A rotifer population breeding-based activated sludge water quality state identification method

Abstract

The application discloses a kind of active sludge water quality state identification method based on rotifer population breeding, it is related to sewage treatment technical field, the method includes: collecting active sludge sample is placed in the culture device that can simulate aeration tank environment;In monitoring period, microscopic imaging is carried out to sample regularly, and time sequence image sequence is obtained;Time sequence image sequence is input into pre-training ViT model, and the type state, position and identity feature vector of rotifer individual are accurately identified and extracted;Based on feature vector, identity matching and continuous tracking of rotifer individual across time point are realized, and trajectory database containing whole life cycle trajectory is constructed;Based on the trajectory database, multiple characteristic parameters related to population breeding are calculated;These parameters are input into pre-trained water quality state classification model, and corresponding water quality state grade is automatically output.The method improves the accuracy and early warning timeliness of water quality state identification result significantly by fine-grained identification, individual tracking and multi-dimensional feature analysis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sewage treatment, in particular to an activated sludge water quality state identification method based on rotifer population proliferation. BACKGROUND

[0002] As the most widely used sewage treatment technology today, the core of activated sludge method relies on the degradation of pollutants by microbial communities. The stability of the system operation and the compliance rate of the effluent quality directly depend on the accurate perception and timely regulation of the ecological-technological state in the aeration tank. In this system, rotifers, as indicator organisms at the level of metazoan, can intuitively reflect the activity of sludge and the water quality. Current rotifer monitoring mainly relies on two methods:

[0003] 1. Artificial microscope observation method: the staff regularly collects activated sludge samples, counts the number of rotifers and observes the morphology through a microscope, judges the reproduction (such as whether there are egg capsules, the proportion of juveniles), and then correlates the water quality state identification. This method requires staff to observe each sample and manually record data, which is extremely inefficient and the data is seriously lagging. And it depends on the experience of personnel, and it is easy to have subjective errors in distinguishing rotifer juveniles / adults, summer eggs / winter eggs, and females / males, resulting in inaccurate water quality state identification results.

[0004] 2. Machine recognition method based on YOLO: by installing a camera at the monitoring point, using the YOLO target detection model to automatically identify rotifers in the image, replacing part of the manual operation to achieve preliminary statistics of the number of rotifers. This method cannot accurately distinguish between rotifer juveniles and adults, summer eggs and winter eggs, and cannot track the proliferation process of individual rotifers; and YOLO is prone to confuse different rotifer individuals with similar shapes, resulting in high error in population number statistics, thus leading to inaccurate water quality state identification results.

[0005] Therefore, due to the single data dimension (mainly rough number) and insufficient information depth (lack of fine-grained classification and dynamic process) of the prior art, the water quality state identification results based on rotifer indicators are inaccurate and not timely, so that the regulation of the sewage treatment process is often based on incomplete or lagging information, which restricts the further improvement of the process operation stability, energy optimization and risk prevention and control capability. SUMMARY

[0006] The technical problem to be solved by the present application is how to improve the accuracy of activated sludge water quality state identification results, and the purpose is to provide an activated sludge water quality state identification method based on rotifer population proliferation, which solves the above problems.

[0007] The present application is realized by the following technical solutions:

[0008] In a first aspect, the present application provides a method for identifying the water quality state of activated sludge based on the population breeding of rotifers, comprising:

[0009] collecting a target sludge sample from an activated sludge aeration tank and placing the target sludge sample in a culture device;

[0010] adjusting the simulated environmental parameters in the culture device to dynamically simulate the actual environmental conditions of the activated sludge aeration tank within a preset monitoring period;

[0011] controlling a microscopic imaging device to perform microscopic imaging on the target sludge sample in the culture device every first preset time interval within the preset monitoring period to obtain a time sequence of images;

[0012] inputting the time sequence of images into a pre-trained visual Transformer model to output the type state, position coordinates and feature vector of each rotifer individual in each image;

[0013] based on the feature vector, performing identity matching and tracking of rotifer individuals across time points to generate a trajectory database of all rotifers; the trajectory database includes the identity ID of all rotifers, the type state and position coordinates of each rotifer individual at different time points;

[0014] based on the trajectory database, calculating a plurality of feature parameters related to population breeding, and inputting the plurality of feature parameters into a pre-trained water quality state classification model to output the water quality state of the target sludge sample.

[0015] Optionally, the collecting a target sludge sample from an activated sludge aeration tank and placing the target sludge sample in a culture device comprises:

[0016] collecting sludge samples from a plurality of sampling points of the activated sludge aeration tank;

[0017] mixing the collected plurality of sludge samples to obtain a mixed sludge sample;

[0018] taking part of the mixed sludge sample as the target sludge sample and placing it in the culture device.

[0019] Optionally, the culture device comprises:

[0020] a culture dish for containing the target sludge sample;

[0021] a semiconductor temperature controller for adjusting the temperature of the target sludge sample;

[0022] a micro air pump for introducing air into the bottom of the culture dish;

[0023] A sterile filter membrane is arranged between the micro air pump and the culture dish for filtering and sterilizing the air supplied by the micro air pump;

[0024] A temperature sensor is arranged for collecting a water temperature monitoring value of the target sludge sample;

[0025] A dissolved oxygen sensor is arranged for collecting a dissolved oxygen concentration monitoring value of the target sludge sample;

[0026] A magnetic stirrer is arranged for driving the culture dish to rotate at a preset speed.

[0027] Optionally, the environmental parameters in the culture device are adjusted to dynamically simulate the actual environmental conditions of the activated sludge aeration tank within a preset monitoring period, including:

[0028] Within the preset monitoring period, the water temperature monitoring value collected by the temperature sensor and the dissolved oxygen concentration monitoring value collected by the dissolved oxygen sensor are synchronously read every second preset time interval;

[0029] The water temperature reference value and the dissolved oxygen concentration reference value collected by the environmental sensor arranged in the activated sludge aeration tank are synchronously acquired;

[0030] The first deviation between the water temperature monitoring value and the water temperature reference value, and the second deviation between the dissolved oxygen concentration monitoring value and the dissolved oxygen concentration reference value are calculated;

[0031] If the absolute value of the first deviation is greater than a first preset deviation threshold, a first control signal is sent to the semiconductor temperature controller to drive the semiconductor temperature controller to adjust so that the water temperature monitoring value approaches the water temperature reference value;

[0032] If the absolute value of the second deviation is greater than a second preset deviation threshold, a second control signal is sent to the micro air pump to drive the micro air pump to adjust so that the dissolved oxygen concentration monitoring value approaches the dissolved oxygen concentration reference value.

[0033] Optionally, the microscopic imaging device includes a motorized stage, a microscope objective lens, and an industrial camera; and the microscopic imaging device is controlled to perform microscopic imaging on the target sludge sample in the culture device every first preset time interval within the preset monitoring period to obtain a time sequence of images, including:

[0034] The motorized stage is controlled to move according to a preset path and stop at a plurality of preset imaging positions in sequence every first preset time interval within the preset monitoring period;

[0035] When the electric stage stays at each fixed imaging position, the industrial camera is controlled to continuously collect multiple frames of images of the target sludge sample magnified by the microscope objective lens;

[0036] All collected images are organized according to the collection time and position information to form a time sequence image sequence.

[0037] Optionally, the visual Transformer model comprises an input layer, a feature extraction layer and an output layer; the time sequence image sequence is input into the pre-trained visual Transformer model, and the type state, position coordinates and feature vector of the rotifer individual in each image are output, comprising:

[0038] Each frame of image in the time sequence image sequence is preprocessed to generate preprocessed image data;

[0039] The preprocessed image data is received by the input layer;

[0040] The preprocessed image data is encoded and feature-extracted by the feature extraction layer to generate a feature vector of each image;

[0041] The feature vector of each image is input into the classification head, positioning head and feature embedding head of the output layer in parallel;

[0042] The feature vector of each image is processed by the classification head to output the type and state of the rotifer individual in the corresponding image;

[0043] The feature vector of each image is processed by the positioning head to output the position coordinates of the rotifer individual in the corresponding image;

[0044] The feature vector of each image is processed by the feature embedding head to output the feature vector of the rotifer individual in the corresponding image.

