A biochemical pool aeration state visual identification method based on water surface turbulent form

Abstract

The present application relates to sewage treatment intelligent monitoring technical field, specifically disclose a kind of biochemical pond aeration state visual identification method based on water surface turmoil form, comprising: by the optical flow spatial consistency detection to water surface to judge whether there is push flow-turbulence motion coupling interference, and generate initial turbulence activity heat map;Water surface mixed motion signal is separated in motion band in time and space two dimensions, and turbulence component is decoupled with surge component, calculates global push flow interference intensity coefficient;Surge submergence degree evaluation is carried out, and pixel-level confidence weight is calculated;Turbulence signal is fused with amplitude correction in pixel-level weighting, reconstructs turbulence intensity field and turbulence activity heat map, and carries out semantic segmentation, outputs turbulence coverage, turbulence intensity index and aeration uniformity index, solve the problem that aeration turbulence visual signal is covered under push flow surge interference, realize the accurate, visual identification of biochemical pond aeration state.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for wastewater treatment, specifically to a visual recognition method for the aeration status of a biological treatment tank based on the turbulence pattern of the water surface. Background Technology

[0002] The aeration system in the aerobic zone of the activated sludge process provides dissolved oxygen and maintains sludge suspension; its operational status directly affects effluent quality and energy consumption. Current aeration status monitoring mainly relies on dissolved oxygen electrodes (single-point measurement, unable to reflect spatial distribution), gas flow meters (only total volume, not sensing equipment operating conditions), and manual inspection (subjective and unquantifiable), which is insufficient to meet the needs of refined control. In recent years, machine vision-based water surface status monitoring has attracted attention due to its intuitive and non-contact nature; however, existing technologies have the following significant shortcomings: First, the amplitude of the directional swell motion generated by the flow generator in the biological treatment tank is much greater than the local turbulence motion formed by the bursting of aeration bubbles. The water surface motion signal exhibits a superposition characteristic where strong swell motion masks weak turbulence. Conventional motion detection algorithms (such as inter-frame difference and optical flow) directly extract motion vectors that mainly reflect the flow swell motion, failing to effectively distinguish between the two motion components with different physical sources, leading to severe distortion in visual monitoring results. Second, some methods attempt to use temporal filtering for signal separation, but ignore the essential difference in spatial scale between the flow swell motion and aeration turbulence: the swell motion exhibits a large-area, continuous, low-frequency characteristic, while the turbulence exhibits a local, discrete, high-frequency characteristic. One-dimensional processing based solely on the time dimension cannot fully decouple the two modes, and swell interference still remains in the separated turbulence component. Third, the swell interference is unevenly distributed throughout the tank, and the degree of submersion varies in different areas due to the influence of the flow generator location, tank shape, and aerator layout. Existing methods generally adopt a global unified processing strategy, lacking pixel-level reliability assessment and differential correction, resulting in low positioning accuracy of active turbulence areas and unreliable evaluation of aeration uniformity; fourth, existing works mostly output only a single aeration intensity estimate or simple segmentation results, lacking a quantitative index system that comprehensively characterizes the aeration state from multiple dimensions such as coverage, intensity, and uniformity, making it difficult to meet the actual needs of process control for multi-dimensional decision-making information.

[0003] Therefore, the present invention provides a visual recognition method for the aeration status of a biological treatment tank based on the turbulence pattern of the water surface. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies and solve at least one of the technical problems mentioned in the background, this invention provides a visual recognition method for the aeration status of a biological tank based on the turbulence morphology of the water surface. Addressing the problem of visual monitoring distortion caused by the obscuring of aeration turbulence by surging waves, this invention decouples turbulence from surging waves through spatiotemporal dual-domain frequency band separation. Then, pixel-level confidence weighting and full-field correction are used to eliminate residual interference, ultimately outputting three indicators: turbulence coverage, intensity, and uniformity, achieving accurate aeration status recognition under strong interference.

[0005] The objective of this invention can be achieved through the following technical solution: a visual recognition method for the aeration state of a biological treatment tank based on the turbulence pattern of water surface, comprising the following steps: By performing optical flow spatial consistency detection on the surface of the biological treatment tank, it is determined whether there is coupling interference between the push flow and the tumbling motion. If so, the energy of the water surface motion is accumulated and normalized over time to generate an initial tumbling activity heatmap. The motion frequency band of the mixed motion signal on the surface of the biological treatment tank is separated in both time and space, the mixed motion signal is decoupled into tumbling component and surging component, and the global thrust interference intensity coefficient is calculated. The flood inundation level of the biological treatment pool is assessed pixel by pixel, and the pixel-level confidence weight is calculated. The tumbling signal on the surface of the biological treatment tank is weighted and fused at the pixel level according to the credibility weighting system of the tumbling signal, and the amplitude is corrected in the surface area of ​​the biological treatment tank by combining the global thrust interference intensity coefficient. The corrected tumbling intensity field and tumbling activity heat map are output. The corrected tumbling activity field is semantically segmented to output tumbling coverage, tumbling intensity index and aeration uniformity index.

[0006] Furthermore, the method for determining whether there is thrust-tumble motion coupling interference is as follows: Continuously acquire video images of the surface of the biological treatment pool, and preprocess each frame of the acquired raw image sequentially; For two adjacent frames of images after preprocessing, the motion vector field of each pixel is calculated. Each pixel outputs a two-dimensional motion vector containing horizontal and vertical displacement components, and the optical flow amplitude field and optical flow direction field are calculated. Within the water surface area of ​​the biological treatment pool, the spatial variation coefficient of the optical flow amplitude field is calculated, and the spatial standard deviation of the optical flow direction field is obtained by using the circular statistical method. Maintain a time-series data queue to store the spatial variation coefficient of optical flow amplitude and the spatial standard deviation of optical flow direction for each frame of image within the past T seconds in chronological order. Calculate the spatial variation coefficient of optical flow amplitude and the spatial standard deviation of optical flow direction in the current frame, and compare them with their historical reference values ​​within the time-series sliding window. When both are lower than 70% of their respective reference values, it is determined that there is a push-flow-tumble motion coupling interference.

[0007] Furthermore, the process of generating the initial tumbling activity heatmap is as follows: For each pixel in the water surface area of ​​the biological treatment pool, the optical flow calculation result corresponding to the most recent L frames is taken. The optical flow calculation result includes the motion displacement amplitude, horizontal displacement component and vertical displacement component. The motion displacement amplitude, horizontal displacement component and vertical displacement component are arranged in frame order to form three time series of length L. For each pixel, maintain multiple Gaussian distribution models, output binary judgments, and combine the binary judgment results of all pixels into a binary image as the foreground dynamic mask of the current frame. The motion displacement amplitude of each pixel in the time series and the dynamic mask of the foreground are accumulated in time sequence and normalized to obtain the normalized optical flow accumulation map and the foreground frequency matrix, respectively. Then, the initial tumbling activity heat map is obtained by pixel-by-pixel weighted fusion.

[0008] Furthermore, the process of separating motion frequency bands in both time and space dimensions is as follows: A time-domain high-pass filter is performed on the motion time sequence of each pixel to separate the high-frequency tumbling component and the low-frequency surge component according to the motion frequency band. The motion vector field is decomposed into two-dimensional frequency domain in the spatial domain, and the low-frequency spatial motion component and the high-frequency spatial motion component are separated by radial bandpass filtering. By combining the low-frequency spatial components in the horizontal and vertical directions with the high-frequency spatial components, a complete low-frequency spatial motion vector field and a high-frequency spatial motion vector field are obtained.

