A method and system for identifying flocs on the surface of sewage

Abstract

The present invention relates to the technical field of image processing, and discloses a method and system for identifying alum flocs on the surface of sewage. A method for identifying alum flocs on the surface of sewage includes: obtaining the control parameters and original water quality parameters during the formation process of alum flocs; extracting the alum floc characteristics of the video of the coagulation and sedimentation process, and identifying and optimizing the key frames based on the genetic algorithm as the images to be identified; using the control parameters, original water quality parameters and key frame image data as inputs, and the alum floc quality as the output to train a CNN neural network model to obtain an alum floc quality identification model; obtaining the suspended solid content through a suspended solid sensor and determining the alum floc generation stage; judging whether the alum floc quality identification result is qualified based on the alum floc quality identification model in the transition stage; when the judgment result is no, adjusting the parameters according to the alum floc quality identification result. This application judges and optimizes the accuracy of various parameters in the coagulation and sedimentation process through the alum floc quality, and improves the sewage treatment efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly relates to a method and system for identifying flocs on the surface of sewage water. Background Art

[0002] The water treatment process in a waterworks generally includes processes such as coagulation, sedimentation, filtration, and disinfection. Among them, the coagulation and sedimentation process is used to remove suspended particles, colloidal substances, and some dissolved organic matters in water. A coagulant is added to the water to cause the fine suspended particles and colloidal substances in the water to aggregate into larger flocs (i.e., flocs), and they are separated from the water by sedimentation using gravity.

[0003] The formation of flocs is the result of charge neutralization and adsorption bridging during the coagulation process. Ideal flocs should have a moderate size, regular shape, and good sedimentation performance. However, affected by factors such as the type of coagulant (such as aluminum salts, iron salts, polyacrylamide), the dosage of coagulant, stirring speed, and raw water quality, the morphology of flocs has diversity, including from fine and loose micro-flocs to larger and compact flocs, thus affecting the overall water treatment effect. Therefore, by effectively identifying flocs, it is convenient to timely control and adjust each link of the coagulation and sedimentation process, and improve the floc formation efficiency.

[0004] Existing image recognition technologies and sensors are difficult to accurately distinguish different morphologies of flocs. Especially when they exist in a mixed state, the shape of flocs is usually irregular, in the shape of snowflakes or clusters, which makes traditional mathematical models based on regular shapes (such as spherical) difficult to apply; extracting effective features from complex morphologies for identification and classification also becomes a difficulty; in addition, the water quality parameters of raw water (such as turbidity, pH value, temperature, dissolved organic matter content, etc.) will also significantly affect the formation and morphology of flocs. Under different water quality conditions, the same dosage of coagulant will produce completely different effects; and the formation of flocs has a lag, and it is difficult to accurately identify the adjustment time node, and the adjustment efficiency is low. Therefore, how to efficiently identify flocs has become an urgent problem to be solved. Summary of the Invention

[0005] The present invention aims to provide a method and system for identifying flocs on the surface of sewage water. By using a genetic algorithm to obtain key frames in the process of floc formation, the quality of flocs is identified based on the key frames, and the accuracy of various parameters in the coagulation and sedimentation process is judged and optimized through the quality of flocs, so as to improve the sewage treatment efficiency.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for identifying flocs on the surface of sewage water, comprising:

[0008] A data acquisition step of acquiring the control parameters and raw water quality parameters in the process of floc formation;

[0009] Image acquisition step: Collect the video of the coagulation and sedimentation process, extract the characteristics of flocs, construct a key frame optimization model based on the genetic algorithm, determine the fitness function based on the floc characteristics, identify and optimize the key frames through the key frame optimization model, and extract the key frames as the images to be recognized;

[0010] Model training step: Use the control parameters, the original water quality parameters, and the key frame image data as the input, and the floc quality as the output to train the CNN neural network model to obtain the floc quality recognition model;

[0011] Stage recognition step: Obtain the suspended solid content through the suspended solid sensor, determine the floc generation stage based on the suspended solid content. The floc generation stage includes the reaction stage and the transition stage, and the transition stage is the stage where the floc stage changes;

[0012] Judgment step: Conduct floc quality recognition in the transition stage, obtain the current stage floc quality recognition result based on the floc quality recognition model, and judge whether the floc quality recognition result is qualified;

[0013] Regulation step: When the judgment result is negative, adjust the parameters according to the floc quality recognition result.

