Visual monitoring system for river waste discharge

Abstract

The invention discloses a visual monitoring system for river waste discharge, relates to the technical field of river waste discharge monitoring, and solves the technical problem that waste discharge targets in rivers cannot be accurately identified and quantified, resulting in deviation in judgment of waste discharge conditions. The waste discharge target and the position and category of the waste discharge target can be accurately recognized, the turbidity difference value is calculated through the turbidity difference feature extraction module, the waste discharge recognition model is built by integrating related feature parameters, the accuracy of waste discharge condition judgment is improved, and the improved YOLOv5 algorithm is improved through network structure optimization, data enhancement strategy improvement and loss function adjustment. The detection capability of a waste discharge target under a complex background is enhanced, and the interference of environmental factors on target detection is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of river waste discharge monitoring, and in particular to a river waste discharge visual monitoring system. Background Art

[0002] According to the patent application with publication number CN117871803A, a wastewater discharge monitoring system is disclosed, which particularly relates to the field of wastewater discharge monitoring technology, including an information acquisition module for periodically acquiring wastewater information at a factory wastewater discharge outlet in a city and river information, tap water information and weather information in the vicinity of the factory wastewater discharge outlet; a wastewater analysis module for analyzing wastewater quality; an impact analysis module for analyzing river quality, tap water quality and water quality influencing parameters; an information storage module for storing wastewater quality, river quality and tap water quality; an adjustment and optimization module for adjusting and optimizing wastewater quality and water quality influencing parameters; a quality analysis module for analyzing wastewater compliance and also for correcting the adjustment process of wastewater quality in the next cycle; and an output module for outputting wastewater compliance.

[0003] Traditional river waste discharge monitoring methods have many drawbacks. Manual sampling and laboratory testing are time-consuming and laborious, with low monitoring frequency, and it is difficult to reflect the river waste discharge status in real time; the sensor-based online monitoring system has a single monitoring parameter, is easily disturbed by the environment, and has a limited monitoring range; some visual monitoring systems do not fully explore the waste discharge characteristics, resulting in poor monitoring accuracy and reliability. In this context, an efficient and accurate river waste discharge monitoring system is urgently needed to deal with the increasingly serious problem of river pollution. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a river waste discharge visual monitoring system, which solves the problem that it is impossible to accurately identify and quantify waste discharge targets in rivers, resulting in deviations in the judgment of waste discharge conditions.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A river waste discharge visual monitoring system, comprising: The data preprocessing module is used to obtain the image data and water quality data transmitted by the multi-source data acquisition module, and perform Gaussian filtering and image enhancement processing on the image data to obtain a preprocessed image, perform outlier correction and normalization processing on the water quality data to obtain preprocessed data, and transmit the two to the target detection and recognition module; The target detection and recognition module is used to use the improved YOLOv5 target detection algorithm to analyze the preprocessed image to generate target detection results and transmit them to the turbidity difference feature extraction module; The turbidity difference feature extraction module is used to obtain the target detection results, determine the sampling points in the waste discharge target area and the periodic background area, obtain the turbidity values ​​of the sampling points, and calculate the corresponding turbidity difference between the two to generate turbidity difference information, then calculate the turbidity change rate and standard deviation, skewness and kurtosis based on the change in the turbidity difference between two adjacent groups of pre-processed images to generate turbidity features, and transmit the turbidity difference information and turbidity features to the waste discharge recognition model construction module; The waste discharge identification model building module is used to fuse the acquired target detection results, turbidity characteristics and turbidity difference information to obtain the input of the waste discharge identification model, and use the support vector machine as the classifier for model training to generate the waste discharge identification model, and transmit it to the monitoring result output and early warning module; The monitoring result output and early warning module is used to calculate the turbidity difference according to the waste discharge identification model, and compare it with the dynamic threshold to generate real-time output monitoring results or normal monitoring results, and perform secondary analysis on the real-time output monitoring results, match the turbidity difference with the preset interval to generate the waste discharge severity, and output it in real time.