[0009] Furthermore, the global push-stream interference intensity coefficient is calculated as follows: The total energy of the low-frequency spatial component and the total energy of the high-frequency spatial component in the entire biochemical pool water surface area are calculated separately. The total energy is calculated by summing the squared values ​​of all pixels in each component matrix. Dividing the total low-frequency energy by the sum of the total low-frequency energy and the total high-frequency energy yields the global thrust interference intensity coefficient.

[0010] Furthermore, the method for evaluating the surge inundation level of the biochemical pool water surface pixel by pixel is as follows: The high-frequency component sequence and low-frequency component sequence of each pixel in the water surface area of ​​the biological treatment pool are respectively squared energy accumulation to obtain the total energy of high-frequency tumbling and the total energy of low-frequency surging. Surge submersion level = Total energy of low-frequency surges / (Total energy of high-frequency tumbling + Total energy of low-frequency surges). If the submersion level is greater than the threshold for determining the dominant state of the surge, then the pixel is determined to be in the dominant state of the surge; otherwise, the pixel is determined to be in the non-dominant state of the surge.

[0011] Furthermore, the pixel-level confidence weights are calculated as follows: A monotonically decreasing function in negative exponential form is constructed with the inundation degree of the surge as the independent variable as the confidence weight function. For each pixel in the water surface area of ​​the biological treatment pool, the confidence weight of the pixel is obtained by substituting the difference between the inundation degree of the surge and the inundation tolerance threshold into the confidence weight function. After calculating the confidence weights for all pixels, the confidence weights of each pixel are arranged according to their spatial location to generate a pixel-level confidence weight map with the same size as the image.

[0012] Furthermore, the reconstruction method of the tumbling intensity field is as follows: For any pixel within the surface area of ​​the biological treatment pool, the high-frequency horizontal component and the high-frequency vertical component at the current moment are squared and then added together. The square root of the sum is then taken to obtain the instantaneous tumbling intensity value at the current moment. Valid pixels that meet the confidence weight requirements within the surface area of ​​the biological treatment pool are selected, and the proportion of valid pixels to the total number of pixels within the surface area of ​​the biological treatment pool is calculated to obtain the effective pixel ratio. If the effective pixel ratio is greater than or equal to the preset effective ratio lower limit threshold, then the tumbling intensity assessment at the current moment is determined to have sufficient statistical confidence, and weighted aggregation is performed: The corrected tumbling intensity index is obtained by weighting the tumbling intensity values ​​of all valid pixels with confidence weights.

[0013] Furthermore, the output process of the corrected tumbling activity heatmap is as follows: Based on the high-frequency spatial motion vector field at the current moment, calculate the instantaneous tumbling amplitude of each pixel point in the water surface area of ​​the biological pool based on the high-frequency spatial components, and obtain the high-frequency spatial component amplitude map. The instantaneous amplitude of the high-frequency spatial components of each pixel in the most recent time window queue is accumulated and normalized in time to obtain a heat map of tumbling activity based on the reconstruction of high-frequency spatial components. A correction function is constructed, and the global thrust interference intensity coefficient is substituted into the correction function to calculate the correction coefficient at the current time. The normalized tumbling activity heatmap is multiplied by the correction coefficient to obtain the corrected tumbling activity heatmap, which is denoted as the corrected tumbling activity field.

[0014] Furthermore, the output methods for the tumbling coverage rate, tumbling intensity index, and aeration uniformity index are as follows: The corrected tumbling activity field is used as the main input channel of the semantic segmentation model, and two auxiliary channels, the grayscale brightness map of the current frame and the foreground dynamic mask, are added to form a three-channel input tensor. Load a pre-trained lightweight U-shaped network semantic segmentation model, feed the three-channel input tensor into the model to perform forward inference calculation, and output a single-channel probability map with the same spatial size as the input image; Perform binarization on the tumbling probability map to obtain a preliminary tumbling region binary mask. Then, perform opening and closing operations in sequence, extract all connected components and remove connected components with an area smaller than the minimum area threshold, and output the final tumbling region binary mask. The tumbling coverage rate is obtained by dividing the total number of pixels with a median value of 1 in the final binary mask of the tumbling area by the total number of pixels in the surface area of ​​the biological pool. The tumble intensity index is directly adopted from the corrected tumble intensity index calculated by spatial weighted aggregation. The surface area of ​​the biological treatment tank is divided into multiple grids. For each grid, the spatial variation coefficient of the energy proportion of all grids is calculated, and the aeration uniformity index is the complement of the variation coefficient.

[0015] The beneficial effects of this invention are as follows: High-pass filtering is performed in the time domain by utilizing the frequency band difference between the surging current and the aeration tumbling, while radial frequency band decomposition is performed in the spatial domain by utilizing the spatial scale difference between the two. This achieves effective decoupling of the mixed motion signals from both temporal and spatial dimensions, improves the purity of the tumbling component extraction, and can still accurately obtain the aeration activity signal under strong surging current interference. By performing time-series statistical modeling of the spatial variation coefficient of optical flow amplitude and the spatial standard deviation of direction, an automatic discrimination mechanism for push-flow-tumble coupling interference is constructed. When both indicators are lower than the historical benchmark, the strong interference condition is accurately identified, providing a basis for subsequent processing path switching. A pixel-level surge submersion assessment and negative exponential confidence weight function are constructed to achieve spatial differential weighted fusion of tumbling signals. The weights are continuous and smooth to avoid inter-frame oscillations. This function can stably suppress the contribution of polluted pixels after the surge submersion exceeds the standard, so that the output heat map and intensity index can more realistically reflect the aeration status. The cumulative optical flow value (reflecting the intensity of motion) and the foreground frequency (reflecting the continuity of motion) are weighted and fused together, and the two types of features constrain each other to improve the signal-to-noise ratio of the heatmap. At the same time, a global push flow interference intensity coefficient is introduced to perform full-field amplitude correction, which cancels the residual surge leakage in the filter transition band and ensures the stability of the semantic segmentation input. The system ultimately outputs three quantifiable business indicators: tumbling coverage rate, tumbling intensity index, and aeration uniformity index. These indicators comprehensively characterize the operating status of the aeration system from three dimensions: spatial coverage, overall intensity, and distribution uniformity. This provides an intuitive and complete basis for refined aeration control, helping to maximize energy conservation and consumption reduction while ensuring effluent quality. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a flowchart of a visual recognition method for aeration status in a biological treatment tank based on water surface turbulence morphology, according to the present invention. Figure 2 This is a logic judgment diagram of a visual recognition method for aeration status of a biological tank based on water surface turbulence morphology in this invention. Figure 3 These are the time-domain power spectra of water surface motion signals collected under the pure flow propulsion condition and the pure aeration condition in this invention. Figure 4 This is the power spectrum of the motion vector field after radial averaging in the spatial frequency domain in this invention; Figure 5 This is the relationship between pixel-level confidence weight and surge inundation level in this invention. Detailed Implementation