[0014] The principle and advantages of this solution are as follows: In practical applications, the data acquisition step obtains the control parameters and original water quality parameters during the formation process of flocs, covering multiple key factors affecting floc formation, providing a rich information basis for subsequent analysis; the image acquisition step uses computer vision technology to automatically extract features and identify key frames, reducing manual intervention. By using genetic algorithms to customize and select the key frames that can best reflect the changes in floc state, the relevance and effectiveness of the data are improved; the model training step uses multi-modal data input, facilitating the model to learn the complex relationships between information from different sources, being able to consider multiple types of data simultaneously, making full use of all available information, improving the prediction accuracy, adjusting the model structure and output form according to specific requirements, and being applicable to different application scenarios; in the stage recognition step, the floc generation stage is recognized. The floc generation stage includes a rapid mixing stage, a slow mixing stage, a sedimentation stage, and a final inspection stage before filtration after the addition of coagulant; during the transition stage, which is the stage of floc stage change, the floc quality is identified, ensuring the integrity of feature collection in the previous stage and the timeliness of identification. The trained model is used to identify the floc quality in the current stage, and based on the set standards, it is judged whether it is qualified, enabling a rapid assessment result of floc quality, facilitating timely measures to be taken, and ensuring the consistency and fairness of the assessment based on the results given by the model rather than subjective judgment; through the regulation step, the above process is continuously cycled, gradually optimizing the entire coagulation and sedimentation process, and improving the stability and efficiency of the system. In this application, by judging and optimizing the accuracy of various parameters in the coagulation and sedimentation process through floc quality, the coagulation effect can be ensured, the effluent quality can be guaranteed, and at the same time, the problem of lag in floc formation can be overcome.

[0015] Preferably, as an improvement, the image acquisition step includes:

[0016] A stage acquisition sub-step, which acquires the current floc generation stage. The floc generation stage includes a rapid mixing stage, a slow mixing stage, a sedimentation stage, and a final inspection stage before filtration after the addition of coagulant;

[0017] A key frame optimization model construction sub-step, which defines a fitness function and constraint conditions in combination with the floc generation stage, randomly selects an initial frame combination as the initial individual of the population, and optimizes the population through selection, crossover, and mutation operations to find the best key frame combination.

[0018] Technical effect: Through the key frame optimization model, it is convenient to find the best key frame combination, achieving the goal of obtaining sufficient information while reducing the amount of image recognition and saving resources.

[0019] Preferably, as an improvement, the floc characteristics include the morphological characteristics, texture characteristics, and aggregation density characteristics of the flocs.

[0020] Technical effect: Facilitating the comprehensive acquisition of floc characteristics.

[0021] Preferably, as an improvement, the fitness function includes:

[0022]

[0023]

[0024]

[0025] wherein, , , are weight values, and are the pixel values of the i-th frame and the (i - 1)-th frame respectively, and N is the total number of pixels; is the time decay factor, T is the time window size, is the amount of change in image features within the time window T; represents the difference in the morphology j between the i-th frame and the (i - 1)-th frame, is the weight value of the morphology j in the floc formation stage d; represents the amount of color change, represents the amount of texture change, is the balance coefficient.

[0026] Technical effect: Through the fitness function, comprehensive consideration is made from multiple perspectives such as morphological changes, color and texture changes, time and space information, and pixel quality to capture the optimal combination of key frames.

[0027] Preferably, as an improvement, the constraint conditions include:

[0028]

[0029]

[0030]

[0031] wherein, is the index of the k-th key frame, is the set minimum interval, is the actually selected number of key frames, is the set maximum number, is the quality score of the i-th frame, is the minimum quality standard.

[0032] Technical effect: Through the above constraint conditions, it is convenient to prevent the key frames from being too dense, and at the same time limit the number of key frames to ensure that the key frames have high image quality.

[0033] Technical effect: It is convenient to extract features from static and dynamic perspectives, thereby improving the integrity and accuracy of features.

