Water quality detection system for water conservancy informatization water conservancy project

Abstract

The invention relates to the technical field of intelligent detection, and discloses a water conservancy informatization hydraulic engineering water quality detection system, which is characterized in that in a preset time period, feature data of a drainage filtering unit and a target pipeline are acquired; wherein the drainage filtering unit comprises a filter screen and a filter screen dust removal device, and the filter screen and the filter screen dust removal device are both installed at a water outlet of a target pipeline; identifying the water outlet images at the fixed time interval based on physical properties to obtain water outlet features in each water outlet image; obtaining the turbidity of the water body in the target pipeline according to a fixed time interval; associating the turbidity and the effluent characteristics at the same moment to obtain a regression mapping model of the turbidity and the effluent characteristics; in the monitoring time period, the monitoring time period is divided into a plurality of monitoring time slices based on the interval operation time of the filter screen dust removal device; and in each monitoring time slice, based on the corresponding water outlet image and the regression mapping model, obtaining the turbidity of the water body in the target pipeline.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection technology, and more specifically, to a water quality detection system for water conservancy information-based water conservancy projects. Background Art

[0002] In water conservancy projects, water bodies usually flow into sedimentation tanks or regulating tanks through water pipelines for further purification. In this process, in order to ensure the stability of downstream water treatment, it is necessary to monitor the suspended particulate matter in the water body transported by the pipeline online to evaluate the turbidity of the water quality, and then dynamically add particulate degradation materials (such as alum) according to the turbidity of the water quality during the treatment of downstream water. For the detection of the turbidity of the water body, it is usually necessary to use a large number of optical turbidity sensors to arrange and then obtain the turbidity of the water body. On the one hand, the optical turbidity sensor will be affected by light pollution, resulting in inaccurate measurement. On the other hand, the optical turbidity sensor will adhere to particulate matter, which will also lead to inaccurate measurement. Therefore, the existing technology has the problems of high measurement cost, high maintenance requirements and low measurement accuracy. Summary of the invention

[0003] The present invention provides a water quality detection system for water conservancy projects with water conservancy informationization, which solves the technical problems raised in the background technology.

[0004] The present invention provides a water quality detection system for water conservancy projects with water conservancy informationization, comprising: A data acquisition module, used to obtain characteristic data of the drainage filter unit and the target pipe within a preset time period; wherein the drainage filter unit includes a filter screen and a filter screen dust removal device, and the filter screen and the filter screen dust removal device are both installed at the water outlet of the target pipe; The characteristic data include: the water outflow image of the water outlet of the target pipeline, the interval operation time of the filter dust removal device, the water pressure of the target pipeline, and the turbidity of the water body in the target pipeline; A data processing module is used to perform semantic recognition on water-out images at fixed time intervals to identify water-out features in each water-out image; wherein the water-out features include: water-out angle, water-out stability and water-out atomization area; The correlation matching module is used to obtain the turbidity of the water body in the target pipeline at fixed time intervals; and to correlate the turbidity and the water discharge characteristics at the same time to obtain a regression mapping model of the turbidity and the water discharge characteristics; The dynamic monitoring module is used to divide the monitoring time period into several monitoring time slices based on the interval operation time of the filter dust removal device within the monitoring time period; in each monitoring time slice, the turbidity of the water body in the target pipeline is obtained based on the corresponding water outlet image and the regression mapping model.

[0005] Furthermore, the out-of-water image is acquired in the horizontal direction and at a fixed angle.

[0006] Further, the water outlet angle includes: performing a single-layer Haar wavelet decomposition on the water outlet image to extract the horizontal high-frequency component matrix H, the vertical high-frequency component matrix V and the diagonal high-frequency component matrix D of the water outlet image; weighting the horizontal high-frequency component matrix H, the vertical high-frequency component matrix V and the diagonal high-frequency component matrix D respectively and merging them to obtain a high-frequency feature matrix; wherein the weighting coefficients are ; Extract local fractal features from the high-frequency feature matrix, specifically including: initializing the high-frequency feature matrix, generating a convolution sliding window, the size of the convolution sliding window is N×N, and the sliding step is , the local fractal dimension in the convolution sliding window is calculated based on the box fractal algorithm to obtain a fractal dimension matrix; wherein, the element in the i-th row and j-th column of the fractal dimension matrix represents the fractal dimension of the pixel in the i-th row and j-th column of the water image; the elements in the fractal dimension matrix whose fractal dimension is greater than the preset dimension threshold are replaced with element 1, otherwise they are replaced with element 0 to obtain a water flow matrix; wherein, the elements with element value 1 in the water flow matrix represent the water flow in the water image; the element 1 in the water flow matrix is ​​detected in the eight-neighborhood, and if the element 1 exists in the eight-neighborhood of the element 1, a connection is established for the corresponding element The connected relationship is established to obtain the connected clusters in the water flow matrix; if the number of element 1 in the connected cluster is less than the preset number threshold, the element 1 in the corresponding connected cluster is replaced by element 0 to obtain the updated water flow matrix; wherein the element 1 in the updated water flow matrix corresponds to the water flow in the water outlet image; the covariance matrix of the updated water flow matrix is ​​obtained, and the eigenvalue decomposition of the covariance matrix is ​​performed to extract the main eigenvector of the covariance matrix; the arc cosine value of the ratio of the dot product of the main eigenvector and the preset standard reference vector to the module length product is calculated to obtain the water outlet angle; wherein the preset standard reference vector is expressed as , the water outlet angle refers to the angle between the water flow and the horizontal line.