[0018] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0019] Example: Please refer to Figure 1 As shown, the visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to the present invention specifically includes the following steps: Step 1: By performing optical flow spatial consistency detection on the surface of the biological treatment tank, determine whether there is coupling interference between the push flow and the tumbling motion. If so, perform time-series accumulation and normalization of the surface motion energy to generate an initial tumbling activity heatmap. In step one, the process of determining whether there is coupling interference between the flow and tumbling motion in the biochemical pool includes: Industrial cameras deployed above the aerobic zone of the biological treatment pond continuously acquire video images of the water surface. Each frame of the acquired raw image undergoes preprocessing, including distortion correction, extraction of the water surface area of ​​the biological treatment pond, and image stabilization. Specifically: Distortion correction utilizes the intrinsic parameter matrix and distortion coefficients obtained during the camera calibration stage to eliminate radial and tangential distortion of the lens, allowing the pixel position to return to its true physical coordinates. Extract the water surface area of ​​the biological treatment pool, load a pre-calibrated binary mask file (1 for water surface area and 0 for non-water surface area), multiply the calibrated image with the mask pixel by pixel, and retain only the water surface area data; Image stabilization processing involves ORB feature point detection and matching between adjacent frames, calculating the global translational offset caused by camera micro-vibration, and correcting the current frame to the same spatial reference as the previous frame through affine transformation. For two adjacent frames of images after preprocessing (the previous frame and the current frame), a dense optical flow algorithm based on polynomial expansion is used to calculate the motion vector field of each pixel. Each pixel outputs a two-dimensional motion vector containing horizontal and vertical displacement components, with the unit being pixels / frame. Based on the motion vector field of each pixel, two derivative physical quantities, the optical flow amplitude field and the optical flow direction field, are further calculated, including: The optical flow amplitude field is obtained by taking the square root of the sum of the squares of the horizontal and vertical displacement components of each pixel, which reflects the intensity of the pixel's motion. The optical flow direction field calculates the arctangent of the horizontal and vertical displacement components of each pixel to obtain the angle value of the pixel's motion direction, which ranges from 0 degrees to 360 degrees, reflecting the direction of the pixel's motion. Within the surface area of ​​the biological treatment pond, the spatial variation coefficient of optical flow amplitude of all pixels is calculated to obtain the spatial variation coefficient of optical flow amplitude field, which reflects the degree of dispersion of motion intensity in spatial distribution. Aeration turbulence is in the form of discrete patches with a large variation coefficient, while push flow and surging waves are in the form of large continuous areas with a small variation coefficient. Within the surface area of ​​the biological treatment pool, the optical flow direction values ​​of all pixels are statistically calculated: Specifically, considering the cyclical nature of orientation angles (0 degrees and 360 degrees represent the same direction), the spatial standard deviation of the orientation field needs to be calculated using a circular statistical method. First, the orientation angles of all pixels are converted into vectors on a unit circle. The average direction of the composite vector is calculated as the orientation mean. Then, the angle difference between the orientation of each pixel and the orientation mean is calculated. The minimum value of the angle difference within the range of -180 degrees to 180 degrees is taken. Finally, the square root of the arithmetic mean of the squares of all angle differences is calculated to obtain the spatial standard deviation of the orientation field. It is understandable that the physical meaning of the spatial standard deviation of the optical flow direction field is: it reflects the spatial consistency of the direction of motion. The direction of the surging and pushing waves is highly consistent, and the standard deviation is small. The direction of the aeration and tumbling is randomly divergent, and the standard deviation is large. Maintain a time-series data queue. Each time-series data queue stores the optical flow amplitude field spatial variation coefficient and optical flow direction field spatial standard deviation calculated for each frame of image within the past T seconds in chronological order. The value of T is set according to the process response speed, for example, 60 seconds. The time-series data queue adopts a first-in-first-out update mechanism: after each new image frame is processed, the newly calculated coefficient of variation and standard deviation of orientation are appended to the tail of the queue, while the earliest data point at the head of the queue is deleted, so that the queue always maintains the amount of data from the most recent T seconds. During the processing of each frame, the following two calculations are performed on the sliding window queue at the current moment: The arithmetic mean of the spatial variation coefficients of all optical flow amplitude fields within the calculation window is used as the baseline value of the variation coefficient at the current moment, reflecting the average level of the spatial dispersion of water surface motion over the past T seconds. The arithmetic mean of the spatial standard deviations of all optical flow directions within the calculation window is used as the reference value of the direction standard deviation at the current moment, reflecting the average level of the consistency of water surface motion direction over the past T seconds; The baseline values ​​for the coefficient of variation and the directional standard deviation are continuously updated as the window is slid. The spatial variation coefficient of the optical flow amplitude field calculated in the current frame is compared with the reference value of the variation coefficient at the current moment. At the same time, the spatial standard deviation of the optical flow direction field calculated in the current frame is compared with the reference value of the direction standard deviation at the current moment. When the spatial coefficient of variation of the optical flow amplitude field is less than 70% of the reference value of the coefficient of variation, and the spatial standard deviation of the optical flow direction field is less than 70% of the reference value of the standard deviation, it is determined that there is a coupling interference of push-flow and tumbling motion on the surface of the biological pool at the current moment, that is, the push-flow surge has completely suppressed the micro-motion signal of aeration and tumbling. It should be noted that the physical meaning of determining whether there is coupling interference between the push flow and tumbling motion on the surface of the biological treatment tank at the current moment is: the water surface movement is abnormally uniform in amplitude and highly consistent in direction, indicating that the push flow and surging waves have completely suppressed the micro-motion signal of aeration and tumbling. In step one, the process of generating the initial tumbling activity heatmap includes: The first point to clarify is that establishing a pixel-level motion displacement time series involves: For each pixel within the surface area of ​​the biological treatment pool, starting from the current moment, backtracking is performed to obtain the optical flow calculation results corresponding to the most recent L frames. The optical flow calculation results include the motion displacement amplitude, horizontal displacement component, and vertical displacement component. For each frame, the motion displacement amplitude (i.e., the magnitude of the optical flow vector, reflecting the motion velocity of the pixel in the current frame), horizontal displacement component (projection of the optical flow vector on the horizontal axis), and vertical displacement component (projection of the optical flow vector on the vertical axis) of the pixel are recorded. The motion displacement amplitude, horizontal displacement component, and vertical displacement component are arranged in frame order to form three time series of length L: motion displacement amplitude time series, horizontal component time series, and vertical component time series. The value of L corresponds to the length of the preset time window. For example, if the preset time window is 10 seconds, L is equal to 250 frames. The maintenance of the time series adopts a sliding window mechanism. After processing each frame of image, the calculated motion displacement amplitude, horizontal displacement component and vertical displacement component are appended to the end of each time series. At the same time, the earliest data point at the beginning of each time series is deleted, so that the time series always maintains the amount of data from the most recent L frames. Secondly, it should be noted that the parallel background modeling and extraction of the foreground dynamic mask are performed as follows: For each pixel, multiple Gaussian distribution models are maintained. Each Gaussian distribution model contains three parameters: mean, variance, and weight. When a new frame of image arrives, the current brightness value of the pixel is matched with each Gaussian distribution model one by one. If the match is successful, the parameters of the corresponding Gaussian distribution model are updated. If the match fails, the old Gaussian distribution model with the smallest weight is replaced with the new Gaussian distribution model. The algorithm outputs a binary judgment: if the brightness change of a pixel cannot be reasonably explained by any background model, it is judged as a moving foreground pixel with a mask value of 1; otherwise, it is judged as a background pixel with a mask value of 0. The binary judgment results of all pixels are combined into a binary image with the same size as the image, which is the foreground dynamic mask of the current frame. In the foreground dynamic mask, the foreground pixels (value 1) represent the area where the water surface has significant movement, and the background pixels (value 0) represent the area where the water surface is relatively still or has only slight disturbance. Thirdly, it should be noted that the temporal accumulation and normalization of optical flow amplitude are as follows: For any pixel, take all L motion displacement amplitudes in the pixel's time series, sort them in ascending order of motion displacement amplitude, and take the motion displacement amplitude corresponding to the 95th percentile as the pixel's cumulative motion amplitude at the current moment. After completing the time series accumulation operation for all pixels, a cumulative amplitude matrix with the same size as the image is obtained. The cumulative amplitude matrix is ​​normalized within the surface area of ​​the biological treatment pool. The specific method is as follows: find the maximum value of the cumulative amplitude of all pixels within the surface area of ​​the biological treatment pool, divide the cumulative amplitude of each pixel by the maximum value, so that the values ​​of all pixels are compressed into a closed interval between 0 and 1. The normalized matrix is ​​denoted as the normalized optical flow cumulative map. Fourthly, it should be noted that the temporal accumulation and normalization of the foreground mask are as follows: Based on the frame-by-frame dynamic foreground mask of the output, for each pixel, the number of times the foreground mask value of the pixel is 1 in all L frames within the time window is counted and divided by L to obtain the foreground occurrence frequency of the pixel within the time window, resulting in a foreground frequency matrix with the same size as the image, located in the interval from 0 to 1. Fifthly, it should be noted that the initial turbulence activity heatmap is generated by dual-channel weighted fusion, specifically as follows: The normalized optical flow accumulation map and the foreground frequency matrix are fused pixel by pixel. For each pixel in the water surface area of ​​the biological treatment pool, the normalized optical flow accumulation value and the foreground frequency value are added to the fusion. The weight of the normalized optical flow accumulation value is the first preset weight, and the weight of the foreground frequency value is the second preset weight. The initial tumbling activity value of the pixel is obtained. The sum of the first preset weight and the second preset weight is equal to 1. For example, the first preset weight is 0.6 and the second preset weight is 0.4. It should be noted that the physical basis for the weighting ratio is as follows: the cumulative optical flow value reflects the intensity of the motion and is more sensitive to the strength of the tumbling, so it is given a larger weight; the foreground frequency reflects the temporal continuity of the motion and has auxiliary value for confirming continuous tumbling, so it is given a smaller weight. The two types of signals complement each other. The optical flow may generate high values ​​due to instantaneous strong disturbances, which need to be constrained by the foreground frequency. The foreground mask may miss weak motions due to imperfections in the background model, which needs to be compensated for by the accumulation of optical flow. After weighted fusion of all pixels, a grayscale image with the same size as the image is obtained, which is the initial heat map of turbulence activity. The grayscale value of each pixel represents the level of turbulence activity of the water surface at that location. The closer the value is to 1, the stronger the turbulence, and the closer the value is to 0, the calmer the water surface. It should be noted that the purpose of determining whether coupling interference exists is to determine whether a high-precision processing path or a simplified processing path should be used subsequently, and to provide basic data for subsequent steps. Step 2: Separate the motion frequency bands of the mixed motion signal on the surface of the biological treatment tank in both time and space, decouple the mixed motion signal into tumbling components and surging components, and calculate the global thrust interference intensity coefficient; In step two, the process of separating the motion frequency bands of the mixed motion signal on the surface of the biological treatment tank in both time and space dimensions includes: The first point to clarify