integrated unit can be realized in the form of hardware or in the form of hardware plus software functional unit.

[0141] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware, and the foregoing program can be stored in a computer readable storage medium, and the program executes the steps including the above-mentioned method embodiments when executed; and the foregoing storage medium includes: mobile storage device, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and various storage program codes.

[0142] Alternatively, the above-mentioned integrated unit of the present application, if realized in the form of a software function module and sold or used as an independent product, can also be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: mobile storage devices, ROM, RAM, magnetic disks or optical disks, and various media that can store program codes.

Claims

1. A method for sink algae identification and removal based on deep learning, characterized in that, The method comprises the following steps: Collecting real-time images of the secondary sedimentation tank water tank and performing preprocessing to obtain initial water tank images; Segmenting the initial water tank images through a preset image segmentation algorithm to obtain regional water tank images; Detecting targets in the regional water tank images through a preset target detection algorithm, identifying a predicted moss area, and obtaining predicted moss feature data, and processing the predicted moss feature data to obtain a predicted moss measurement index corresponding to the regional water tank images; Comparing the predicted moss measurement index with a preset moss measurement threshold value; If the predicted moss measurement index is less than the preset moss measurement threshold value, it is determined that the predicted moss area does not contain moss; If the predicted moss measurement index is greater than or equal to the preset moss measurement threshold value, it is determined that the predicted moss area contains moss, and the number and position data of moss pixels in the predicted moss area are obtained; Processing the number and position data of moss pixels in combination with the regional water tank images to obtain a water tank cleanliness evaluation index; Comparing the water tank cleanliness evaluation index with a preset water tank cleanliness threshold value to obtain a cleaning state of the water tank, and generating a cleaning scheme according to the cleaning state; The processing of the number and position data of moss pixels in combination with the regional water tank images to obtain a water tank cleanliness evaluation index comprises: Extracting the number of regional pixels of the regional water tank images from the regional water tank images; Comparing the number of moss pixels with the number of regional pixels to obtain a regional moss coverage rate corresponding to the regional water tank images, and then performing mean value processing to obtain a water tank moss coverage rate; Obtaining the total number of regional water tank images within a preset range, and determining the number of regional water tank images containing moss within the preset range according to the position data; Comparing the number of regional water tank images with the total number to obtain a regional moss density corresponding to the preset range, and then performing mean value processing to obtain a water tank moss density; Processing the regional moss coverage rate to obtain moss distribution uniformity data; Inputting the water tank moss coverage rate, water tank moss density, and moss distribution uniformity data into a preset water tank cleanliness evaluation model for processing to obtain a water tank cleanliness evaluation index; The preset water tank cleanliness evaluation model is obtained by training the water tank moss coverage rate, water tank moss density, and moss distribution uniformity data of historical samples and the corresponding water tank cleanliness evaluation index.

2. The sink algae identification and removal method based on deep learning according to claim 1, characterized in that, The processing of the number and position data of moss pixels in combination with the regional water tank images to obtain a water tank cleanliness evaluation index comprises: Detecting targets in the regional water tank images through a preset target detection algorithm, identifying a predicted moss area, and obtaining predicted moss feature data, including color feature data and texture feature data; The color feature data includes a color histogram and a color moment, and the color moment includes a first-order color moment, a second-order color moment, and a third-order color moment; The texture feature data includes an LBP histogram and a GLCM feature matrix; According to the color histogram and color moment and the LBP histogram and GLCM feature matrix, processing is performed to obtain a predicted moss measurement index corresponding to the area sink image. 3.The sink algae identification and removal method based on deep learning according to claim 2, characterized in that, The processing according to the color histogram and color moment and the LBP histogram and GLCM feature matrix to obtain a predicted moss measurement index corresponding to the area sink image comprises: combining the color histogram, first-order color moment, second-order color moment and third-order color moment into a joint feature vector, and inputting the joint feature vector into a preset moss identification model for processing to obtain a moss color probability value; comparing the LBP histogram and GLCM feature matrix with a preset moss LBP histogram and a preset moss GLCM feature matrix respectively to obtain a local binary similarity value and a gray level co-occurrence similarity value; According to the moss color probability value, local binary similarity value and gray level co-occurrence similarity value, processing is performed to obtain a predicted moss measurement index corresponding to the area sink image. 4.The sink algae identification and removal method based on deep learning according to claim 3, characterized in that, The threshold comparison of the sink cleanliness evaluation index with the preset sink cleanliness threshold value to obtain the cleaning state of the sink, and generating a cleaning scheme according to the cleaning state, comprises: The threshold comparison of the sink cleanliness evaluation index with the preset sink cleanliness threshold value, wherein the preset sink cleanliness threshold value comprises a first preset sink cleanliness threshold value and a second preset sink cleanliness threshold value, and the first preset sink cleanliness threshold value is smaller than the second preset sink cleanliness threshold value; If the sink cleanliness evaluation index is less than or equal to the first preset sink cleanliness threshold value, it is determined that the cleaning state is clean, and the cleaning scheme is no cleaning; If the sink cleanliness evaluation index is greater than the first preset sink cleanliness threshold value and less than or equal to the second preset sink cleanliness threshold value, it is determined that the cleaning state is slightly contaminated, and the cleaning scheme is to clean the sink by the linkage cleaning equipment; If the sink cleanliness evaluation index is greater than the second preset sink cleanliness threshold value, it is determined that the cleaning state is severely contaminated, and the cleaning scheme is to clean the sink by the linkage cleaning equipment and manual operation and maintenance.