[0045] Optionally, the preprocessed image data is generated by preprocessing each frame of image in the time sequence image sequence, comprising:

[0046] Each frame of image in the time sequence image sequence is subjected to size normalization processing to obtain images with uniform preset size;

[0047] For the images with uniform preset size, a dense area in which the rotifer distribution density exceeds a preset density threshold is identified and determined;

[0048] The dense area is subjected to image segmentation to generate a plurality of sub-image blocks with fixed size;

[0049] The non-dense area in the image and the sub-image blocks segmented from the dense area jointly constitute the preprocessed image data.

[0050] Optionally, the plurality of characteristic parameters include population change rate, asexual reproduction efficiency, sexual reproduction proportion, average reproduction cycle, larva survival rate, adult mortality rate, and deformity rate.

[0051] Optionally, the water quality state classification model is trained in the following manner:

[0052] A historical monitoring data set containing a plurality of samples is obtained; each sample contains actual values of the plurality of characteristic parameters identified and tracked based on a historical time sequence image sequence;

[0053] For each characteristic parameter in the plurality of characteristic parameters, a plurality of water quality state threshold ranges corresponding to the characteristic parameter are preset according to the rotifer ecology response principle; the water quality state threshold ranges include at least a normal range, a slightly deviated range, and an abnormal range;

[0054] For each sample in the historical monitoring data set, state determination is performed according to the threshold range in which the actual values of the characteristic parameters of the sample are located;

[0055] Based on the state determination result and a predetermined comprehensive decision rule, a water quality state category label corresponding to each sample is automatically generated;

[0056] The actual values of the plurality of characteristic parameters are taken as input features, and the corresponding water quality state category labels are taken as supervision targets, so as to perform supervised learning training on an initial classification model, thereby obtaining a trained water quality state classification model.

[0057] Optionally, based on the state determination result and the predetermined comprehensive decision rule, the water quality state category label corresponding to each sample is automatically generated, including:

[0058] If the actual values of the plurality of characteristic parameters in the sample are all located in the corresponding normal threshold range, the water quality state label of the corresponding sample is labeled as a high-quality state;

[0059] If the actual values of 1 to 2 characteristic parameters in the sample are located in the corresponding slightly deviated threshold range, and the actual values of other characteristic parameters are all located in the corresponding normal threshold range, the water quality state label of the corresponding sample is labeled as a good state;

[0060] If the actual values of 1 to 3 characteristic parameters in the sample are located in the corresponding abnormal threshold range, the water quality state label of the corresponding sample is labeled as a slightly polluted state;

[0061] If the actual values of 4 or more characteristic parameters in the sample are located in the corresponding abnormal threshold range, the water quality state label of the corresponding sample is labeled as a severely polluted state.

[0062] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0063] The present application provides a kind of active sludge water quality state identification method based on rotifer population breeding, this method by introducing visual Transformer (ViT) model, utilize its powerful global attention mechanism, realize the accurate, automated identification of fine-grained biological characteristics, solve the subjective error problem of traditional method identification rough, rely on artificial experience.Based on the feature vector extracted by ViT model, the identity matching and tracking of rotifer individuals across time points are carried out, and the trajectory database of all rotifers is generated, a unique identity ID is established for the rotifer individuals, and the type state and position coordinates of the rotifer individuals in the monitoring period are continuously recorded.Based on the trajectory database of rotifer individuals in whole life cycle, a plurality of characteristic parameters related to population breeding are calculated, and the key life activities and dynamic response processes of rotifer population are quantified from multiple dimensions.These multiple characteristic parameters are input into the pre-trained water quality state classification model, and the model can comprehensively weigh the synergistic changes of various indexes by learning the complex nonlinear mapping relationship between the multidimensional characteristics and water quality levels in historical data, so as to realize more accurate and more robust intelligent discrimination than artificial experience or single threshold rule, thereby significantly improving the accuracy of water quality state identification result.The method upgrades the water quality identification basis from static "quantity statistics" to dynamic "process response analysis", thereby significantly improving the identification accuracy, sensitivity and timeliness, and realizing the change from "after-event discovery" to "pre-event warning". BRIEF DESCRIPTION OF DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the example embodiments of the present application, the following will briefly introduce the drawings needed to be used in the examples. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor. In the drawings:

[0065] Figure 1 The flowchart of the active sludge water quality state identification method based on rotifer population breeding provided by the example of the present application is shown in the figure;

[0066] Figure 2 The structure diagram of the culture device provided by the example of the present application is shown in the figure;

[0067] Figure 3 The schematic diagram of the image acquisition process provided by the example of the present application is shown in the figure;

[0068] Figure 4 The structure diagram of the visual Transformer model provided by the example of the present application is shown in the figure;

[0069] Figure 5Part of the schematic diagram of rotifer individual trajectory data provided for the embodiments of the present application;

[0070] Figure 6 The schematic diagram of all rotifer key events provided for the embodiments of the present application;

[0071] Figure 7 The structural schematic diagram of the activated sludge water quality state identification system based on rotifer population breeding provided for the embodiments of the present application. DETAILED DESCRIPTION

[0072] In order to make the objects, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application in combination with embodiments and drawings, and the illustrative embodiments of the present application and the description thereof are only used to explain the present application, and do not limit the present application.

[0073] For a water plant, rotifers live in sludge. The survival and reproduction of rotifers directly depend on key indicators of water quality, and changes in different indicators will quickly reflect on the population characteristics of rotifers:

[0074] 1. Dissolved oxygen (DO): Most rotifers are aerobic organisms, and their population will thrive when DO≥2mg / L; when DO<0.5mg / L, the population number will decrease sharply, only a small amount of low-oxygen-tolerant species (such as rotifer) will survive, and even trigger sexual reproduction to produce winter eggs.

[0075] 2. Organic load (F / M): When the organic load in activated sludge is too high, bacteria will reproduce in large numbers, rotifers will have sufficient food but the environment will be turbid, and the population may grow in the short term but will decline later due to the decrease of dissolved oxygen; when the load is too low, rotifers will lack food and their reproduction will be inhibited, and the population will decrease.

[0076] 3. Pollutants (toxic substances, heavy metals): Rotifers have thin body walls and no shell protection, and are sensitive to pollutants such as heavy metals, antibiotics, and pesticides, and low-concentration pollutants can cause the reproduction rate of rotifers to decrease, the mortality rate of juveniles to increase, and the deformity rate to increase (such as body bending and loss of cilia), and the population will die out in severe pollution.

[0077] Rotifers feed on bacteria, algae, and small protozoa, and can control the over-reproduction of bacteria and prevent sludge from swelling (the number of rotifers often decreases sharply when the sludge swells); when the rotifer population is thriving, it indicates that the sludge has good settling performance, high organic pollutant degradation efficiency, and high probability of meeting water quality standards; on the contrary, the number of rotifers decreases sharply or disappears, which often indicates that the sludge has lost its function (such as poisoning and anoxia). Therefore, the core function of activated sludge is to degrade organic pollutants, and rotifers, as "top consumers of metazoans", can reflect the "activity" and "stability" of the sludge.

[0078] In summary, rotifers are key water quality indicator organisms in activated sludge systems. Current monitoring methods, such as manual microscopy or general object detection models like YOLO, have coarse recognition granularity and lack temporal correlation, resulting in low accuracy in identifying activated sludge water quality status.