[0034] Preferably, as an improvement, the model training steps include:

[0035] A model selection sub-step, obtaining a training data set and model parameters based on the current floc generation stage;

[0036] An image preprocessing sub-step, enhancing the image and performing labeling;

[0037] A model training sub-step, creating a branch structure to process numerical data and image data, splicing the outputs of the numerical data branch and the image data branch to form a final feature representation, adding an output layer to the spliced features according to task requirements, and training the model.

[0038] Technical effect: It is convenient to use different models at different stages, highlighting the stage characteristics and improving the accuracy of floc recognition.

[0039] Preferably, as an improvement, the model training steps further include a data augmentation sub-step, randomly transforming the input data to obtain an augmented data set.

[0040] Technical effect: Generating more training samples, thereby enhancing the generalization ability of the model.

[0041] Preferably, as an improvement, the model training steps further include a feature weight dynamic adjustment sub-step, dynamically adjusting the importance weights of each input feature by learning a meta-model, using a reinforcement learning algorithm to gradually optimize the combination method of input features according to environmental feedback, and dynamically adjusting the input feature weights of the model based on the optimization results.

[0042] Technical effect: It is convenient to improve the performance of the model when dealing with complex and changeable tasks.

[0043] Preferably, as an improvement, the floc quality recognition model further includes a secondary output layer. When the judgment result obtained in the judgment step is negative, the secondary output is triggered, and the secondary output is a control method.

[0044] Technical effect: It is convenient to make quick adjustments.

[0045] It also includes a recognition system for sewage surface flocs, which uses the above-mentioned recognition method for sewage surface flocs. Description of the Drawings

[0046] Figure 1 It is a schematic flow chart of a recognition method for sewage surface flocs. Detailed Embodiments

[0047] The following is a further detailed description through specific embodiments:

[0048] The embodiment is basically as shown in the appendix Figure 1 A method for identifying alum flocs on the surface of sewage, including:

[0049] Data acquisition step: Obtain the control parameters and original water quality parameters during the formation of alum flocs. The control parameters include the type and dosage of coagulant, stirring speed, and water flow speed. The original water quality parameters include turbidity, pH value, temperature, and dissolved organic matter content. The control parameters and original water quality parameters are closely related to the formation of alum flocs. The type and dosage of coagulant are determined according to the raw water quality. Many particles in water carry negative charges and repel each other, making it difficult to settle. After adding a positively charged coagulant, it can neutralize the negative charges on the particle surface, reduce the electrostatic repulsion, and promote the collision and aggregation of particles. Some polymer coagulants (such as polyacrylamide) have a long-chain structure and can "bridge" multiple particles together through physical adsorption to form larger flocs. Different coagulants have different optimal pH ranges. For example, aluminum salt coagulants have the best effect between pH 6.0 - 7.5, while iron salt coagulants have a better effect between pH 5.0 - 8.5. Rapid stirring helps the coagulant to disperse evenly, and slow stirring helps the formation and growth of alum flocs. The higher the temperature, the stronger the Brownian motion between particles, which is beneficial to the formation of alum flocs, but too high a temperature will also cause incomplete hydrolysis of the coagulant. Too much or too little coagulant will affect the quality of alum flocs. Too much will cause the counterion effect and make the particles redisperse; too little will not be able to effectively neutralize the negative charges on the particle surface. By obtaining the control parameters and original water quality parameters, it is convenient to analyze and adjust the influencing factors when the quality of alum flocs is poor later.

[0050] Stage identification step: Obtain the suspended solid content through a suspended solid sensor, and determine the alum floc generation stage based on the suspended solid content. The alum floc generation stage includes a reaction stage and a transition stage, and the transition stage is the stage where the alum floc stage changes. The transition stage is used for subsequent identification of the quality of alum flocs. The reaction stage is the rapid mixing stage, slow mixing stage, precipitation stage, and final inspection stage before filtration after the coagulant is added. In this embodiment, the alum floc generation stage is identified by the suspended solid content, and the reaction stage of the alum flocs is identified by setting a threshold range of the suspended solid content. The stage where the suspended solid content fluctuates back and forth within the threshold or the suspended solid content reaches the transition threshold is defined as the transition stage, and the transition threshold is determined according to historical data.