5. The sink algae identification and removal method based on deep learning according to claim 4, characterized in that, Further comprising: obtaining a mean value of the predicted moss measurement index corresponding to the historical same period of the real-time image acquisition time; comparing the predicted moss measurement index with the mean value of the predicted moss measurement index to obtain a moss measurement deviation rate; threshold comparison of the moss measurement deviation rate with a preset moss measurement deviation rate threshold value; If the moss measurement deviation rate is less than or equal to the preset moss measurement deviation rate threshold value, it is determined that the predicted moss measurement index is normal; If the moss measurement deviation rate is greater than the preset moss measurement deviation rate threshold value, it is determined that the predicted moss measurement index is abnormal. 6.The sink algae identification and removal method based on deep learning according to claim 5, wherein, If the moss measurement deviation rate is greater than the preset moss measurement deviation rate threshold value, it is determined that the predicted moss measurement index is abnormal, and then further comprising: obtaining water quality evaluation data and environmental evaluation data of the secondary sedimentation tank sink at the real-time image acquisition time; obtaining historical water quality evaluation data and historical environmental evaluation data of the secondary sedimentation tank sink corresponding to the historical same period of the real-time image acquisition time; According to the water quality evaluation data and environmental evaluation data combined with the historical water quality evaluation data and historical environmental evaluation data, a green moss degree correction coefficient is obtained; According to the green moss degree correction coefficient, the predicted green moss degree index is corrected to obtain an optimized green moss degree index.

7. A sink algae recognition and removal system based on deep learning, characterized in that, The device comprises a memory and a processor, and the memory comprises a deep learning-based sink green moss identification and removal method program. When the deep learning-based sink green moss identification and removal method program is executed by the processor, the following steps are implemented: Real-time images of the sink of the secondary sedimentation tank are collected and preprocessed to obtain initial sink images; The initial sink images are segmented by a preset image segmentation algorithm to obtain regional sink images; A target detection algorithm is used to detect the regional sink images to identify a predicted green moss region and obtain predicted green moss feature data. The predicted green moss feature data are processed to obtain a predicted green moss degree index corresponding to the regional sink images; The predicted green moss degree index is compared with a preset green moss degree threshold value; If the predicted green moss degree index is less than the preset green moss degree threshold value, it is determined that there is no green moss in the predicted green moss region; If the predicted green moss degree index is greater than or equal to the preset green moss degree threshold value, it is determined that there is green moss in the predicted green moss region, and the number and position data of green moss pixels in the predicted green moss region are obtained; The number and position data of green moss pixels are combined with the regional sink images to process the sink cleanliness evaluation index; The sink cleanliness evaluation index is compared with a preset sink cleanliness threshold value to obtain the cleaning state of the sink, and a removal scheme is generated according to the cleaning state; The processing of the number and position data of green moss pixels combined with the regional sink images to obtain the sink cleanliness evaluation index comprises: The number of regional pixels of the regional sink images is extracted from the regional sink images; The number of green moss pixels is compared with the number of regional pixels to obtain the regional green moss coverage rate of the regional sink images, and the sink green moss coverage rate is obtained by averaging processing; The total number of regional sink images in a preset range is obtained, and the number of regional sink images with green moss in the preset range is determined according to the position data; The number of regional sink images is compared with the total number to obtain the regional green moss density corresponding to the preset range, and the sink green moss density is obtained by averaging processing; The regional green moss coverage rate is processed to obtain green moss distribution uniformity data; The sink green moss coverage rate, sink green moss density, and green moss distribution uniformity data are input into a preset sink cleanliness evaluation model for processing to obtain the sink cleanliness evaluation index; The preset sink cleanliness evaluation model is trained by obtaining the sink green moss coverage rate, sink green moss density, and green moss distribution uniformity data of historical samples and corresponding sink cleanliness evaluation indexes.

8. The deep learning-based sink mildew identification and removal system of claim 7, wherein, The target detection algorithm is used for target detection on the area water tank image, a predicted moss area is identified, and predicted moss feature data is obtained, the predicted moss feature data is processed, and a predicted moss measurement index corresponding to the area water tank image is obtained, including: The target detection algorithm is used for target detection on the area water tank image, a predicted moss area is identified, and predicted moss feature data is obtained, including color feature data and texture feature data; The color feature data includes a color histogram and color moments, and the color moments include first-order color moments, second-order color moments, and third-order color moments; The texture feature data includes an LBP histogram and a GLCM feature matrix; The color histogram, color moments, LBP histogram, and GLCM feature matrix are processed, and a predicted moss measurement index corresponding to the area water tank image is obtained.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium includes a water tank moss identification and removal method program based on deep learning, and when the water tank moss identification and removal method program based on deep learning is executed by the processor, the steps of the water tank moss identification and removal method based on deep learning in any one of claims 1 to 6 are implemented.

Images