[0079] Therefore, the present application provides a method for identifying the water quality status of activated sludge based on the population proliferation of rotifers. Please refer to Figure 1 A flowchart of the method for identifying the water quality status of activated sludge based on the population proliferation of rotifers is provided in the present application. The following will introduce Figure 1 the method for identifying the water quality status of activated sludge based on the population proliferation of rotifers as shown in the figure.

[0080] S1, collect a target sludge sample from the activated sludge aeration tank and place it in a culture device.

[0081] In one possible embodiment, sludge samples are collected from multiple sampling points in the activated sludge aeration tank; the collected sludge samples are mixed and treated to obtain a mixed sludge sample; and part of the mixed sludge sample is taken as a target sludge sample and placed in a culture device.

[0082] In the specific implementation process, the sampling process is as follows:

[0083] S1.1, sampling point selection: determine multiple sampling points (e.g., 3) at the mixed liquor zone of the activated sludge aeration tank, 1 / 3 height from the bottom, and uniform water flow disturbance.

[0084] S1.2, sample collection: collect 15 mL of sludge mixed liquor (i.e., sludge sample) at each sampling point and inject it into a sterile centrifuge tube.

[0085] S1.3, sample mixing: pour the sludge samples from multiple sampling points into a stirrer and stir for 2 minutes to obtain a mixed sludge sample.

[0086] S1.4, sample processing: take 5 mL of the mixed sludge sample as a target sludge sample and inject it into a dedicated culture device, and the remaining sample is reserved for re-inspection.

[0087] In the embodiments of the present application, the "multi-point mixed sampling" strategy effectively overcomes the accidental errors caused by the micro-uniformity of activated sludge (such as zoogleal ball distribution and local flow dead angle) in single-point sampling. The target sludge sample obtained can represent the overall status of the rotifer population in the entire aeration tank section to the greatest extent, providing a statistically reliable analysis basis for subsequent continuous and accurate monitoring for up to 12 hours. The uniform mixing treatment makes the rotifer individuals uniformly distributed in the sample, avoiding the systematic bias caused by uneven initial distribution on subsequent microscopic imaging and individual counting, which ensures the accuracy of the population data at the starting time (t=0) of monitoring, making the dynamic indicators such as population change rate and reproduction efficiency calculated subsequently more real and reliable.

[0088] S2, adjust the simulated environment parameters in the culture device to dynamically simulate the actual environment conditions of the activated sludge aeration tank within a preset monitoring period.

[0089] Please refer to Figure 2 for a structural schematic diagram of a culture device provided in the embodiments of the present application. The culture device includes a culture dish, a semiconductor temperature controller, a micro air pump, a sterile filter membrane, a temperature sensor, a dissolved oxygen sensor, and a magnetic stirrer, which are introduced as follows.

[0090] The culture dish is used to hold the target sludge sample and is made of transparent material, which facilitates non-destructive in-situ observation by the microscopic imaging equipment above;

[0091] The semiconductor temperature controller is used to adjust the temperature of the target sludge sample;

[0092] The micro air pump is used to introduce air into the bottom of the culture dish;

[0093] The sterile filter membrane is arranged between the micro air pump and the culture dish and is used to filter and sterilize the air introduced by the micro air pump;

[0094] The detection end of the temperature sensor is immersed in the target sludge sample and is used to collect the water temperature monitoring value of the target sludge sample;

[0095] The detection end of the dissolved oxygen sensor is immersed in the target sludge sample and is used to collect the dissolved oxygen concentration monitoring value of the target sludge sample;

[0096] The magnetic stirrer is arranged below the culture dish and is used to drive the culture dish to rotate at a preset speed. The preset speed (such as 30 r / min) is within the tolerable range of rotifers.

[0097] In the embodiments of the present application, a culture device integrating environmental simulation, precise control, in-situ observation and pollution prevention is provided. A semiconductor temperature controller combined with an immersion type temperature sensor realizes fast response and precise closed-loop control of the sample temperature. A micro air pump combined with the feedback of a dissolved oxygen sensor precisely adjusts the rate and duration of the air input to stably maintain the required dissolved oxygen level. A sterile filter effectively prevents the introduction of bacteria from the air source, maintains the original structure and balance of the microbial community (including rotifers and their food chain) in the sample, and enables the observed changes in the state of rotifers to be more accurately attributed to the original water quality or process conditions rather than accidental biological pollution. The low-speed stirring of the magnetic stirrer prevents the sedimentation and aggregation of sludge flocs and rotifers, ensures the uniformity and representativeness of the sample, and simulates the weak water flow movement in the aeration tank with gentle disturbance, thereby improving the ecological simulation degree.

[0098] In a possible embodiment, within a preset monitoring period, the water temperature monitoring value collected by the temperature sensor and the dissolved oxygen concentration monitoring value collected by the dissolved oxygen sensor are synchronously read every second preset time interval. The water temperature reference value and the dissolved oxygen concentration reference value collected by the environmental sensors deployed in the activated sludge aeration tank are synchronously obtained. The first deviation between the water temperature monitoring value and the water temperature reference value and the second deviation between the dissolved oxygen concentration monitoring value and the dissolved oxygen concentration reference value are calculated. If the absolute value of the first deviation is greater than a first preset deviation threshold, a first control signal is sent to the semiconductor temperature controller to drive the semiconductor temperature controller to adjust so that the water temperature monitoring value approaches the water temperature reference value. If the absolute value of the second deviation is greater than a second preset deviation threshold, a second control signal is sent to the micro air pump to drive the micro air pump to adjust so that the dissolved oxygen concentration monitoring value approaches the dissolved oxygen concentration reference value.

[0099] In the specific implementation process, within a preset monitoring period (such as 12 hours), the system automatically performs an environmental calibration cycle every second preset time interval (such as 30 minutes), and the calibration cycle includes the following sub-steps:

[0100] Firstly, the readings of the temperature sensor and the dissolved oxygen sensor deployed in the culture device are synchronously read to obtain the current water temperature monitoring value and the dissolved oxygen concentration monitoring value DOm.

[0101] Secondly, through a data communication interface, the real-time data collected by the environmental sensors (including temperature sensors and dissolved oxygen sensors) deployed in the activated sludge aeration tank are synchronously obtained as the water temperature reference value and the dissolved oxygen concentration reference value at the current time within the monitoring period.

[0102] Then, the first deviation between the water temperature monitoring value and the water temperature reference value Tr is calculated. And the second deviation between the dissolved oxygen concentration monitoring value and the dissolved oxygen concentration benchmark value. :

[0103]

[0104]

[0105] If the absolute value of the first deviation If the deviation exceeds a first preset threshold (e.g., 1°C), the system sends a first control signal to the semiconductor temperature controller. The direction (heating or cooling) and amplitude of the first control signal are determined based on the first deviation. The sign and magnitude are determined to drive the semiconductor temperature controller to work, so that... rapidly approaching If the absolute value of the first deviation If the value is less than or equal to the first preset deviation threshold (e.g., 1°C), the parameters of the semiconductor temperature controller will remain unchanged.

[0106] If the absolute value of the second deviation If the deviation exceeds a second preset threshold (e.g., 0.5 mg / L), the system sends a second control signal to the micro air pump, adjusting the pump's ventilation rate or start-stop cycle to ensure... rapidly approaching If the absolute value of the second deviation If the value is less than or equal to the second preset deviation threshold (e.g., 0.5 mg / L), the parameters of the micro air pump will remain unchanged.