[0051] The suspended solid content changes significantly at each stage of floc formation. During the rapid mixing stage after the addition of the coagulant, due to the action of the coagulant, some tiny particles will be temporarily dispersed, resulting in a slight increase or relatively stable suspension solid content in the water. During this stage, the suspended solid content in the water is monitored by an on-line suspended solid sensor. At the end of the rapid mixing stage, the suspended solid content remains at a relatively stable level or shows a slight increase. Compared with the suspended solid content before treatment, there is no significant decrease or substantial increase in the suspended solid content during this stage. In the slow mixing stage, primary aggregates gradually agglomerate to form larger and loosely structured flocs. As the flocs form, the suspended solid particles in the water start to combine together to form larger particle aggregates. As the flocs grow, the suspended solid content will gradually decrease because some particles have been combined into larger flocs, reducing the number of suspended solids. Continuously monitoring the change of the suspended solid content, the suspended solid content gradually decreases during this stage. In the sedimentation stage, the formed flocs further mature and separate from the water to achieve solid-liquid separation. At this time, the suspended solid content decreases significantly, and most of the flocs have settled to the bottom of the tank. Over time, samples are taken at different depths in the sedimentation tank to measure the suspended solid content in the supernatant. The suspended solid content in the supernatant approaches the lowest point during this stage. In the final inspection stage before filtration, most of the flocs in the water should have settled, but there are still a small number of residual particles. Based on a high-precision suspended solid sensor, the suspended solid content in the water is accurately measured. The suspended solid content at this stage should be very low (e.g., less than 5 mg / L).

[0052] Image acquisition step: Collect the video of the coagulation and sedimentation process, construct a key frame optimization model based on the genetic algorithm, identify and optimize the key frames through the key frame optimization model, and extract the key frames as the images to be recognized; the image acquisition step includes a stage acquisition sub-step and a key frame optimization model construction sub-step.

[0053] Stage acquisition sub-step: Obtain the current floc formation stage, which includes the rapid mixing stage after the addition of the coagulant, the slow mixing stage, the sedimentation stage, and the final inspection stage before filtration. The identification of flocs is carried out at several key stages during the tap water treatment process. Each stage has its specific purpose and significance. By carrying out the identification at these stages, the effectiveness and stability of the coagulation and sedimentation process can be ensured, thereby improving the efficiency of the entire water treatment system and the quality of the effluent.

[0054] The rapid mixing stage after the addition of the coagulant is the initial stage of the coagulation reaction, which usually lasts from 30 seconds to 2 minutes. After the coagulant is added to the raw water, rapid mixing is required to quickly and evenly disperse the coagulant into the water and fully contact the suspended particles in the water. By identifying the water sample after mixing in this stage, it is possible to judge whether the coagulant is evenly dispersed, whether there are local high or low concentrations, preliminarily observe whether microflocs begin to form, and their morphology and size. Ensuring the even dispersion of the coagulant and good initial reaction is the basis for the success of subsequent steps. After rapid mixing, it enters the slow mixing stage (also known as the flocculation stage). In this stage, slow stirring is used to promote the gradual growth of microflocs to form larger flocs. In this stage, the growth of the flocs is evaluated, the size, shape and density of the flocs are observed, and it is judged whether they are suitable for sedimentation. The stirring conditions are optimized, and the stirring speed and time are adjusted according to the morphology of the flocs to ensure the best flocculation effect; the slow mixing stage is the key stage for the formation of flocs, which directly determines the quality and sedimentation performance of the flocs. By identifying in this stage, problems (such as insufficient or excessive stirring) can be discovered and corrected in time to ensure that the formed flocs have good sedimentation characteristics. This stage usually lasts for 15 - 30 minutes. After slow mixing, the water sample enters the sedimentation tank, and the flocs sink to the bottom of the tank under the action of gravity to form a sludge layer, while the clear water flows out from the upper part. The sedimentation stage is the key link for removing suspended solids, which directly affects the turbidity and water quality of the effluent. By identifying in this stage, the operating state of the sedimentation tank is ensured, and the decline in the effluent quality caused by too thick sludge layer or poor sedimentation of flocs is avoided. Identification is usually carried out within 5 - 10 minutes after the start of sedimentation. After sedimentation, the water sample will enter the filtration system for further purification. The final inspection before this is to ensure that the water sample after sedimentation meets the standards for entering the filtration system and prevent the filter from being blocked. Identification is carried out before entering the filtration system after sedimentation is completed. The identification includes the turbidity of the effluent and fine flocs or particles.