[0006] As a further solution of the present invention, it also includes a multi-source data acquisition module for collecting river image data and relevant water quality data such as turbidity, and transmitting them to the data preprocessing module at the same time. The image data is acquired through a high-definition camera, and the water quality data is measured through a turbidity sensor.

[0007] As a further solution of the present invention, the data preprocessing module performs Gaussian filtering and image enhancement processing on the image data to obtain the preprocessed image in the following specific manner: Gaussian filtering is used to denoise the image. The specific formula is: And G(x, y) is the weight coefficient of the Gaussian filter template at the coordinate (x, y), e is a natural constant, σ is the standard deviation of the Gaussian distribution, (x 0 ,y 0 ) is the center coordinate of the template, and weighted average is performed on each pixel in the image and its neighboring pixels; Count the number of pixels at each gray level in the image to obtain the gray histogram of the image, calculate the gray transformation function based on the histogram, map the gray value of the original image to the new gray value through the transformation function, and obtain the enhanced image; The preprocessed image is obtained by combining the two processes.

[0008] As a further solution of the present invention, the data preprocessing module performs outlier correction and normalization processing on the water quality data to obtain preprocessed data in a specific manner: Outlier detection is performed based on the interquartile range. The first quartile Q1 and the third quartile Q3 of the turbidity data are calculated. Then, the interquartile range IQR = Q3 - Q1 is calculated. Values in the data that are less than Q1 - 1.5IQR or greater than Q3 + 1.5IQR are regarded as outliers, and the outliers are corrected using linear interpolation to obtain the corrected data; The obtained corrected data is normalized. The normalization formula is , where x norm is the normalized turbidity value, x is the original turbidity value, x max and x min are the minimum and maximum values of the turbidity data respectively, and the preprocessed data is obtained.

[0009] As a further solution of the present invention, the specific method for correcting outliers using linear interpolation to obtain the corrected data is as follows: For each outlier, search forward and backward to find the two nearest normal turbidity data points. Obtain the index of the outlier denoted as i, the index of the previous normal data point as j, and the index of the next normal data point as k, and j < i < k. Obtain the normal data points (j, y j ) and (k, y k ). According to the two-point form straight line equation formula, on the x-axis, the linear relationship between x j and x k is . According to the position x = i of the outlier, calculate the corrected value y i . Finally, replace the outlier in the original dataset with the obtained corrected value y i to obtain the corrected data.

[0010] As a further solution of the present invention, the specific method for the turbidity difference feature extraction module to generate turbidity difference information is as follows: Determine the waste discharge target area and the surrounding background area. At the same time, select m and k sampling points in the waste discharge target area and the surrounding background area respectively, and calculate the average turbidity value Tp of the waste discharge target area and the corresponding average turbidity value Tp1 of the surrounding background area. According to the formula calculate the turbidity difference and generate turbidity difference information.

[0011] As a further solution of the present invention, the specific method for the turbidity difference feature extraction module to generate turbidity features is as follows: Obtain the preprocessed image. Then, calculate the change amount of the turbidity difference between two adjacent groups of preprocessed images, and obtain the corresponding time interval. Calculate the turbidity change rate according to the formula; Analyze and calculate all the preprocessed images, and calculate the mean value of all the turbidity change rates denoted as the turbidity change rate RΔT , then the standard deviation, skewness and kurtosis of the turbidity values ​​in the waste discharge target area are calculated, and the turbidity characteristics are obtained by integrating the turbidity change rate.

[0012] As a further solution of the present invention, the waste discharge identification model building module generates the waste discharge identification model in the following specific manner: The acquired target detection results, turbidity features and turbidity difference information are fused, the target category is encoded using a one-hot encoding method, and is concatenated with the target position coordinates (x, y) and turbidity difference feature parameters into a one-dimensional vector as the input of the waste discharge recognition model; Use the training set to train the SVM model. After the training is completed, use the test set to evaluate the performance of the model to generate a normal signal or an abnormal signal. For the generated evaluation abnormal signals, the model is adjusted and optimized until the evaluation results are normal.