[0107] In this embodiment, by periodically and automatically executing the aforementioned closed-loop control process of "monitoring-comparison-adjustment," the system can dynamically and accurately track and simulate the actual environment of the original activated sludge aeration tank, including temperature and dissolved oxygen concentration within the cultivation device, throughout the entire long-term monitoring period. This effectively ensures that the rotifers grow and reproduce under conditions highly consistent with their native environment during the monitoring period, thereby ensuring that subsequent water quality analyses based on their population behavior and state changes have high ecological authenticity and data reliability.

[0108] S3. Within the preset monitoring period, every first preset time interval, control the microscopic imaging device to perform microscopic imaging on the target sludge sample in the culture device to obtain a time-series image sequence.

[0109] In one possible embodiment, to acquire high-quality dynamic images of rotifers, the microscopic imaging apparatus includes:

[0110] Electric platform;

[0111] An optical system is fixedly installed on the motorized stage, including a microscope objective (such as a biological microscope lens with a magnification of 40x) and an industrial camera (such as a high-definition camera with a resolution of 2048x1536 pixels); the microscope objective is used to optically magnify the target sludge sample in the culture device; and the industrial camera is used to collect images of the target sludge sample magnified by the microscope objective.

[0112] An illumination system is fixedly installed on the motorized stage and used to provide illumination to the target sludge sample. The illumination system can adopt a ring-shaped LED light source and adopt oblique illumination to reduce reflection.

[0113] In the embodiments of the present application, a micro-imaging device with an integrated design is provided, by fixing the optical system and the illumination system on the same motorized stage, it is ensured that the relative position relationship between the illumination angle, light intensity, and light path and the sample remains constant during movement and shooting. This fundamentally eliminates problems such as uneven lighting, shadow changes, or focus drift caused by relative displacement of components, and provides high-quality and consistent input images for subsequent models.

[0114] In a possible embodiment, the specific steps of S3 include:

[0115] In a preset monitoring period, every first preset time interval, the motorized stage is controlled to move according to a preset path and stop at multiple preset imaging positions in turn; when the motorized stage stops at each fixed imaging position, the industrial camera is controlled to collect continuous multiple frames of images of the target sludge sample magnified by the microscope objective; and all collected images are organized according to the collection time and position information to form a time sequence image sequence.

[0116] In the specific implementation process, first, according to the size and shape of the culture dish in the culture device, multiple (such as 10) imaging position coordinates are pre-set and stored in the control system of the micro-imaging device, to ensure that the union of the micro-imaging fields corresponding to the multiple preset imaging positions covers more than 90% of the area of the culture dish.

[0117] Please refer to Figure 3 An image collection process provided by the embodiments of the present application is shown in the following figure. After the target sludge sample is placed in the culture device and a preset delay (such as 30 minutes) is started, the rotifers in the target sludge sample are allowed to recover from the sampling disturbance and adapt to the new culture environment, to ensure that the recorded initial state is more representative. After the delay ends, in a preset monitoring period (such as 12 hours), the system enters an automatic periodic collection cycle with a fixed first preset time interval (such as 30 minutes) as a period:

[0118] The motorized stage is controlled to move sequentially according to the preset path and stop at the preset imaging positions in turn. After the motorized stage is stably stopped at each imaging position, the industrial camera is triggered to capture multiple frames (e.g., 5 frames) of images of the target sludge sample in the fixed field of view. By capturing multiple images in a very short time (milliseconds), the missed detection in a single frame caused by the rapid and random movement of the rotifer individuals is effectively offset, and the capture probability and statistical integrity of the rotifer individuals in the area at each stop position are significantly improved. At the same time of capturing each frame of image, a precise time stamp and imaging position coordinates are automatically attached to each frame of image.

[0119] After completing the collection work of all cycles in a monitoring period, the system automatically sorts and organizes all images according to the time stamp and imaging position coordinates of the images, and finally generates a complete and ordered time sequence image sequence.

[0120] It should be noted that the first preset time interval and the second preset time interval can be the same or different.

[0121] In the embodiments of the present application, a standardized image collection process is constructed, which not only realizes the continuous observation of the rotifer population dynamics with high spatiotemporal resolution and unattended, but also completely replaces the traditional discrete and manual sampling and microscopic examination mode. The time sequence image sequence generated thereby completely records the dynamic change process of the rotifer population in the target sludge sample in the spatial distribution and time dimension within the preset monitoring period, and provides high-quality continuous image data stream for accurately quantifying the life cycle events (hatching, spawning, death) of rotifer individuals and the population reproduction characteristics (reproduction rate, survival rate).

[0122] S4, inputting the time sequence image sequence into a pre-trained visual Transformer model to output the type state, position coordinates and feature vector of each rotifer individual in the image.

[0123] Please refer to Figure 4 The structural diagram of the visual Transformer model provided in the embodiments of the present application is shown. The visual Transformer (ViT) model includes an input layer, a feature extraction layer and an output layer, and the output layer includes three branches of a classification head, a positioning head and a feature embedding head.

[0124] In a possible embodiment, the specific steps of S4 include:

[0125] The pre-processing is performed on each image in the time sequence image sequence to generate pre-processed image data; the pre-processed image data is received through an input layer; the pre-processed image data is encoded and feature extracted through a feature extraction layer to generate a feature vector of each image; the feature vector of each image is input into a classification head, a positioning head and a feature embedding head of an output layer in parallel; the feature vector of each image is processed through the classification head to output the type and state of the rotifer individual in the corresponding image; the feature vector of each image is processed through the positioning head to output the position coordinates of the rotifer individual in the corresponding image; and the feature vector of each image is processed through the feature embedding head to output the feature vector of the rotifer individual in the corresponding image.

[0126] In the specific implementation process, first, the time sequence image sequence is pre-processed, the pre-processed image data is received through the input layer of the ViT model, the feature extraction layer of the ViT model is constructed based on the Transformer encoder architecture, the image data received by the input layer is hierarchically encoded and feature abstracted, and a feature vector containing multi-scale semantic information is output. Then, the feature vector is simultaneously fed to three branches of the output layer for parallel processing.

[0127] In the classification head branch, the feature vector of each image is decoded and classified mapped to output the class (such as female, male and larva) and state (such as survival, death and egg laying) of each detected rotifer individual in the current image.

[0128] In the positioning head branch, the feature vector of each image is decoded and coordinate-regressed to output the bounding box coordinates, i.e. the position coordinates, of each rotifer individual in the current image pixel coordinate system.

[0129] In the feature embedding head branch, the feature vector of each image is transformed to calculate and output a fixed-dimension feature vector for each detected rotifer individual. The feature vector aims to uniquely represent the visual appearance of the rotifer individual in the Euclidean space, and its core use is to support subsequent individual identity matching and continuity tracking across different time frame images.

[0130] In the embodiments of the present application, the ViT model is optimized, and through multi-task parallel learning of the feature vector, the three tasks of classification, positioning and feature extraction can promote each other, the positioning information helps the classification to focus on the correct area, and the semantic information of the classification guides the feature extraction to be more discriminative. This design is particularly beneficial to maintaining high recognition and segmentation accuracy in complex sludge images with dense rotifers and partial occlusion. Through the above steps, the ViT model can simultaneously extract the type and state, position coordinates and feature vector for tracking of the rotifer individual from each image, thereby providing a complete and structured data basis for subsequent individual trajectory construction and population dynamics analysis.

[0131] In a possible embodiment, each frame of the time-series image sequence is preprocessed to generate preprocessed image data, including:

[0132] Each frame of the time-series image sequence is subjected to size normalization processing to obtain images with a unified preset size; for the images with the unified preset size, a dense area with a rotifer distribution density exceeding a preset density threshold is identified and determined; the dense area is subjected to image segmentation to generate a plurality of sub-image blocks of a fixed size; the non-dense area in the image and the sub-image blocks obtained by segmenting the dense area jointly constitute the preprocessed image data.