[0055] The key frame optimization model construction sub-step combines the floc generation stage to define the fitness function and constraint conditions, randomly selects the initial frame combination as the initial individuals of the population, and optimizes the population through selection, crossover and mutation operations to find the best key frame combination.

[0056] The said floc characteristics include the morphological characteristics, texture characteristics and aggregation density characteristics of the flocs. The morphological characteristics include the area, perimeter and shape factor of the flocs:

[0057]

[0058]

[0059]

[0060] Among them, A is the area of the floc, P is the perimeter, is the area of flocs, and C is the boundary of the flocs; is the shape factor, which is used to describe the regularity degree of the flocs. For circular flocs, the shape factor is close to 1; for irregular shapes, the shape factor is smaller.

[0061] The texture features include:

[0062]

[0063] Among them, is the gray value of the central pixel, is the gray value of the neighborhood pixel, and s() is the sign function.

[0064] The aggregation density feature is the degree of aggregation of flocs in a certain area, which is closely related to the effect of the coagulant. The connected component labeling algorithm is used to count the number of flocs in a specific area, and the number of flocs per unit area or the coverage area ratio is calculated to obtain:

[0065]

[0066] Among them, is the number of flocs, is the total area considered.

[0067] The fitness function includes:

[0068]

[0069]

[0070]

[0071] Among them, , , are weight values, and are the pixel values of the i-th frame and the (i - 1)-th frame respectively, and N is the total number of pixels. Considering that the generation of flocs is a dynamic process, not only the feature changes within a single frame need to be concerned, but also the change trend in the time series needs to be considered. Time series analysis methods (such as autoregressive model AR, moving average model MA, etc.) are used to analyze the time series features of floc generation. is the time decay factor, which is used to emphasize the importance of recent frames relative to early frames, T is the size of the time window, is the change amount of the image features within the time window T. By considering the time factor, the trend of floc generation is captured in the time dimension to provide a more comprehensive perspective for selecting key frames. represents the difference in the image morphology j between the i-th frame and the (i - 1)-th frame, $w_{ij}$ is the weight value of the $j$-th morphology in the alum floc formation stage $d$. During the alum floc formation process, its morphology will undergo significant changes. These changes are used to define the fitness function to capture the key morphological transition points. By comparing the changes in the morphological features (area, perimeter, and shape factor) of the alum floc region between adjacent frames, the importance of each frame is evaluated, which can effectively capture the moments when significant morphological changes occur during the alum floc formation process, thus helping to select the most representative key frames; represents the color change amount, represents the texture change amount. The color and texture of the alum flocs are also important features. As the alum flocs grow, their color will become darker and the texture will become more complex. The gray-level co-occurrence matrix (GLCM) or local binary pattern (LBP) is used to extract the texture features, and the color histogram is combined to analyze the color changes, $\beta$ is the balance coefficient, which is used to adjust the influence ratio of color and texture on the fitness.

[0072] The constraint conditions include:

[0073]

[0074]

[0075]

[0076] Among them, $k$ is the index of the $k$-th key frame, $m$ is the set minimum interval. To avoid selecting frames that are too close as key frames, a minimum interval constraint is set to ensure that there is at least a fixed number of frame intervals between every two key frames. This constraint can prevent the key frames from being too dense and improve the efficiency of key frame selection; $N$ is the actual number of selected key frames, $M$ is the set maximum number. To control the number of key frames and avoid an excessive number of key frames causing an overly heavy burden on subsequent processing, a maximum number constraint is set to ensure the efficiency of the processing process; $q_i$ is the quality score of the $i$-th frame, $q_{min}$ is the minimum quality standard. To ensure the quality of the key frames, according to quality indicators such as clarity and contrast, it is ensured that the selected key frames have high image quality for subsequent feature extraction and analysis.