[0013] As a further solution of the present invention, the specific manner in which the monitoring result output and early warning module generates real-time output monitoring results or normal monitoring results is: According to the obtained waste identification model, the turbidity difference is calculated and recorded as Then, obtain historical data to set dynamic thresholds and obtain the turbidity difference and the waste discharge judgment threshold T threshold For comparison, if the turbidity difference Less than the waste discharge judgment threshold T threshold , then a normal monitoring result is generated. If the turbidity difference Greater than the waste discharge judgment threshold T threshold , then real-time output monitoring results are generated.

[0014] As a further solution of the present invention, the specific manner in which the monitoring result output and early warning module performs secondary analysis on the real-time output monitoring result is: Continuously collect river turbidity data and calculate turbidity difference in real time according to the established algorithm , calculate the turbidity difference and the waste discharge judgment threshold T threshold The difference is recorded as the turbidity difference, and the obtained turbidity difference is matched with the corresponding preset interval to generate the corresponding waste discharge severity and output it in real time.

[0015] The present invention provides a visual monitoring system for river waste discharge. Compared with the prior art, it has the following beneficial effects: The present invention can accurately identify waste discharge targets and their locations and categories by integrating turbidity difference features, combining advanced machine vision technology and image processing algorithms, calculating turbidity differences using a turbidity difference feature extraction module, and constructing a waste discharge recognition model based on relevant feature parameters, thereby improving the accuracy of waste discharge judgment. The improved YOLOv5 algorithm enhances the detection capability of waste discharge targets in complex backgrounds and reduces the interference of environmental factors on target detection through network structure optimization, data enhancement strategy improvement, and loss function adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a block diagram of the system principle of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] See also Figure 1 The present application provides a river waste discharge visual monitoring system, including a multi-source data acquisition module, a data preprocessing module, a target detection and recognition module, a turbidity difference feature extraction module, a waste discharge recognition model construction module and a monitoring result output and early warning module, and in combination with the attached Figure 1 It can be known that the above functional modules are electrically connected in a unidirectional manner.

[0019] Multi-source data acquisition module, which is used to collect river image data and related water quality data such as turbidity, and transmit them to the data preprocessing module. Among them, image acquisition uses high-definition cameras, and through reasonable installation layout, it ensures that clear and comprehensive river surface images can be obtained; turbidity data acquisition uses turbidity sensors to measure the turbidity value of the river in real time.

[0020] Data preprocessing module, which is used to preprocess the acquired image data and water quality data, wherein the operations of preprocessing the image data include denoising and image enhancement. The specific method of denoising the image data is to use Gaussian filtering to denoise the image, and to eliminate noise by weighted averaging each pixel point in the image and its neighboring pixels. The formula of Gaussian filtering is: , and in the above formula G(x, y) is the weight coefficient of the Gaussian filter template at the coordinate (x, y), e is a natural constant, σ is the standard deviation of the Gaussian distribution, (x 0 ,y 0 ) is the center coordinate of the template; The specific method for image enhancement is to perform enhancement processing on the image using histogram equalization. First, count the number of pixels for each gray level in the image to obtain the gray histogram of the image. Then, calculate the gray transformation function based on the histogram, and map the gray values of the original image to new gray values through the transformation function to obtain the enhanced image; Based on the denoising processing and image enhancement processing, a corresponding preprocessed image is obtained; Perform preprocessing operations on the obtained water quality data. The preprocessing operations include outlier detection and correction and normalization processing. The specific process of outlier detection and correction is to use the method based on the interquartile range (IQR) for outlier detection. First, calculate the first quartile Q1 and the third quartile Q3 of the turbidity data. Then, calculate the interquartile range IQR = Q3 - Q1. Consider the values in the data that are less than Q1 - 1.5IQR or greater than Q3 + 1.5IQR as outliers, and use the method of linear interpolation to correct the outliers. The specific correction process is as follows; For each outlier, search forward and backward to find the two nearest normal turbidity data points. Assume the index of the outlier is i, the index of its previous normal data point is j, and the index of the next normal data point is k, and j < i < k. Let the normal data points be (j, y j ), and (k, y k ). According to the two-point form straight line equation formula, on the x-axis (here represented as the index order of the data points), the linear relationship between x j and x k is . For the position x = i of the outlier, the corrected value y i can be calculated accordingly. Finally, replace the outlier in the original dataset with the obtained corrected value y i to obtain the corrected data; At the same time, perform normalization processing on the obtained corrected data to normalize the turbidity data to the range of [0, 1]. The normalization formula is . Here, x and x in the above analysis process do not represent the same meaning, but refer to a kind of variable. Among them, x norm is the normalized turbidity value, x is the original turbidity value, x max and x min are the minimum and maximum values of the turbidity data respectively to obtain the preprocessed data; At the same time, transmit the obtained preprocessed image and preprocessed data to the target detection and recognition module.