[0133] In the specific implementation process, size normalization processing is performed on all images to scale or crop them to a preset size, and rotifer distribution density analysis is performed on the images that have completed size normalization. The images can be scanned based on region proposal of a lightweight convolutional network or based on target preliminary screening of background difference to identify areas in the images where rotifers may exist. The number of rotifers preliminarily detected in a unit area is counted and compared with a preset density threshold, so as to determine and mark a dense area in the image where the rotifer distribution density exceeds the threshold. The preset density threshold can be calibrated according to the characteristics of the actual sludge sample and the identification accuracy requirement.

[0134] Further, each dense area is divided into a plurality of sub-image blocks of a preset fixed size, for example, 100x100 pixels. Finally, for the non-dense area in the image that is not marked as a dense area, the whole or a block of a larger size is retained as an input unit. For each dense area, a plurality of sub-image blocks of a fixed size generated by segmentation of the dense area are used as input units. All these image units (including the blocks of the non-dense area and the sub-image blocks of the dense area) are jointly organized into an image block sequence as input of the feature extraction layer.

[0135] In the embodiments of the present application, size normalization processing can eliminate the image size difference caused by factors such as shooting position and focal length fine adjustment, and ensure that all input data remain consistent in spatial dimensions, laying a foundation for stable processing of subsequent models. By dividing the area with high target concentration into smaller processing units with relatively sparse targets, the probability of multiple rotifer individuals overlapping and shielding each other in the same processing unit is effectively reduced, thereby avoiding recognition errors or missed detection caused by feature confusion of the model.

[0136] S5, based on the feature vector, performing identity matching and tracking of rotifer individuals across time points to generate a trajectory database of all rotifers.

[0137] In the implementation process, based on the feature vector, an individual tracking algorithm is used to generate a trajectory database of all rotifers. The trajectory database includes the identity ID of all rotifers, the type state and position coordinates of the rotifer individual at different time points.

[0138] The individual tracking algorithm includes the following steps:

[0139] (1) Individual registration step: at the first image recognition moment (t=0.5h) of the monitoring period, a globally unique ID is assigned to each identified rotifer individual, and the feature vector, type state and position coordinates corresponding to this moment are stored as initial records in the trajectory database.

[0140] (2) Cross-frame matching step: at each subsequent image recognition moment (every 30 minutes) during the monitoring period, the similarity of the feature vector of each rotifer individual identified at the current moment to the most recent feature vector corresponding to all existing IDs in the trajectory database is calculated. If the highest similarity reaches or exceeds the preset matching threshold, it is determined that the current individual and the corresponding ID are the same rotifer individual, and the latest position coordinates, type state and feature vector of the ID in the trajectory database are updated.

[0141] (3) New individual marking step: if the feature vector of the rotifer individual identified at the current moment is lower than the preset matching threshold, a new globally unique ID is assigned to the individual, and its feature vector, type state, position coordinates and appearance time are stored as new records in the trajectory database.

[0142] (4) Disappeared individual determination step: for any ID recorded in the trajectory database, if the corresponding individual cannot be successfully matched in the cross-frame matching step at consecutive image recognition moments, it is determined that the rotifer individual corresponding to the ID has disappeared, and the last successful matching time is recorded as the disappearance time.

[0143] Please refer to Figure 5 , a part of the rotifer individual trajectory data provided by the embodiment of the present application. As shown in the figure, the trajectory data is presented in a structured table form, and the state records of three rotifer individuals with unique identity (ID001, ID008, ID007) at different time points within a 12-hour monitoring period are exemplarily shown. Each record contains at least the following fields:

[0144] (1) Time: records the specific moment of the state.

[0145] (2) Type state: describes the biological category and life state of the rotifer individual at the corresponding moment, such as "female (adult, alive)", "juvenile (alive)", "female (adult, with summer eggs)", etc.

[0146] (3) Coordinates: the spatial position of rotifer individuals in the image.

[0147] (4) Event records: key events in the life cycle of rotifer individuals, such as "egg-laying event (summer egg)", "hatching event", "development event (adult)", "death event", etc. Key events are identified by tracking the type state sequence corresponding to each ID.

[0148] S6, based on the trajectory database, calculate a plurality of characteristic parameters related to population reproduction, and input the plurality of characteristic parameters into a pre-trained water quality state classification model to output the water quality state of the target sludge sample.

[0149] The plurality of characteristic parameters include the following:

[0150] (1) Population number change rate: refers to the relative change degree of the total number of rotifer population in the preset monitoring period.

[0151]

[0152] Wherein, is the total number of rotifer individuals identified at the start time (t=0.5h) of the monitoring period, is the total number of rotifer individuals surviving at the end time (t=12.5h) of the monitoring period.

[0153] (2) Asexual reproduction efficiency: refers to the efficiency of rotifer offspring produced by parthenogenesis (summer egg production) in the monitoring period.

[0154]

[0155] Wherein, is the total number of larvae produced by summer egg hatching in the monitoring period, is the total number of adult females identified at the start time (t=0.5h) of the monitoring period.

[0156] (3) Sexual reproduction proportion: refers to the proportion of eggs produced by sexual reproduction (winter egg production) in the monitoring period.

[0157]

[0158] Wherein, is the total number of winter eggs identified in the monitoring period, is the total number of summer eggs identified in the monitoring period.

[0159] (4) Average reproduction cycle: refers to the average time interval between two consecutive egg-laying events of egg-laying females in the monitoring period (usually summer eggs).

[0160]

[0161] wherein, is the time interval between two consecutive ovipositions of the same type by a single female during the monitoring period, is the total number of females that oviposited at least once during the monitoring period.

[0162] (5) Juvenile survival rate: refers to the proportion of juveniles that successfully developed into adults from the eggs hatched during the monitoring period.

[0163]

[0164] wherein, is the total number of individuals that successfully developed into adults from the juveniles during the monitoring period, is the total number of juveniles produced by all hatching events during the monitoring period.

[0165] (6) Adult mortality rate: refers to the proportion of adult rotifer deaths during the monitoring period.

[0166]

[0167] wherein, is the total number of adults that died during the monitoring period, is the total number of adults identified at the start of the monitoring period (t=0.5h).

[0168] (7) Malformation rate: refers to the proportion of individuals with morphological deformities among the surviving rotifer individuals at the end of the monitoring period.

[0169]

[0170] wherein, is the total number of individuals with morphological deformities among the surviving individuals at the end of the monitoring period (t=12.5h).

[0171] Please refer to Figure 6 for the schematic diagram of all key events of rotifers provided by the embodiments of the present application. The following introduces how to calculate each characteristic parameter in combination with Figure 5 .

[0172] (1) Population number change rate:

[0173] From Figure 6 , it can be seen that there are 7 rotifers at t=0.5h: ID001, ID002, ID003, ID004, ID005, ID006, ID007, i.e. . The total number of rotifers surviving at t=12.5h is 10: ID001, ID002, ID004, ID006, ID008, ID009, ID011, ID012, ID013, ID014, i.e. .

[0174]

[0175] (2) Efficiency of asexual reproduction:

[0176] From Figure 6 it can be seen that the "summer egg hatching" event in the event corresponds to 6 juveniles: ID008 (t=2.5h), ID009 (t=2.5h), ID010 (t=3.0h), ID011 (t=5.5h), ID012 (t=5.5h), ID013 (t=7.0h), that is . There are 4 female insects in the event t=0.5h: ID001, ID002, ID003, ID004, that is .

[0177]

[0178] (3) Proportion of sexual reproduction:

[0179] From Figure 6 it can be seen that the "winter egg laying" event in the event corresponds to 1 rotifer: ID003 (t=6.0h), that is . The "summer egg laying" event in the event corresponds to 5 rotifers: ID001 (2 times) + ID002 (2 times) + ID003 (1 time), that is .