is that time-domain high-pass filtering is specifically as follows: An infinite impulse response high-pass filter is applied to the time series of motion displacement amplitudes of each pixel. The process of determining the filter cutoff frequency includes: When the biological treatment tank is in pure flow propulsion mode (the aeration blower is turned off and only the propulsion device is turned on), the water surface video is continuously collected for no less than 20 minutes. For no less than 500 pixels randomly selected from the water surface area of ​​the biological treatment tank, the power spectral density of the motion displacement amplitude time series is calculated. The power spectrum of all pixels is averaged to obtain the typical power spectral distribution curve of the propulsion surge. The energy concentration frequency band of the curve is the characteristic frequency band of the propulsion surge. When the biological tank is in pure aeration mode (the propeller is turned off and only the aeration blower is turned on), repeat the above operation. The energy concentration frequency band of the typical power spectrum distribution curve of aeration turbulence is the characteristic frequency band of aeration turbulence. The power spectrum curves of the surging and aeration are plotted on the same frequency axis, and the center frequency value of the overlapping frequency band of the two power spectrum curves is taken as the cutoff frequency of the time-domain high-pass filter. After the infinite impulse response high-pass filtering, the amplitude sequence of each pixel is separated into a high-frequency tumbling component above the cutoff frequency and a low-frequency surge component below the cutoff frequency. The same filtering operation is performed on the horizontal component sequence and the vertical component sequence to obtain the high-frequency horizontal component, high-frequency vertical component, low-frequency horizontal component and low-frequency vertical component of each pixel. For example, such as Figure 3 As shown, Figure 3 The time-domain power spectra of water surface motion signals collected under pure thrust flow and pure aeration conditions are shown. The energy of thrust flow swell is concentrated in the low-frequency range (approximately 0.1-1.5Hz), while the energy of aeration tumbling is concentrated in the high-frequency range (approximately 2-8Hz). The intersection of the two power spectrum curves is the cutoff frequency of the time-domain high-pass filter. Components above this frequency are retained as tumbling components, while components below this frequency are classified as swell components, thus achieving preliminary separation of the two in the time domain. Secondly, it should be noted that the spatial frequency band decomposition is as follows: Take the horizontal and vertical component matrices of the motion vector field at the current moment, and perform two-dimensional discrete Fourier transforms on them respectively to transform the spatial domain data to the spatial frequency domain. In the process of designing radial bandpass filter banks in the frequency domain and determining the boundary frequency of spatial frequency band decomposition, specifically, under pure flow and pure aeration conditions, two-dimensional power spectrum analysis is performed on the horizontal and vertical components of the motion vector field of a large number of frames, respectively. The two-dimensional power spectra under pure flow and pure aeration conditions are averaged radially to obtain one-dimensional radial power spectrum curves. The intersection frequency of the two curves or the center frequency of the overlapping frequency band is taken as the boundary frequency of spatial frequency band decomposition. Multiply the frequency domain complex matrix by the transfer function matrices of the low-frequency bandpass filter and the high-frequency bandpass filter respectively, and then perform two-dimensional inverse Fourier transform to obtain the low-frequency spatial components and high-frequency spatial components in the horizontal and vertical directions respectively. Combine the low-frequency spatial components and high-frequency spatial components in the horizontal and vertical directions to obtain the complete low-frequency spatial motion vector field and high-frequency spatial motion vector field. For example, such as Figure 4 As shown, Figure 4 The power spectrum of the motion vector field after radial averaging in the spatial frequency domain is given. The surging waves exhibit large-area continuous motion with a low spatial frequency, while the aeration tumbling exhibits local discrete patches with a higher spatial frequency. The intersection of the two spectral lines is defined as the boundary frequency of spatial frequency band decomposition. Based on this, a radial bandpass filter is designed to decompose each frame of the motion vector field into a low-frequency spatial component (surging waves) and a high-frequency spatial component (tumbling), thereby achieving mode decoupling in the spatial domain. In step two, the calculation process of the global push flow interference intensity coefficient includes: The total energy of the low-frequency spatial component and the total energy of the high-frequency spatial component in the entire biochemical pool water surface area are calculated separately. The total energy is calculated by summing the squared values ​​of all pixels in each component matrix. Divide the total low-frequency energy by the sum of the total low-frequency energy and the total high-frequency energy to obtain the global thrusting interference intensity coefficient. The closer it is to 1, the higher the proportion of thrusting and surging energy in the global motion signal at the current moment. The closer it is to 0, the higher the proportion of aeration and tumbling energy. It should be noted that the function of signal decoupling is to simultaneously perform frequency band cutting on the mixed motion signal in both time and space dimensions, separating the pure tumbling component and swell component. Step 3: Assess the flooding level of the biological treatment pool pixel by pixel and calculate the pixel-level confidence weight; In step three, the process of evaluating the surge inundation level of the biological treatment pool surface pixel by pixel includes: For each pixel in the surface area of ​​the biological treatment pool, the horizontal high-frequency component sequence is squared and summed within the time window to obtain the horizontal high-frequency energy. The vertical high-frequency component sequence is squared and summed to obtain the vertical high-frequency energy. The horizontal high-frequency energy and the vertical high-frequency energy are added together to obtain the total high-frequency tumbling energy of the pixel. For each pixel in the surface area of ​​the biological treatment pool, the horizontal low-frequency component sequence is squared and summed within the time window to obtain the horizontal low-frequency energy. The vertical low-frequency component sequence is squared and summed to obtain the vertical low-frequency energy. The horizontal low-frequency energy and the vertical low-frequency energy are added together to obtain the total low-frequency surge energy of the pixel. For any pixel, the surge inundation level = total energy of low-frequency surges / (total energy of high-frequency tumbling + total energy of low-frequency surges); It is understandable that the physical meaning of the swell submersion degree is: the value ranges from 0 to 1. When the movement of a pixel is mainly contributed by the swell, the low-frequency swell energy is much greater than the high-frequency turbulence energy, and the submersion degree approaches 1. When the movement of a pixel is mainly contributed by the turbulence of aeration, the high-frequency turbulence energy is much greater than the low-frequency swell energy, and the submersion degree approaches 0. When the two types of motion energy are equal, the submersion degree is close to 0.5. The submersion of the surge is compared with the surge dominance state determination threshold. If the submersion degree is greater than the surge dominance state determination threshold, the pixel is determined to be in the surge dominance state. Otherwise, the pixel is determined to be in the non-surge dominance state. The process for determining the threshold for the dominant state of surging is as follows: When the biological tank is in pure aeration mode (with the flow generator off), the statistical distribution of the submersion degree of all pixels in the entire tank is calculated, and the 95th percentile is taken as the upper limit of the submersion degree under the tumbling mode. When the biological tank is in pure flow mode (with aeration off), the statistical distribution of the submersion degree of all pixels in the entire tank is calculated, and the 5th percentile is taken as the lower limit of the submersion degree under the surging mode. The midpoint between the upper limit of the submersion degree and the lower limit of the submersion degree is taken as the threshold for determining the dominant state of surging. In step three, the calculation process of the pixel-level confidence weight includes: A monotonically decreasing function in negative exponential form is constructed with surge submergence as the independent variable. The function is: ω(S)=exp(-λ∙max(0,S-S0)), where S is the surge submergence of the pixel, S0 is the submergence tolerance threshold, λ is the suppression intensity coefficient, exp is an exponential function with the natural constant e as the base, and max(0,S-S0) means taking the maximum value between S-S0 and 0. Reasons for choosing the function form: The negative exponential function curve is smooth and continuous. At the threshold point, the function value and its first derivative are continuous and there are no jumps. This can avoid the oscillation of inter-frame intensity caused by sudden changes in weights. The function form is simple, requiring only one subtraction, one multiplication and one exponentiation operation. Its impact on inference latency is negligible. The physical meaning of the parameters is clear. The two parameters control when to start suppression and how strong the suppression is, respectively. Among them, the flooding tolerance threshold S0 represents the maximum pixel-level flooding degree that the system can tolerate. When the flooding degree of a pixel does not exceed the flooding tolerance threshold, the tumbling signal is considered to be within an acceptable range and no suppression is applied. When the flooding degree exceeds the flooding tolerance threshold, the weight attenuation is applied to the pixel. The flooding tolerance threshold and the surge dominance state determination threshold are numerically the same, but their functions are different: the surge dominance state determination threshold is used to classify pixels into surge-dominant / non-surge-dominant categories, while the flooding tolerance threshold serves as the dividing point of the continuous weight function. In engineering implementation, the same value is used for both. The suppression intensity coefficient λ controls the rate at which the weight decreases with increasing submersion after exceeding the submersion tolerance threshold. Its value directly determines the shape of the weighting function. If the coefficient is too small, suppression is insufficient, and even with high submersion, a large weight is retained, allowing surge interference to still contaminate the churning intensity. If the coefficient is too large, suppression is too steep, and the weight approaches zero shortly after the submersion threshold is exceeded, potentially leading to excessive sacrifice of information about the actual churning area. The determination process includes: First, the system was operated in pure flow mode in the biological tank (the aeration blower was turned off, and only the flow promoter was turned on). The suppression intensity coefficient was gradually increased, and the confidence weight distribution of each pixel was observed. When the suppression intensity coefficient increased to the point that the average confidence weight of the entire tank was lower than 0.05, the suppression intensity coefficient value was recorded as the lower limit candidate value. The physical meaning of the test is that in a pure surge environment with no turbulence signal, the confidence weight of the system should suppress the effective contribution of all pixels to close to zero. Secondly, under normal aeration conditions (both the propeller and the aerator are turned on normally), the inhibition intensity coefficient is increased from the lower limit candidate value. After each increase, the weighted aggregated tumbling intensity index is calculated and compared with the aeration intensity manually evaluated at the same time. When further increasing the coefficient begins to cause the tumbling intensity index to decrease by more than 5% of the amplitude that the pure tumbling signal should have, the inhibition intensity coefficient value is recorded as the upper limit candidate value. Finally, the geometric mean of the lower limit candidate value and the upper limit candidate value is taken as the final determined suppression intensity coefficient; For each pixel in the surface area of ​​the biological treatment pool, the flood inundation degree is subtracted from the flood inundation tolerance threshold. If the difference is less than or equal to 0, the confidence weight of the pixel is directly assigned to 1. If the difference is greater than 0, the difference is multiplied by the inhibition intensity coefficient λ, and then the negative value is used as the exponent of the natural constant e. The exponential function value is calculated to obtain the confidence weight of the pixel. After calculating the confidence weights for all pixels, the weight values ​​of each pixel are arranged according to their spatial position to generate a pixel-level confidence weight map with the same size as the image. It should be noted that the role of the credibility assessment is to quantify the degree to which each pixel is submerged by the surge and convert it into a credibility weight for the tumbling signal. Step 4: Perform pixel-level weighted fusion of the tumbling signal on the surface of the biological treatment tank according to the credibility weight system of the tumbling signal, and perform amplitude correction in combination with the global thrust interference intensity coefficient, and output the corrected tumbling intensity field and tumbling activity heat map. For example, such as Figure 5 As shown, Figure 5 The relationship between pixel-level confidence weight ω and surge flooding level S is expressed as a negative exponential function: ω=exp(-λ·max(0,S-S0)), where S0 is the flooding tolerance threshold and λ is the suppression intensity coefficient. When S does not exceed S0, the weight remains at 1, indicating that the surge signal of that pixel is reliable. When S exceeds S0, the weight decreases exponentially with the increase of