[0021] The target detection and recognition module uses the improved YOLOv5 target detection algorithm to analyze the preprocessed image, identify the waste discharge targets in the image, and the waste discharge targets include sewage outlets, floating pollutants, etc., and determine their locations and categories. At the same time, the target detection results are generated and transmitted to the turbidity difference feature extraction module. Specifically, the improved YOLOv5 model is trained using the annotated river waste discharge image dataset. The dataset is annotated in VOC format, and the annotation content includes the category of the waste discharge target (such as sewage outlets, pollutants, etc.) and its location information. During the training process, the stochastic gradient descent (SGD) algorithm is used to optimize the parameters of the model, setting the initial learning rate to 0.001, the momentum factor to 0.9, and the weight decay coefficient to 0.0005. To prevent the model from overfitting, the L1 and L2 regularization methods are used, and the learning rate is dynamically adjusted during the training process.

[0022] Turbidity difference feature extraction module, which is used to analyze the obtained target detection results, determine the waste discharge target area and the surrounding background area based on the obtained target detection results, and select n sampling points in the waste discharge target area, where n=1, 2, ..., m, where m represents the number of sampling points, and obtain the turbidity value T corresponding to the sampling point n target，n Then calculate the average turbidity value Tp of the waste discharge target area, and select the same number of sampling points in the surrounding background area as o, and o=1, 2, ..., k, where k represents the number of sampling points in the surrounding background area, and similarly obtain the turbidity value T corresponding to the sampling point o background，o , and calculate the average turbidity value Tp1 corresponding to the surrounding background area, then the corresponding turbidity difference , and generate turbidity difference information; At the same time, the preprocessed images are obtained, and the preprocessed images obtained here are multiple groups. Then, the change in the turbidity difference between two adjacent groups of preprocessed images is calculated, and the corresponding time interval is obtained. The turbidity change rate is calculated according to the formula. Similarly, all preprocessed images are analyzed and calculated, and the average of all turbidity change rates is calculated and recorded as the turbidity change rate R. ΔT Then, the standard deviation, skewness and kurtosis of the turbidity values ​​in the waste discharge target area are calculated. These parameters can reflect the degree of dispersion of the turbidity values ​​in the target area, the symmetry of the distribution and the sharpness of the peak value, which is helpful to further analyze the characteristics of the waste discharge target and obtain the turbidity characteristics by combining the turbidity change rate. At the same time, the turbidity characteristics and turbidity difference information are transmitted to the waste discharge identification model construction module.