[0180]

[0181] (4) Average reproduction cycle:

[0182] From Figure 6 it can be seen that the "egg laying event" in the event corresponds to 3 female insects: ID001, ID002, ID003, that is . The interval between the two egg laying of a single female is as follows:

[0183] ID001: Second summer egg (t=5.5h) - first summer egg (t=2.0h) = 3.5h

[0184] ID002: Second summer egg (t=7.0h) - first summer egg (t=3.0h) = 4.0h

[0185] ID003: Only 1 summer egg + 1 winter egg (winter egg and summer egg are not calculated in the reproduction cycle) = 0h

[0186]

[0187] (5) Juvenile survival rate:

[0188] FromFigure 6 It can be seen that the "development event (juvenile -> adult)" in the event corresponds to a total of 4 rotifers: ID006 (t=3.0h), ID008 (t=6.5h), ID011 (t=9.5h), and ID012 (t=9.5h), that is, . The "hatching" in the event corresponds to a total of 9 juveniles: ID006 (initial juvenile), ID007 (initial juvenile), ID008, ID009, ID010, ID011, ID012, ID013, and ID014, that is, .

[0189]

[0190] (6) Adult mortality rate:

[0191] From Figure 6 it can be seen that the "adult death" in the event corresponds to a total of 2 rotifers: ID003 (t=9.0h, female) and ID005 (t=4.0h, male), that is, . The event t=0.5h corresponds to a total of 5 rotifers: ID001 (female), ID002 (female), ID003 (female), ID004 (female), and ID005 (male), that is, .

[0192]

[0193] (7) Malformation rate:

[0194] From Figure 6 it can be seen that the "malformation" in the event corresponds to a total of 2 rotifers: ID007 (juvenile) and ID012 (juvenile -> adult), that is, .

[0195]

[0196] After calculating multiple feature parameters, the numerical values of multiple feature parameters (such as population number change rate, asexual reproduction efficiency, sexual reproduction proportion, juvenile survival rate, malformation rate, etc.) are organized according to the required format of the model to form a feature vector to be discriminated. The feature vector is input into the water quality state classification model that has been trained, and the model performs operation and classification on the input features according to the decision boundary or discriminant function learned inside, and outputs the water quality state of the target sludge sample.

[0197] Further, according to the water quality state and multiple feature parameters related to population reproduction, a structured monitoring report (12h rotifer population monitoring report) can be generated, containing the following data:

[0198] (1) Index data: final values of population change rate, asexual reproduction efficiency, sexual reproduction proportion, larval survival rate, deformity rate, etc.

[0199] (2) Time series curve: shows the dynamic change trend of key indicators (such as total population, male-female ratio, egg type ratio) within 12 hours in the form of a chart.

[0200] (3) Water quality status: such as: high-quality status, good status, slight pollution status, serious pollution status, etc.

[0201] When the water quality status is slightly polluted or seriously polluted, a warning signal can be automatically generated, which is sent to the sewage treatment system. The sewage treatment system can start the emergency aeration equipment to quickly improve the dissolved oxygen in the aeration tank. It can also notify the operator through sound and light alarm or short message, prompting the operator to intervene in the investigation. It can also record the event log to provide data support for process optimization and accident tracing.

[0202] In one possible embodiment, the water quality status classification model is trained in the following way:

[0203] A historical monitoring data set containing multiple groups of samples is obtained; each group of samples contains actual values of multiple feature parameters identified and tracked based on historical time series images; for each feature parameter in the multiple feature parameters, a plurality of water quality status threshold ranges corresponding to the feature parameter are preset according to the response principle of rotifer ecology; the water quality status threshold range includes at least a normal range, a slight deviation range and an abnormal range; for each sample in the historical monitoring data set, the state is determined according to the threshold range in which the actual value of each feature parameter is located; based on the result of state determination and the predetermined comprehensive decision rule, a corresponding water quality status category label is automatically generated for each sample; the actual values of the multiple feature parameters are used as input features, and the corresponding water quality status category label is used as a supervision target to supervise the learning and training of the initial classification model, and a trained water quality status classification model is obtained.

[0204] In the specific implementation process, a historical monitoring data set is obtained. Each sample in the data set is derived from the continuous monitoring results of historical activated sludge samples for 12 hours, and the actual values of the multiple feature parameters contained therein are accurately calculated after processing the historical time series images by S5 and S6. These feature parameters at least include: population change rate, asexual reproduction efficiency, sexual reproduction proportion, larval survival rate and deformity rate.

[0205] For each of the above feature parameters, a set of numerical threshold ranges for state determination is preset according to the in-depth study of the environmental response of rotifer ecology. The threshold system defines three state intervals for each parameter, for example:

[0206] (1) Population change rate:

[0207] The normal threshold range is [20%, +∞); the slightly deviated threshold range is [10%, 20%); and the abnormal threshold range is [0, 10%].

[0208] (2) Asexual reproduction efficiency:

[0209] The normal threshold range is [100%, +∞); the slightly deviated threshold range is [90%, 100%); and the abnormal threshold range is [0, 90%].

[0210] (3) Sexual reproduction proportion:

[0211] The normal threshold range is [0, 10%); the slightly deviated threshold range is [10%, 20%); and the abnormal threshold range is [20, 100%].

[0212] (4) Larval survival rate:

[0213] The normal threshold range is [80%, 100%); the slightly deviated threshold range is [70%, 80%); and the abnormal threshold range is [0, 70%].

[0214] (5) Malformation rate:

[0215] The normal threshold range is [0, 3%); the slightly deviated threshold range is [3%, 8%); and the abnormal threshold range is [8%, 100%].

[0216] For each sample in the data set, the actual value of each feature parameter of the sample is compared with the preset threshold range to determine the state (normal, slightly deviated or abnormal) of each feature parameter. According to a predetermined comprehensive decision rule, the state determination results of multiple parameters are summarized to automatically assign a water quality state category label to the sample.

[0217] In a possible embodiment, the comprehensive decision rule is as follows:

[0218] If the actual values of multiple feature parameters in the sample are all within the corresponding normal threshold range, the water quality state label of the corresponding sample is marked as a high-quality state.

[0219] If the actual values of 1 to 2 feature parameters in the sample are within the corresponding slightly deviated threshold range, and the actual values of other feature parameters are within the corresponding normal threshold range, the water quality state label of the corresponding sample is marked as a good state.

[0220] If the actual values of 1 to 3 feature parameters in the sample are within the corresponding abnormal threshold range, the water quality state label of the corresponding sample is marked as a slightly polluted state.

[0221] If the actual values of 4 or more characteristic parameters in the sample are within the corresponding abnormal threshold range, the water quality state label of the corresponding sample is marked as a serious pollution state.

[0222] After completing the automatic labeling, a high-quality supervised learning data set is obtained. With the actual values of multiple characteristic parameters in this data set as input features (X) and the automatically generated water quality state category labels as supervised targets (Y), the selected initial machine learning classification model (such as random forest, gradient boosting tree, or support vector machine, etc.) is trained. Through optimization algorithm, the model learns the mapping relationship from multi-dimensional biological features to water quality categories, and finally obtains the trained water quality state classification model.

[0223] Traditional machine learning model training requires a large amount of manually labeled data, which is costly and inconsistent. The embodiment of the present application efficiently generates a large number of high-quality and standard-consistent training samples through an automatic and rule-based labeling method, breaking through the data labeling bottleneck. And the core knowledge (expressed as quantifiable threshold and executable rules) based on rotifer ecology response recognized in the field is converted into a fully automatic and batch data labeling engine, ensuring the consistency and scientificity of the label, and the model trained further internalizes expert knowledge and has the ability to handle complex nonlinear relationships, which is more intelligent and adaptable than directly applying fixed rules, providing a solid technical core for accurate and automated research and judgment of water quality state.