S, which smoothly suppresses the contribution of pixels severely polluted by surges, avoids abrupt changes in weight between frames due to hard thresholds, and ensures the stability of subsequent weighted fusion. In step four, the output process of the corrected tumbling intensity field includes: The first point to clarify is that the calculation of the tumbling intensity characterization value at each pixel location is as follows: For any pixel in the surface area of ​​the biological treatment pool, the high-frequency horizontal component and the high-frequency vertical component at the current moment are squared and added together. The square root of the sum is then taken to obtain the instantaneous tumbling intensity characterization value of the pixel at the current moment. It is understandable that the physical meaning of the instantaneous tumbling intensity characterization value is: reflecting the instantaneous motion velocity of a pixel in the current frame contributed by aeration tumbling, with the unit being pixels / frame. Unlike the accumulated energy value of the time window used for submersion assessment, the transient response value at the current moment is used here, so that the subsequent tumbling intensity index can respond quickly to real-time changes in aeration intensity. Before starting spatial weighted aggregation, all pixels in the surface area of ​​the biological pool are first screened for validity. The screening condition is that the confidence weight of the pixel must be greater than the preset minimum positive threshold, such as 0.05. Within the surface area of ​​the biological treatment pond, pixels with a confidence weight greater than a preset threshold are selected and marked as valid pixels. The proportion of valid pixels to the total number of pixels within the surface area of ​​the biological treatment pond is calculated to obtain the proportion of valid pixels. The preset threshold is a very small positive threshold, such as 0.05. If the effective pixel ratio is greater than or equal to the preset effective ratio lower limit threshold (e.g., 0.15), the tumbling intensity assessment at the current moment is determined to have sufficient statistical confidence; otherwise, the corrected tumbling intensity field is marked as invalid. For tumbling intensity with sufficient statistical confidence, weighted aggregation is performed. Specifically, the tumbling intensity characterization value of each pixel is taken and multiplied by the corresponding confidence weight to obtain the weighted tumbling intensity contribution value of the pixel. Summing up the weighted tumbling intensity contribution values ​​of all valid pixels yields the total weighted tumbling intensity. Summing up the confidence weights of all valid pixels yields the total weights. Dividing the total weighted tumbling intensity by the total weights yields the corrected tumbling intensity index at the current moment. In step four, the output process of the corrected tumbling activity heatmap includes: The first point to clarify is that the heatmap of turbulence activity is reconstructed based on high-frequency components in the spatial domain, specifically as follows: From the output of the spatial frequency band decomposition, extract the high-frequency spatial motion vector field at the current moment, which includes the high-frequency spatial component matrix in the horizontal direction and the high-frequency spatial component matrix in the vertical direction; For each pixel in the biochemical pool water surface area, the horizontal and vertical high-frequency spatial component values ​​of the pixel are taken, squared and added together. The square root of the sum is then taken to obtain the instantaneous tumbling amplitude of the pixel based on the spatial high-frequency components at the current moment. After the calculation is completed for all pixels, a high-frequency spatial component amplitude map with the same size as the biochemical pool water surface area is obtained. The instantaneous amplitude of the high-frequency spatial components of each pixel in the most recent time window queue is accumulated and normalized in the same way as when constructing the initial tumbling activity heatmap, to obtain the tumbling activity heatmap reconstructed based on the high-frequency spatial components. Secondly, it should be noted that the calculation of the correction coefficient is as follows: Construct a correction function in negative exponential form: when the global push interference intensity coefficient is less than or equal to the global interference tolerance threshold, the correction coefficient is always equal to 1; otherwise, the correction coefficient is equal to the value of an exponential function with the natural constant as the base and the negative value of the product of the threshold excess and the global correction intensity coefficient as the exponent. It should be noted that the global interference tolerance threshold represents the maximum degree of global push flow interference that the system can tolerate. The value is the same as the surge dominance state judgment threshold, but it is for global macroscopic energy rather than pixel-level local energy. Substitute the global push flow interference intensity coefficient into the correction function to calculate the correction coefficient at the current moment. The correction intensity coefficient controls the rate at which the correction strength increases with the degree of interference after the tolerance threshold is exceeded. Understandably, the physical necessity of global correction lies in the fact that even after high-pass filtering through spatial frequency band decomposition, due to the non-ideal finite roll-off characteristics of the filter transfer function in the transition band, low-frequency surge energy may still leak into the high-frequency spatial component channel with a very small residual amount. Under extreme conditions where the overall proportion of surge energy is extremely high, although this residual leakage is small, its absolute amount may still reach a level comparable to the real tumbling signal. Global correction offsets the impact of residual leakage on the input amplitude of the subsequent semantic segmentation model by compressing the overall amplitude of the heat map. Multiply the normalized tumbling activity heatmap by the correction coefficient to obtain the corrected tumbling activity heatmap, and denote the corrected tumbling activity heatmap as the corrected tumbling activity field. It should be noted that the corrected tumbling intensity index and the corrected tumbling activity field have the same physical meaning, both reflecting tumbling signals, but they are different in form: the corrected tumbling intensity index is a scalar, representing a single-value quantization of the average tumbling intensity of the entire pool, while the corrected tumbling activity field is a two-dimensional matrix, representing the spatial distribution of the tumbling activity level at each location in the entire pool. It should be noted that the role of feature reconstruction is to reconstruct the separated pure tumbling signal into two forms: a scalar expression of the average intensity of the entire pool and a matrix expression of the spatial distribution of the entire pool. Step 5: Perform semantic segmentation on the corrected tumbling activity field and output tumbling coverage, tumbling intensity index and aeration uniformity index; In step five, the process of semantically segmenting the corrected tumbling activity field includes: The corrected turbulence activity field is used as the main input channel of the semantic segmentation model. In addition to the main input channel, two auxiliary input channels are added: Auxiliary channel one is the grayscale brightness map of the current frame, which converts the color image of the current frame into a grayscale image. The brightness value of each pixel is used as the input value of auxiliary channel one, providing the model with the original visual information of the water surface texture, which helps to distinguish the broken water surface texture in the turbulence area from the smooth water surface in the non-turbulence area. Auxiliary channel two is the foreground dynamic mask, which provides the model with independent evidence of the existence of motion, and corroborates the heat information of the heat map. Arrange the main input channel, auxiliary channel one, and auxiliary channel two in the order of main channel, brightness channel, and mask channel to form a three-channel input tensor; The system loads a pre-trained lightweight U-shaped network semantic segmentation model. The model adopts an encoder-decoder symmetric structure and its parameters have been optimized during the training phase before system deployment. The training dataset covers labeled images under different working conditions, different lighting conditions, and different seasons. The three-channel input tensor is fed into the semantic segmentation model to perform forward inference calculation. After the encoder performs layer-by-layer downsampling, the decoder performs layer-by-layer upsampling, and the skip connection features are fused, the model outputs a single-channel probability map with the same spatial size as the input image. The value of each pixel in the probability map is between 0 and 1, representing the probability that the pixel belongs to the turbulent region. The closer the value is to 1, the more confident the model is that the position is a turbulent region. The closer the value is to 0, the more confident the model is that the position is not a turbulent region. Perform binarization on the tumbling probability map, set the binarization threshold to 0.50, and compare the probability value of each pixel in the probability map with this threshold. Pixels with a probability value greater than or equal to 0.50 are classified as turbulent areas and assigned a value of 1 at the corresponding position in the binary mask. Pixels with a probability value less than 0.50 are classified as non-turbulent areas and assigned a value of 0 at the corresponding position in the binary mask, thus obtaining a preliminary turbulent area binary mask. Perform two morphological operations sequentially on the initial binary mask: Opening operation: First, perform an erosion operation on the binary mask, and perform a logical AND operation between each pixel with a value of 1 and its surrounding pixels. Only when all pixels in the neighborhood are 1 will the pixel remain 1, otherwise it will be set to 0. Then, perform a dilation operation on the erosion result, and propagate each pixel with a value of 1 to its surrounding neighborhood, so that all pixels in the neighborhood are set to 1. Closing operation: First, perform a dilation operation on the binary mask, and then perform an erosion operation on the dilation result; For the binary mask after morphological post-processing, extract all connected components, calculate the number of pixels contained in each connected component, i.e. the area of ​​the connected component, set the minimum area threshold to 200 pixels, determine the connected components with an area less than 200 pixels as noise regions, change the value of the pixels in them from 1 to 0, remove them from the tumbling mask, and output the final tumbling region binary mask. It should be noted that the minimum area threshold of 200 pixels corresponds to an actual pool surface area of ​​approximately 0.15 square meters. The physical meaning is that isolated churning patches smaller than the minimum area threshold cannot physically correspond to the working area of ​​a complete aeration disc. They are more likely to originate from local disturbances of floating objects on the water surface or image noise. The total number of pixels with a value of 1 in the final binary mask of the turbulent area is counted, which is the number of pixels identified as the turbulent area. The total number of pixels in the surface area of ​​the biological pool is counted, which is the total number of pixels covered by the mask. The turbulent area pixel count is divided by the total number of pixels in the surface area of ​​the biological pool to obtain the turbulent coverage rate. It is understandable that the physical meaning of tumbling coverage rate is the proportion of the water surface area that is effectively covered by tumbling at the current moment. When the tumbling coverage rate is high, it indicates that most water surface areas are aerated and tumbling, and the aerators are working normally. When the tumbling coverage rate is low, it indicates that a large area of ​​water surface lacks tumbling, which may be due to insufficient air volume or failure of aerators in a large area. The tumble intensity index is directly adopted from the corrected tumble intensity index calculated by spatial weighted aggregation. The surface area of ​​the biological treatment tank is divided into N columns along the length and M rows along the width, forming a total of N by M rectangular grids. Each grid covers a fixed rectangular area of ​​the tank surface. For example, it can be divided into 10 columns along the length and 4 rows along the width, for a total of 40 grids. Each grid corresponds to an area of ​​about 2 meters by 2 meters on the actual tank surface. For each grid, the sum of the high-frequency turbulence component energy values ​​of all pixels in the grid is counted to obtain the total high-frequency energy of the grid. The total high-frequency energy of all grids is added together to obtain the total high-frequency turbulence energy of the entire biochemical pool surface area. The total high-frequency energy of each grid is divided by the total high-frequency turbulence energy of the entire area to obtain the high-frequency energy percentage of the grid. Calculate the arithmetic mean of the energy percentage of all grids. Since the sum of the energy percentages of all grids is 1, the average value is always equal to the reciprocal of the total number of grids. Calculate the spatial variation coefficient of the energy proportion of all grids, and define the aeration uniformity index as the complement of the variation coefficient: set an upper limit value of the variation coefficient, which represents the maximum degree of non-uniformity that the system can tolerate, divide the variation coefficient by the upper limit value of the variation coefficient to obtain a normalized ratio, and then subtract the normalized ratio from 1 to obtain the aeration uniformity index. When the coefficient of variation is zero (the energy of each grid is completely equal), the aeration uniformity index is 1, which represents the most uniform ideal state. When the coefficient of variation reaches the upper limit, the aeration uniformity index is 0, which means that the system-defined minimum non-uniformity boundary has been reached. When the coefficient of variation is between the two, the uniformity index changes linearly between 0 and 1. It should be noted that the purpose of the three output parameters is to transform visual features into three business indicators that can serve process decision-making.