[0077] The selection operation is based on the fitness function, and individuals with higher fitness are selected as the parents of the next generation. The preferred selection methods in this application include, but are not limited to, the roulette wheel selection method:

[0078]

[0079] Among them, $p_i$ is the probability that the $i$-th individual is selected.

[0080] The crossover operation generates new individuals by swapping parts of the genes of two parents. In this application, the one-point crossover method is preferably adopted:

[0081]

[0082] The mutation operation increases the diversity of the population by randomly changing parts of the genes of individuals. In this application, the mutation rate is set to 0.1.

[0083] Model training step: Using the regulation parameters, original water quality parameters, and key frame image data as inputs, and the floc quality as the output, train the CNN neural network model to obtain the floc quality recognition model. Specifically, the model training step includes an image preprocessing sub-step, a model selection sub-step, and a model training sub-step.

[0084] Model selection sub-step: Based on the current floc generation stage, obtain the training data set and model parameters, and perform labeling. The model parameters include feature weight parameters, which facilitate focusing on different influencing factors at different stages and improve the model accuracy.

[0085] Image preprocessing sub-step: Perform enhancement processing on the image and perform labeling. The enhancement processing includes background removal and image enhancement. The background noise is removed through a mixture Gaussian model to highlight the foreground flocs:

[0086]

[0087] Among them, is the current frame, is the background model, is the foreground floc.

[0088] Use methods such as histogram equalization and filtering denoising to enhance the image contrast and clarity. Specifically:

[0089]

[0090] Among them, is the gray value of the input image, is the gray value of the output image, is the gray probability density function of the input image.

[0091] Model training sub-step: Create a branch structure to process numerical data and image data. Use a fully connected layer to process the standardized numerical data, where the numerical data includes control parameters and original water quality parameters. Use a CNN convolutional neural network to extract image features. Concatenate the outputs of the numerical data branch and the image data branch to form the final feature representation. Add an output layer to the concatenated features according to the task requirements. In this embodiment, the task requirement is multi-classification, that is, quality assessment of different levels. Select a loss function, an optimizer, and evaluation metrics, and perform model training.

[0092] The model training step further includes a data augmentation sub-step, which randomly transforms the input data, mainly including randomly transforming the input image data. The random transformation includes rotation, flipping, and cropping to obtain an augmented data set.

[0093] The model training step further includes a feature weight dynamic adjustment sub-step, which dynamically adjusts the importance weights of each input feature by learning a "meta-model", uses a reinforcement learning algorithm to gradually optimize the combination method of the input features according to the environmental feedback, and dynamically adjusts the input feature weights of the model based on the optimization results. Specifically:

[0094] If the model is g(x;u), where x = [x1, x2,..., x n represents the input features, and u = [u1, u2,..., u n represents the corresponding weight vector. The goal is to find the optimal weight vector u * .

[0095] Input the weighted input features into the model:

[0096]

[0097] Define a loss function to measure the difference between the model recognition result value and the true label, and find the optimal weight vector through the optimization objective:

[0098]

[0099] Among them, ) is the loss function, is the optimal weight vector.

[0100] By dynamically adjusting the weights of the input features, the model can automatically adjust its attention to different features according to the specific application scenario or data characteristics, thereby improving the adaptability and robustness of the model. For example, in the water treatment process, during certain periods, water quality parameters are more critical than control parameters, while in other periods, it is the opposite. Dynamically adjusting the weights can help the model better capture these changes.

[0101] Judgment step: During the transition stage, the quality of flocs is identified, and based on the floc quality identification model, the floc quality identification result at the current stage is obtained, and it is judged whether the floc quality identification result is qualified; Regulation step: When the judgment result is negative, parameter adjustment is performed according to the floc quality identification result. The floc quality identification model further includes a secondary output layer. When the judgment result obtained in the judgment step is negative, secondary output is triggered, and the secondary output is a regulation method, and the regulation method includes adjusting the coagulant dosage, optimizing the stirring conditions, and adjusting the pH value.

[0102] It also includes a recognition system for the flocs on the surface of sewage, which uses the above-mentioned recognition method for the flocs on the surface of sewage.