[0224] In summary, the embodiment of the present application provides a method for identifying the water quality state of activated sludge based on the reproduction of rotifer population, which has the following beneficial effects:

[0225] 1. The present application breaks through the limitations of traditional target detection models in fine-grained recognition. By using an optimized ViT model, it utilizes its powerful global attention mechanism to achieve accurate identification of rotifer sub-morphology (summer egg / winter egg, male / female, juvenile / adult, abnormal / normal individual) in activated sludge. At the same time, the high-dimensional feature vector output by the model lays a reliable data foundation for subsequent time series tracking, fundamentally solving the core problems of recognition confusion and detail loss caused by insufficient feature extraction capability of existing methods.

[0226] 2. Based on the feature vector generated by the ViT model, a trajectory database is generated by calculating the feature similarity. In a monitoring period of up to 12 hours, continuous tracking of the same rotifer individual is achieved, recording the entire life cycle of a single rotifer from hatching, development to reproduction / death, solving the problem of traditional models that cannot associate time series data and are prone to missing or mislabeling.

[0227] 3. Based on 12-hour individual trajectory data, a population change rate, asexual reproduction efficiency, sexual reproduction proportion, larval survival rate, and deformity rate are constructed, and complex biological behaviors are converted into quantifiable data. Further, by training a water quality state recognition classification model, an intelligent mapping relationship from multi-dimensional quantifiable indicators to water quality state is established, filling the gap in the prior art that only provides simple number statistics and cannot achieve deep water quality diagnosis and forward-looking warning.

[0228] Based on the same inventive concept, please refer to Figure 7 The application also provides an activated sludge water quality state recognition system based on rotifer population reproduction, which comprises:

[0229] A sludge sample collection module is used to collect a target sludge sample from an activated sludge aeration tank and place the target sludge sample in a culture device.

[0230] A sample constant-temperature culture module is used to adjust the simulated environmental parameters in the culture device to dynamically simulate the actual environmental conditions of the activated sludge aeration tank within a preset monitoring period.

[0231] A dynamic imaging module is used to control the microscopic imaging device to perform microscopic imaging on the target sludge sample in the culture device every first preset time interval within the preset monitoring period, and obtain a time sequence image sequence.

[0232] A rotifer recognition and tracking module is used to input the time sequence image sequence into a pre-trained visual Transformer model, output the type state, position coordinates, and feature vector of each rotifer individual in each image, and perform identity matching and tracking of the rotifer individuals across time points based on the feature vector to generate a trajectory database of all rotifers. The trajectory database includes the identity ID of all rotifers, the type state and position coordinates of each rotifer individual at different time points.

[0233] A population feature extraction module is used to calculate a plurality of feature parameters related to population reproduction based on the trajectory database.

[0234] A water quality state recognition module is used to input the plurality of feature parameters into a pre-trained water quality state classification model, and output the water quality state of the target sludge sample.

[0235] It should be noted that the modules in the activated sludge water quality state recognition system based on rotifer population reproduction in this embodiment correspond one by one to the steps in the activated sludge water quality state recognition method based on rotifer population reproduction in the foregoing embodiments, and therefore the specific embodiments of this embodiment can refer to the embodiments of the foregoing activated sludge water quality state recognition method based on rotifer population reproduction, which will not be described here again.

[0236] The active sludge water quality state identification system based on rotifer population propagation provided by the application has a wide application prospect in the following multiple fields due to its high precision, automation and intelligence.

[0237] 1. Intelligent operation and precise regulation and control of municipal sewage treatment plants

[0238] The rotifer state in the active sludge aeration tank of a sewage plant is an important indicator of water quality. When monitoring with the YOLO model in the past, it is difficult to accurately distinguish rotifer larvae / adults, summer eggs / winter eggs, and females / males, and it is also difficult to track the individual propagation process. When the application is applied, a camera is installed in the aeration tank and connected to the system, which can accurately identify rotifer reproduction conditions (such as summer egg hatching and larval survival), the number of male worms (an increase is a signal of water quality deterioration), and automatically generate 12-hour monitoring data and issue warnings (such as prompting "water quality deviation, adjust aeration intensity"), thereby reducing manpower and making water quality compliance more stable.

[0239] 2. Toxicity early warning and emergency response of industrial wastewater treatment stations

[0240] Industrial wastewater may contain toxic substances, and rotifers are very sensitive to such substances. Taking a wastewater treatment station of a chemical plant as an example, after applying the system, the rotifer state can be monitored in real time. If a large number of deformed rotifers or a sudden increase in rotifer mortality rate occurs, the system will quickly identify and prompt "toxic substances may have been mixed into the wastewater", and the staff can promptly investigate to avoid the discharge of substandard wastewater and prevent the wastewater from damaging the entire treatment system.

[0241] 3. On-site rapid detection and evaluation of environmental protection monitoring agencies

[0242] The environmental protection department needs to randomly check the sewage treatment situation in various places on a daily basis. In the past, samples had to be collected and taken back to the laboratory for analysis, which took a long time. Now, using the system, active sludge samples can be collected on site through a small camera and a notebook computer, and the reproduction of rotifers can also be directly analyzed to determine the water quality state level on the spot, greatly improving the detection efficiency.

[0243] 4. Water quality monitoring and precise feeding of the aquaculture industry

[0244] Rotifers in aquaculture ponds are natural bait for fish and shrimp, and their number and survival state can also reflect water quality. After adapting and adjusting the system, it can be installed in the breeding pond: on the one hand, it can monitor the water quality (such as a decrease in the number of rotifers, which may indicate that the water quality is too rich or too lean); on the other hand, it can prompt bait feeding, such as "the current number of rotifers is sufficient, and no additional bait needs to be fed", helping breeders reduce costs and reduce the pollution of bait residues on water quality.

[0245] In summary, the present application is not only suitable for the traditional municipal and sewage treatment technical field, but also can be extended to the fields of environmental monitoring, ecological research and aquaculture, etc. which need to make state judgment by microorganism indexes, by adapting the core technology of the present application, i.e. the ecosystem health degree evaluation based on fine-grained biological visual recognition, thus showing strong technical versatility and wide application potential.

[0246] Based on the same inventive concept, the present application further provides a computer device, which comprises a processor, a memory and a computer program stored in the memory, and the computer program realizes the aforementioned activated sludge water quality state identification method based on rotifer population breeding when being run by the processor.

[0247] Based on the same inventive concept, the present application further provides a computer storage medium, which stores a computer program, and the computer program realizes the aforementioned activated sludge water quality state identification method based on rotifer population breeding when being run by a processor.

[0248] In some embodiments, the computer readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disc or CD-ROM, etc.; and can also be various devices comprising one or any combination of the above memories. The computer can be various computing devices including smart terminals and servers.

[0249] In some embodiments, the executable instructions can be in the form of programs, software, software modules, scripts or codes, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and can be deployed in any form, including being deployed as independent programs or as modules, components, subroutines or other units suitable for use in a computing environment.

[0250] As an example, the executable instructions can but not necessarily correspond to files in a file system, can be stored in a part of a file storing other programs or data, for example, stored in one or more scripts in a hypertext markup language (HTML) document, stored in a single file dedicated to the program in question, or stored in multiple cooperative files (for example, files storing one or more modules, subroutines or code portions).

[0251] As an example, the executable instructions can be deployed to execute on one computing device, or on multiple computing devices located at one site, or on multiple computing devices distributed at multiple sites and interconnected through a communication network.

[0252] It should be noted that, in this document, the terms "comprising", "comprises" or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or system that includes the element.