[0020] The technical solution and advantages of this application are as follows: By performing optical flow spatial consistency detection on the surface of the biological treatment tank, it is determined whether there is coupling interference between the push flow and tumbling motion. If so, the surface motion energy is accumulated and normalized over time to generate an initial tumbling activity heatmap; the mixed motion signal of the biological treatment tank surface is separated into motion frequency bands in both time and space, decoupling the mixed motion signal into tumbling and surging components, and calculating the global push flow interference intensity coefficient; the surging submersion degree of the biological treatment tank surface is evaluated pixel by pixel, and the pixel-level confidence weight is calculated; the tumbling signal of the biological treatment tank surface is weighted and fused at the pixel level according to the tumbling signal confidence weight system, and amplitude correction is performed in combination with the global push flow interference intensity coefficient, outputting the corrected tumbling intensity field and tumbling activity heatmap; the corrected tumbling activity field is semantically segmented, and the tumbling coverage, tumbling intensity index and aeration uniformity index are output. By performing optical flow spatial consistency detection on the water surface to determine the presence of thrust-tumbling motion coupling interference, an initial tumbling activity heatmap is generated. The mixed motion signal on the water surface is separated into spatiotemporal dual-domain motion mode frequency bands to decouple the tumbling component from the surge component, and the global thrust interference intensity coefficient is calculated. Surge submersion assessment is performed, and a tumbling signal credibility weighting system is constructed. Pixel-level weighted fusion and full-field amplitude correction are applied to the tumbling signal to reconstruct the tumbling intensity field and tumbling activity heatmap, and semantic segmentation is performed to output tumbling coverage, tumbling intensity index, and aeration uniformity index. This solves the problem of visual signals of aeration tumbling being masked under thrust-surge interference, achieving accurate and visual recognition of the aeration status in the biological treatment tank.