[0023] The waste discharge recognition model construction module is used to fuse the acquired target detection results, turbidity features and turbidity difference information. The specific fusion method is to encode the target category using the One-Hot Encoding method, and splice it with the target position coordinates (x, y) and the turbidity difference feature parameters into a one-dimensional vector as the input of the waste discharge recognition model; For example, a waste discharge target is detected, and its category is "industrial wastewater discharge outlet". After one-hot encoding, the category is encoded as [1, 0, 0] (assuming that there are only three categories in the data set: "industrial wastewater discharge outlet", "domestic sewage discharge outlet", and "agricultural non-point source discharge"). The target position coordinates are (x=300, y=400), the x coordinate range of the monitoring area is [0, 500], and the y coordinate range is [0, 600]. After normalization, x norm =0.6,y norm =0.67, and the coordinates of the center of the monitoring area are (250, 300), then the relative coordinate difference Δx=300-250=50, Δy=400-300=100, and the normalized relative coordinate difference Δx nom =0.2,Δy nom =0.33; At the same time, the calculated turbidity difference ΔT=12.5, the turbidity change rate R ΔT =0.25, after minimum-maximum scaling (and scaling range is [0, 1]) ΔT scaled =0.7, ΔT scaled =0.5. After correlation analysis, the product of turbidity difference and turbidity difference change rate is selected as the new feature, that is, 0.7×0.5=0.35. This information is concatenated into a one-dimensional vector. According to the order of unique hot encoding, absolute coordinates, relative coordinate difference, scaled turbidity difference features and new features, it is [1, 0, 0, 0.6, 0.67, 0.2, 0.33, 0.7, 0.5, 0.35]. This one-dimensional vector can be used as the input feature vector of the waste recognition model.

[0024] Next, support vector machine (SVM) is used as the classifier of the waste identification model, where SVM is a binary classification model based on statistical learning theory, which can find an optimal classification hyperplane in high-dimensional space to separate samples of different categories. When training the SVM model, radial basis kernel function (RBF) is used as the kernel function, and the optimal penalty parameter C and kernel function parameter are selected by cross-validation method. The fused feature vector is divided into a training set and a test set, and the training set is used to train the SVM model. After the training is completed, the test set is used to evaluate the performance of the model to generate an evaluation result, and the evaluation result includes evaluating normal signals or evaluating abnormal signals; For the generated normal evaluation signal, the waste discharge identification model is directly generated. For the generated abnormal evaluation signal, the model is adjusted and optimized until the evaluation result is normal. At the same time, the waste discharge identification model is generated and then transmitted to the monitoring result output and early warning module.

[0025] The monitoring result output and early warning module is used to calculate the turbidity difference according to the waste discharge identification model obtained. , and the turbidity difference here is different from the turbidity difference analyzed above. Here is the predicted turbidity difference. Then, historical data is obtained to set the dynamic threshold. First, the historical monitoring data is analyzed, the distribution of turbidity difference in different time periods is statistically analyzed, and the turbidity difference threshold range under different confidence levels is calculated. Then, according to the current real-time data of river flow, flow velocity, etc., combined with seasonal changes and other factors, the current waste discharge judgment threshold T is dynamically adjusted within the threshold range through the linear interpolation algorithm. threshold ; The turbidity difference and the waste discharge judgment threshold T threshold For comparison, if the turbidity difference Less than the waste discharge judgment threshold T threshold , then the normal monitoring results are generated and output in real time. Greater than the waste discharge judgment threshold T threshold , it means that there is waste discharge in the river, and the monitoring results are output in real time, including the location, category, turbidity difference characteristics and severity of waste discharge. At the same time, the obtained turbidity difference is analyzed again, and the specific method of the secondary analysis is as follows: Continuously collect river turbidity data and calculate turbidity difference in real time according to the established algorithm For example, the system collects and calculates data every 15 minutes to ensure that changes in waste discharge can be captured in a timely manner and the turbidity difference can be calculated. and the waste discharge judgment threshold T threshold The difference is recorded as the turbidity difference, and the obtained turbidity difference is matched with the corresponding preset interval. The setting of the preset interval is determined by the operator based on a large amount of actual data, and the corresponding waste discharge severity is generated and output in real time.