[0253] The above-mentioned sequence numbers of embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0254] The above detailed description has further explained the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above is only a specific embodiment of the present application, and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for identifying the water quality status of activated sludge based on rotifer population reproduction, characterized in that, include: A target sludge sample is collected from the activated sludge aeration tank and placed in a culture device. Within a preset monitoring period, the simulated environmental parameters within the cultivation device are adjusted to dynamically simulate the actual environmental conditions of the activated sludge aeration tank. Within a preset monitoring period, every first preset time interval, the microscopic imaging device is controlled to perform microscopic imaging on the target sludge sample in the culture device to obtain a time-series image sequence. The time-series image sequence is input into a pre-trained visual Transformer model, which outputs the type, state, location coordinates, and feature vector of each rotifer individual in the image. Based on the feature vector, cross-time point identity matching and tracking are performed on individual rotifers to generate a trajectory database of all rotifers; the trajectory database includes the identity ID of all rotifers, the type status and location coordinates of each individual rotifer at different time points; Based on the trajectory database, multiple feature parameters related to population reproduction are calculated, and the multiple feature parameters are input into a pre-trained water quality status classification model to output the water quality status of the target sludge sample. The water quality status classification model is trained in the following way: Obtain a historical monitoring dataset containing multiple sets of samples; each set of samples contains the actual values ​​of the multiple feature parameters obtained based on historical time-series image sequence recognition and tracking. For each of the multiple characteristic parameters, multiple water quality state threshold ranges are preset according to the ecological response principle of rotifers; the water quality state threshold ranges include at least the normal range, the slightly deviated range, and the abnormal range; For each sample in the historical monitoring dataset, the status is determined based on the threshold range of the actual values ​​of its various characteristic parameters. Based on the results of the state determination and the predetermined comprehensive decision-making rules, a corresponding water quality state category label is automatically generated for each sample; Using the actual values ​​of the aforementioned multiple feature parameters as input features and the corresponding water quality state category labels as supervision targets, supervised learning training is performed on the initial classification model to obtain a trained water quality state classification model.

2. The method for identifying the water quality status of activated sludge based on rotifer population reproduction according to claim 1, characterized in that, The step of collecting target sludge samples from an activated sludge aeration tank and placing the target sludge samples in a culture device includes: Sludge samples were collected from multiple sampling points in the activated sludge aeration tank. Multiple sludge samples were mixed to obtain a homogenized sludge sample. A portion of the mixed sludge sample is taken and placed in a culture device as the target sludge sample.

3. The method for identifying the water quality status of activated sludge based on rotifer population reproduction according to claim 1, characterized in that, The culture apparatus includes: Petri dishes are used to hold the target sludge sample. A semiconductor temperature controller is used to regulate the temperature of the target sludge sample. A miniature air pump is used to introduce air into the bottom of the petri dish; A sterile filter membrane is disposed between the micro air pump and the petri dish to filter and sterilize the air introduced by the micro air pump. A temperature sensor is used to collect the water temperature monitoring value of the target sludge sample; A dissolved oxygen sensor is used to collect the dissolved oxygen concentration monitoring value of the target sludge sample; A magnetic stirrer is used to drive the petri dish to rotate at a preset speed.

4. The method for identifying the water quality status of activated sludge based on rotifer population reproduction according to claim 3, characterized in that, The step of adjusting the environmental parameters within the cultivation device within a preset monitoring period to dynamically simulate the actual environmental conditions of the activated sludge aeration tank includes: Within a preset monitoring period, at every second preset time interval, the water temperature monitoring value collected by the temperature sensor and the dissolved oxygen concentration monitoring value collected by the dissolved oxygen sensor are read synchronously. Simultaneously acquire the water temperature baseline value and dissolved oxygen concentration baseline value collected by the environmental sensors deployed in the activated sludge aeration tank; Calculate the first deviation between the monitored water temperature value and the reference water temperature value, and the second deviation between the monitored dissolved oxygen concentration value and the reference dissolved oxygen concentration value; If the absolute value of the first deviation is greater than the first preset deviation threshold, a first control signal is sent to the semiconductor temperature controller to drive the semiconductor temperature controller to adjust so that the water temperature monitoring value approaches the water temperature reference value. If the absolute value of the second deviation is greater than the second preset deviation threshold, a second control signal is sent to the micro air pump to drive the micro air pump to adjust so that the dissolved oxygen concentration monitoring value approaches the dissolved oxygen concentration reference value.

5. The method for identifying the water quality status of activated sludge based on rotifer population reproduction according to claim 1, characterized in that, The microscopic imaging device includes a motorized stage, a microscope objective, and an industrial camera; within a preset monitoring period, at first preset time intervals, the microscopic imaging device is controlled to perform microscopic imaging on the target sludge sample in the culture device to obtain a time-series image sequence, including: Within the preset monitoring period, every first preset time interval, the electric stage is controlled to move along a preset path and stop at multiple preset imaging positions in sequence; When the electric stage is stopped at each fixed imaging position, the industrial camera is controlled to continuously acquire multiple frames of images of the target sludge sample after being magnified by the microscope objective. All acquired images are organized according to acquisition time and location information to form a time-series image sequence.

6. The method for identifying the water quality status of activated sludge based on rotifer population reproduction according to claim 1, characterized in that, The visual Transformer model includes an input layer, a feature extraction layer, and an output layer; the step of inputting the temporal image sequence into the pre-trained visual Transformer model and outputting the type state, location coordinates, and feature vector of each rotifer individual in the image includes: Each frame of the time-series image is preprocessed to generate preprocessed image data; The preprocessed image data is received through the input layer; The preprocessed image data is encoded and features are extracted through the feature extraction layer to generate a feature vector for each image; The feature vectors of each image are input in parallel into the classification head, localization head, and feature embedding head of the output layer; The classification head processes the feature vector of each image and outputs the type and state of the rotifer individuals in the corresponding image. The positioning head processes the feature vector of each image and outputs the position coordinates of the rotifer in the corresponding image. The feature embedding head processes the feature vector of each image and outputs the feature vector of the corresponding rotifer individual in the image.

7. The method for identifying the water quality status of activated sludge based on rotifer population reproduction according to claim 6, characterized in that, The step of preprocessing each frame of the time-series image sequence to generate preprocessed image data includes: Each frame of the time-series image is normalized to obtain an image with a uniform preset size; For the image with a uniform preset size, identify and determine dense areas in the image where the rotifer distribution density exceeds a preset density threshold; The dense region is segmented to generate multiple fixed-size sub-image blocks; The non-dense regions in the image and the sub-image blocks obtained by segmenting the dense regions together constitute the preprocessed image data.

8. The method for identifying the water quality status of activated sludge based on rotifer population reproduction according to claim 1, characterized in that, The aforementioned multiple characteristic parameters include population size change rate, asexual reproduction efficiency, sexual reproduction ratio, average reproductive cycle, juvenile survival rate, adult mortality rate, and deformity rate.

9. The method for identifying the water quality status of activated sludge based on rotifer population reproduction according to claim 1, characterized in that, Based on the state determination results and predetermined comprehensive decision-making rules, a corresponding water quality state category label is automatically generated for each sample, including: If the actual values ​​of multiple feature parameters in a sample are all within the corresponding normal threshold range, then the water quality status label of the corresponding sample will be marked as excellent. If the actual values ​​of one or two feature parameters in a sample are slightly off the corresponding threshold range, and the actual values ​​of other feature parameters are all within the corresponding normal threshold range, then the water quality status label of the corresponding sample will be marked as good. If the actual values ​​of 1 to 3 feature parameters in a sample are within the corresponding abnormal threshold range, the water quality status label of the corresponding sample will be marked as slightly polluted. If the actual values ​​of four or more feature parameters in a sample are within the corresponding abnormal threshold range, the water quality status label of the corresponding sample will be marked as severely polluted.

Images