[0103] The above are only embodiments of the present invention. Specific technical solutions and / or common knowledge such as characteristics well known in the art are not described in detail herein. It should be noted that for those skilled in the art, without departing from the technical solution of the present invention, several deformations and improvements can be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application should be subject to the content of its claims, and the specific implementation manners and the like recorded in the specification can be used to interpret the content of the claims.

Claims

1. A method for identifying alum bloom on the surface of sewage, characterized in that: include: Data acquisition step, obtaining the control parameters and original water quality parameters in the process of alum floc formation; The image acquisition step includes collecting a video of the coagulation and sedimentation process, extracting the features of the alum flower, constructing a key frame optimization model based on a genetic algorithm, determining a fitness function based on the features of the alum flower, identifying and optimizing key frames through the key frame optimization model, and extracting the key frames as images to be identified; The image acquisition step comprises: The stage acquisition sub-step is to acquire the current alum floc reaction stage, wherein the alum floc reaction stage includes a rapid mixing stage after the addition of coagulant, a slow mixing stage, a sedimentation stage, and a final inspection stage before filtration; The key frame optimization model construction sub-step defines the fitness function and constraints in combination with the alum flower reaction stage, randomly selects the initial frame combination as the initial individual of the population, optimizes the population through selection, crossover and mutation operations, and finds the best key frame combination; The fitness function includes: in, , , is the weight value, and are the pixel values ​​of the i-th frame and the i-1-th frame respectively, and N is the total number of pixels; is the time decay factor, T is the time window size, is the change of image features within the time window T; represents the difference between the image morphology j of the i-th frame and the i-1-th frame, is the weight value of morphology j in the alum flower reaction stage d; Indicates the amount of color change, Indicates the amount of texture change, is the balance coefficient; The constraints include: in, is the index of the kth keyframe, is the minimum interval set, is the number of keyframes actually selected, is the maximum number set, is the quality score of the i-th frame, is the minimum quality standard; The model training step takes the control parameters, original water quality parameters and the image to be identified as input, takes the alum mass as output, trains the CNN neural network model, and obtains the alum mass identification model; the control parameters include the type and dosage of coagulant, stirring speed, and water flow speed; The stage identification step is to obtain the suspended solid content through the suspended solid sensor, and determine the alum floc formation stage through the suspended solid content. The alum floc formation stage includes a reaction stage and a transition stage. The transition stage is the stage when the suspended solid content reaches the transition threshold; A judgment step is to identify the quality of the alum flower in the transition stage, obtain the alum flower quality identification result of the current stage based on the alum flower quality identification model, and judge whether the alum flower quality identification result is qualified; In the control step, when the judgment result is no, the parameters are adjusted according to the alum flower quality identification result.

2. A method for identifying alum bloom on sewage surface according to claim 1, characterized in that: The alum flower characteristics include morphological characteristics, texture characteristics, and aggregation density characteristics of alum flowers.

3. A method for identifying alum bloom on sewage surface according to claim 1, characterized in that, The model training step includes: The model selection sub-step obtains the training data set and model parameters based on the current bloom generation stage; The image preprocessing sub-step is to enhance the image and label it; In the model training sub-step, a branch structure is created to process numerical data and image data, the outputs of the numerical data branch and the image data branch are concatenated to form the final feature representation, an output layer is added to the concatenated features according to task requirements, and model training is performed.

4. A method for identifying alum bloom on sewage surface according to claim 1, characterized in that: The model training step also includes a data enhancement sub-step, which randomly transforms the input data to obtain an enhanced data set.

5. A method for identifying alum bloom on sewage surface according to claim 1, characterized in that: The model training step also includes a feature weight dynamic adjustment sub-step, which dynamically adjusts the importance weight of each input feature by learning a meta-model, uses a reinforcement learning algorithm to gradually optimize the combination of input features according to environmental feedback, and dynamically adjusts the input feature weights of the model based on the optimization results.

6. A method for identifying alum bloom on sewage surface according to claim 1, characterized in that: The alum flower quality identification model also includes a secondary output layer. When the judgment result obtained in the judgment step is no, the secondary output is triggered, and the secondary output is a control method.

7. A system for identifying alum bloom on the surface of sewage, characterized in that: A method for identifying alum bloom on the surface of sewage as described in any one of claims 1-6 is used.

Images