[0026] Assuming that a river is in normal condition, the dynamic turbidity threshold T is calculated based on historical data and real-time flow, flow rate and other factors. threshold The difference in turbidity is 50NTU (turbidity unit). When the difference in turbidity exceeds the threshold value within the range of 0-20%, it is classified as light waste discharge. In a certain monitoring, the calculated difference in turbidity is 55NTU. , which is in the light waste discharge interval. When the turbidity difference is 1.5-2 times the threshold, it is classified as moderate waste discharge. At another moment, the turbidity difference is 75NTU, and the multiple value calculated according to the formula is 1.5, which is just in the moderate waste discharge range, indicating that the waste discharge situation is already obvious and needs to be taken seriously. When the turbidity difference is greater than 2 times the threshold, it is classified as severe waste discharge. If the subsequent monitoring shows that the turbidity difference reaches 120NTU, the multiple value calculated according to the formula is 2.4 times, which is much greater than 2 times the threshold, and it belongs to severe waste discharge. The river ecology is facing serious threats.

[0027] Some of the data in the above formulas are calculated by taking their numerical values, and are not substituted into parameter units for calculation. At the same time, the contents not described in detail in this specification belong to the existing technologies known to those skilled in the art.

[0028] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A visual monitoring system for river waste discharge, characterized in that: include: The data preprocessing module is used to obtain the image data and water quality data transmitted by the multi-source data acquisition module, and perform Gaussian filtering and image enhancement processing on the image data to obtain a preprocessed image, perform outlier correction and normalization processing on the water quality data to obtain preprocessed data, and transmit the two to the target detection and recognition module; The target detection and recognition module is used to use the improved YOLOv5 target detection algorithm to analyze the preprocessed image to generate target detection results and transmit them to the turbidity difference feature extraction module; The turbidity difference feature extraction module is used to obtain the target detection results, determine the sampling points in the waste discharge target area and the periodic background area, obtain the turbidity values ​​of the sampling points, and calculate the corresponding turbidity difference between the two to generate turbidity difference information, then calculate the turbidity change rate and standard deviation, skewness and kurtosis based on the change in the turbidity difference between two adjacent groups of pre-processed images to generate turbidity features, and transmit the turbidity difference information and turbidity features to the waste discharge recognition model construction module; The waste discharge identification model building module is used to fuse the acquired target detection results, turbidity characteristics and turbidity difference information to obtain the input of the waste discharge identification model, and use the support vector machine as the classifier for model training to generate the waste discharge identification model, and transmit it to the monitoring result output and early warning module; The monitoring result output and early warning module is used to calculate the turbidity difference according to the waste discharge identification model, and compare it with the dynamic threshold to generate real-time output monitoring results or normal monitoring results, and perform secondary analysis on the real-time output monitoring results, match the turbidity difference with the preset interval to generate the waste discharge severity, and output it in real time.

2. A river waste discharge visual monitoring system according to claim 1, characterized in that: It also includes a multi-source data acquisition module for collecting river image data and related water quality data such as turbidity, and transmitting them to the data preprocessing module at the same time. The image data is acquired through a high-definition camera, and the water quality data is measured through a turbidity sensor.

3. A river waste discharge visual monitoring system according to claim 1, characterized in that: The specific method of the data preprocessing module performing Gaussian filtering and image enhancement processing on the image data to obtain the preprocessed image is: Gaussian filtering is used to denoise the image. The specific formula is: And G(x, y) is the weight coefficient of the Gaussian filter template at the coordinate (x, y), e is a natural constant, σ is the standard deviation of the Gaussian distribution, (x0, y0) is the center coordinate of the template, and weighted averaging is performed on each pixel point in the image and its neighboring pixels; Count the number of pixels at each gray level in the image to obtain the gray histogram of the image, calculate the gray transformation function based on the histogram, map the gray value of the original image to the new gray value through the transformation function, and obtain the enhanced image; The preprocessed image is obtained by combining the two processes.