[0021] The embodiments of the present invention have been described in detail above, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A visual recognition method for aeration status in a biological treatment tank based on water surface turbulence morphology, characterized in that: Includes the following steps: By performing optical flow spatial consistency detection on the surface of the biological treatment tank, it is determined whether there is coupling interference between the push flow and the tumbling motion. If so, the energy of the water surface motion is accumulated and normalized over time to generate an initial tumbling activity heatmap. The motion frequency band of the mixed motion signal on the surface of the biological treatment tank is separated in both time and space, the mixed motion signal is decoupled into tumbling component and surging component, and the global thrust interference intensity coefficient is calculated. The flood inundation level of the biological treatment pool is assessed pixel by pixel, and the pixel-level confidence weight is calculated. The tumbling signal on the surface of the biological treatment tank is weighted and fused at the pixel level according to the credibility weighting system of the tumbling signal, and the amplitude is corrected in the surface area of ​​the biological treatment tank by combining the global thrust interference intensity coefficient. The corrected tumbling intensity field and tumbling activity heat map are output. The corrected tumbling activity field is semantically segmented to output tumbling coverage, tumbling intensity index and aeration uniformity index.

2. The visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to claim 1, characterized in that: The method for determining whether there is coupling interference between the push-flow and tumbling motion is as follows: Continuously acquire video images of the surface of the biological treatment pool, and preprocess each frame of the acquired raw image sequentially; For two adjacent frames of images after preprocessing, the motion vector field of each pixel is calculated. Each pixel outputs a two-dimensional motion vector containing horizontal and vertical displacement components, and the optical flow amplitude field and optical flow direction field are calculated. Within the water surface area of ​​the biological treatment pool, the spatial variation coefficient of the optical flow amplitude field is calculated, and the spatial standard deviation of the optical flow direction field is obtained by using the circular statistical method. Maintain a time-series data queue to store the spatial variation coefficient of optical flow amplitude and the spatial standard deviation of optical flow direction for each frame of image within the past T seconds in chronological order. Calculate the spatial variation coefficient of optical flow amplitude and the spatial standard deviation of optical flow direction in the current frame, and compare them with their historical reference values ​​within the time-series sliding window. When both are lower than 70% of their respective reference values, it is determined that there is a push-flow-tumble motion coupling interference.