4. A river waste discharge visual monitoring system according to claim 1, characterized in that: The specific method in which the data preprocessing module performs outlier correction and normalization processing on the water quality data to obtain preprocessed data is: Outlier detection was performed based on the interquartile range. The first quartile Q1 and the third quartile Q3 of the turbidity data were calculated. Then the interquartile range IQR = Q3-Q1 was calculated. The values ​​less than Q1-1.5IQR or greater than Q3+1.5IQR in the data were regarded as outliers. The outliers were corrected by linear interpolation to obtain corrected data. The corrected data is normalized, and the normalization formula is: , where x norm is the normalized turbidity value, x is the original turbidity value, and x max and x min are the minimum and maximum values ​​of the turbidity data, respectively, and the preprocessed data are obtained.

5. A river waste discharge visual monitoring system according to claim 4, characterized in that: The specific method of using the linear interpolation method to correct the outliers to obtain the corrected data is: For each outlier, search forward and backward to find the two nearest normal turbidity data points. Obtain the index of the outlier denoted as i, the index of the previous normal data point as j, and the index of the next normal data point as k, where j < i < k. Obtain the normal data points (j, y j ), and (k, y k ). According to the two-point form straight-line equation formula, on the x-axis, the linear relationship between x j and x k is . Based on the position of the outlier x = i, calculate the correction value y i . Finally, replace the outlier in the original dataset with the obtained correction value y i to obtain the corrected data.

6. A river waste discharge visual monitoring system according to claim 1, characterized in that: The specific method of generating turbidity difference information by the turbidity difference feature extraction module is as follows: Determine the waste discharge target area and the surrounding background area, select m and k sampling points in the waste discharge target area and the surrounding background area respectively, and calculate the average turbidity value Tp of the waste discharge target area and the average turbidity value Tp1 of the surrounding background area respectively. According to the formula The turbidity difference is calculated and turbidity difference information is generated.

7. A river waste discharge visual monitoring system according to claim 1, characterized in that: The specific method of generating turbidity features by the turbidity difference feature extraction module is: The preprocessed image is obtained, and then the change in the turbidity difference between two adjacent groups of preprocessed images is calculated, and the corresponding time interval is obtained, and the turbidity change rate is calculated according to the formula; All preprocessed images are analyzed and calculated, and the mean of all turbidity change rates is calculated and recorded as the turbidity change rate R ΔT , then the standard deviation, skewness and kurtosis of the turbidity values ​​in the waste discharge target area are calculated, and the turbidity characteristics are obtained by integrating the turbidity change rate.

8. A river waste discharge visual monitoring system according to claim 1, characterized in that: The specific method of generating the waste discharge identification model by the waste discharge identification model building module is as follows: The acquired target detection results, turbidity features and turbidity difference information are fused, the target category is encoded using a one-hot encoding method, and is concatenated with the target position coordinates (x, y) and turbidity difference feature parameters into a one-dimensional vector as the input of the waste discharge recognition model; Use the training set to train the SVM model. After the training is completed, use the test set to evaluate the performance of the model to generate a normal signal or an abnormal signal. For the generated evaluation abnormal signals, the model is adjusted and optimized until the evaluation results are normal.

9. A river waste discharge visual monitoring system according to claim 1, characterized in that: The specific method of the monitoring result output and early warning module to generate real-time output monitoring results or normal monitoring results is: According to the obtained waste identification model, the turbidity difference is calculated and recorded as Then, obtain historical data to set dynamic thresholds and obtain the turbidity difference and the waste discharge judgment threshold T threshold For comparison, if the turbidity difference Less than the waste discharge judgment threshold T threshold , then a normal monitoring result is generated. If the turbidity difference Greater than the waste discharge judgment threshold T threshold , then real-time output monitoring results are generated.

10. A river waste discharge visual monitoring system according to claim 1, characterized in that: The specific method of the monitoring result output and early warning module to perform secondary analysis on the real-time output monitoring results is: Continuously collect river turbidity data and calculate turbidity difference in real time according to the established algorithm , calculate the turbidity difference and the waste discharge judgment threshold T threshold The difference is recorded as the turbidity difference, and the obtained turbidity difference is matched with the corresponding preset interval to generate the corresponding waste discharge severity and output it in real time.

Images