3. The visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to claim 2, characterized in that: The process of generating the initial tumbling activity heatmap is as follows: For each pixel in the water surface area of ​​the biological treatment pool, the optical flow calculation result corresponding to the most recent L frames is taken. The optical flow calculation result includes the motion displacement amplitude, horizontal displacement component and vertical displacement component. The motion displacement amplitude, horizontal displacement component and vertical displacement component are arranged in frame order to form three time series of length L. For each pixel, maintain multiple Gaussian distribution models, output binary judgments, and combine the binary judgment results of all pixels into a binary image as the foreground dynamic mask of the current frame. The motion displacement amplitude of each pixel in the time series and the dynamic mask of the foreground are accumulated in time sequence and normalized to obtain the normalized optical flow accumulation map and the foreground frequency matrix, respectively. Then, the initial tumbling activity heat map is obtained by pixel-by-pixel weighted fusion.

4. The visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to claim 1, characterized in that: The process of separating motion frequency bands in both time and space dimensions is as follows: A time-domain high-pass filter is performed on the motion time sequence of each pixel to separate the high-frequency tumbling component and the low-frequency surge component according to the motion frequency band. The motion vector field is decomposed into two-dimensional frequency domain in the spatial domain, and the low-frequency spatial motion component and the high-frequency spatial motion component are separated by radial bandpass filtering. By combining the low-frequency spatial components in the horizontal and vertical directions with the high-frequency spatial components, a complete low-frequency spatial motion vector field and a high-frequency spatial motion vector field are obtained.

5. The visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to claim 4, characterized in that: The global push-stream interference intensity coefficient is calculated as follows: The total energy of the low-frequency spatial component and the total energy of the high-frequency spatial component in the surface area of ​​the biological treatment pool are calculated separately. The total energy is calculated by summing the squared values ​​of all pixels in each component matrix. Dividing the total low-frequency energy by the sum of the total low-frequency energy and the total high-frequency energy yields the global thrust interference intensity coefficient.

6. The visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to claim 1, characterized in that: The method for evaluating the flood inundation level of the biological treatment pool surface pixel by pixel is as follows: The high-frequency component sequence and low-frequency component sequence of each pixel in the water surface area of ​​the biological treatment pool are respectively squared energy accumulation to obtain the total energy of high-frequency tumbling and the total energy of low-frequency surging. Surge submersion level = Total energy of low-frequency surges / (Total energy of high-frequency tumbling + Total energy of low-frequency surges). If the submersion level is greater than the threshold for determining the dominant state of the surge, then the pixel is determined to be in the dominant state of the surge; otherwise, the pixel is determined to be in the non-dominant state of the surge.

7. The visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to claim 6, characterized in that: The pixel-level confidence weights are calculated as follows: A monotonically decreasing function in negative exponential form is constructed with the inundation degree of the surge as the independent variable as the confidence weight function. For each pixel in the water surface area of ​​the biological treatment pool, the confidence weight of the pixel is obtained by substituting the difference between the inundation degree of the surge and the inundation tolerance threshold into the confidence weight function. After calculating the confidence weights for all pixels, the confidence weights of each pixel are arranged according to their spatial location to generate a pixel-level confidence weight map with the same size as the image.

8. The visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to claim 1, characterized in that: The reconstruction method of the tumbling intensity field is as follows: For any pixel within the surface area of ​​the biological treatment pool, the high-frequency horizontal component and the high-frequency vertical component at the current moment are squared and then added together. The square root of the sum is then taken to obtain the instantaneous tumbling intensity value at the current moment. Valid pixels that meet the confidence weight requirements within the surface area of ​​the biological treatment pool are selected, and the proportion of valid pixels to the total number of pixels within the surface area of ​​the biological treatment pool is calculated to obtain the effective pixel ratio. If the effective pixel ratio is greater than or equal to the preset effective ratio lower limit threshold, then the tumbling intensity assessment at the current moment is determined to have sufficient statistical confidence, and weighted aggregation is performed: The corrected tumbling intensity index is obtained by weighting the tumbling intensity values ​​of all valid pixels with confidence weights.

9. The visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to claim 8, characterized in that: The output process of the corrected tumbling activity heatmap is as follows: Based on the high-frequency spatial motion vector field at the current moment, calculate the instantaneous tumbling amplitude of each pixel point in the water surface area of ​​the biological pool based on the high-frequency spatial components, and obtain the high-frequency spatial component amplitude map. The instantaneous amplitude of the high-frequency spatial components of each pixel in the most recent time window queue is accumulated and normalized in time to obtain a heat map of tumbling activity based on the reconstruction of high-frequency spatial components. A correction function is constructed, and the global thrust interference intensity coefficient is substituted into the correction function to calculate the correction coefficient at the current time. The normalized tumbling activity heatmap is multiplied by the correction coefficient to obtain the corrected tumbling activity heatmap, which is denoted as the corrected tumbling activity field.

10. The visual recognition method for aeration status of a biological treatment tank based on water surface turbulence morphology according to claim 1, characterized in that: The output methods for the tumbling coverage rate, tumbling intensity index, and aeration uniformity index are as follows: The corrected tumbling activity field is used as the main input channel of the semantic segmentation model, and two auxiliary channels, the grayscale brightness map of the current frame and the foreground dynamic mask, are added to form a three-channel input tensor. Load a pre-trained lightweight U-shaped network semantic segmentation model, feed the three-channel input tensor into the model to perform forward inference calculation, and output a single-channel probability map with the same spatial size as the input image; Perform binarization on the tumbling probability map to obtain a preliminary tumbling region binary mask. Then, perform opening and closing operations in sequence, extract all connected components and remove connected components with an area smaller than the minimum area threshold, and output the final tumbling region binary mask. The tumbling coverage rate is obtained by dividing the total number of pixels with a median value of 1 in the final binary mask of the tumbling area by the total number of pixels in the surface area of ​​the biological pool. The tumble intensity index is directly adopted from the corrected tumble intensity index calculated by spatial weighted aggregation. The surface area of ​​the biological treatment tank is divided into multiple grids. For each grid, the spatial variation coefficient of the energy proportion of all grids is calculated, and the aeration uniformity index is the complement of the variation coefficient.

Images