[0007] Furthermore, water outlet smoothness includes: obtaining an updated water flow matrix of the water outlet image and a difference matrix of the water flow matrix; and calculating the ratio of the number of element 1s in the difference matrix to the number of elements in the difference matrix to obtain the water outlet smoothness corresponding to the water outlet image.

[0008] Furthermore, the water outlet atomization area includes: filtering the water outlet image to obtain a standard image; calculating the difference matrix between the water outlet image and the standard image; threshold classifying the elements in the difference matrix; if the element value is greater than a preset pixel threshold, it is replaced with 1, otherwise it is replaced with 0 to obtain a binary matrix; calculate the ratio of the number of element 1s in the binary matrix to the number of elements in the binary matrix to obtain the water outlet atomization area.

[0009] Furthermore, a regression mapping model of turbidity and water outlet characteristics is obtained, including: constructing a feature vector of the corresponding water outlet image based on the normalized water outlet angle, water outlet stability and water outlet atomization area; initializing a training matrix of G rows for G moments of the interval operation time of a filter dust removal device, and establishing training samples for each moment based on the training matrix, specifically including: obtaining the feature vectors of the 1st to gth moments, and filling them into the 1st to gth rows of the training matrix respectively, and filling the gth to Gth rows of the training matrix based on element 0 to obtain the training sample of the gth moment; wherein 1≤g≤G; establishing sample labels for the training samples of the G moments through the turbidity at the corresponding moments, and obtaining the regression mapping model through training based on the training samples and the sample labels; wherein the regression mapping model is constructed based on a long short-term memory network and iteratively updated based on a mean square error loss function.

[0010] Furthermore, the regression mapping model is configured with an initialization unit, and the initialization unit is used to initialize the memory unit of the regression mapping model based on the interval operation time of the filter dust removal device.

[0011] Furthermore, the turbidity of the water in the target pipeline is obtained, including: The outflow images from the 1st to the kth moment in the monitoring time slice are obtained to obtain the corresponding outflow characteristics; the training matrix is ​​filled based on the normalized outflow characteristics from the 1st to the kth moment and input into the regression mapping model to obtain the turbidity at the kth moment in the monitoring time slice.

[0012] The beneficial effects of the present invention are as follows: based on the fact that when water is discharged from a pipe through a filter screen, the discharged water flow has the characteristics of an increased water flow angle, water flow vibration, and an increased atomization area of ​​downstream water bombarded by the water flow due to the nonlinear blockage of the filter screen by particles, a model is constructed to reflect the mapping relationship between turbidity and water outlet angle, water outlet stability, and water outlet atomization area, thereby quantitatively analyzing turbidity based on the water outlet image. Compared with the prior art, the present invention reduces the detection cost of water turbidity, has low maintenance requirements, and has accurate measurement results. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a module diagram of a water quality detection system for water conservancy projects in water conservancy informationization of the present invention; Figure 2 It is a schematic diagram of a usage scenario of the present invention. DETAILED DESCRIPTION

[0014] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0015] like Figure 1 As shown, a water quality detection system for water conservancy projects with water conservancy information technology includes: A data acquisition module, used to obtain characteristic data of the drainage filter unit and the target pipe within a preset time period; wherein the drainage filter unit includes a filter screen and a filter screen dust removal device, and the filter screen and the filter screen dust removal device are both installed at the water outlet of the target pipe; The characteristic data include: the water outflow image of the water outlet of the target pipeline, the interval operation time of the filter dust removal device, the water pressure of the target pipeline, and the turbidity of the water body in the target pipeline; A data processing module is used to perform semantic recognition on water-out images at fixed time intervals to identify water-out features in each water-out image; wherein the water-out features include: water-out angle, water-out stability and water-out atomization area; The correlation matching module is used to obtain the turbidity of the water body in the target pipeline at fixed time intervals; and to correlate the turbidity and the water discharge characteristics at the same time to obtain a regression mapping model of the turbidity and the water discharge characteristics; The dynamic monitoring module is used to divide the monitoring time period into several monitoring time slices based on the interval operation time of the filter dust removal device within the monitoring time period; in each monitoring time slice, the turbidity of the water body in the target pipeline is obtained based on the corresponding water outlet image and the regression mapping model.

[0016] In one embodiment of the present invention, the water-emerging image is acquired in a horizontal direction and at a fixed angle.

[0017] Specifically, it is stipulated that the water outlet image must be acquired in the horizontal direction and at a fixed angle. This ensures the consistency of angle and direction of all acquired images, which is convenient for subsequent feature extraction and analysis. Shooting in the horizontal direction and at a fixed angle can eliminate image distortion or angle deviation caused by changes in camera position or posture, and ensure that the images acquired at each moment are comparable. Key features such as water outlet angle, water outlet stability, and water outlet atomization area need to be extracted from the water outlet image. In the process of system calibration and establishing a regression mapping model, a fixed acquisition angle helps reduce external interference factors and ensure the consistency of training data.

[0018] In one embodiment of the present invention, the water outlet angle includes: performing a single-layer Haar wavelet decomposition on the water outlet image to extract the horizontal high-frequency component matrix H, the vertical high-frequency component matrix V and the diagonal high-frequency component matrix D of the water outlet image; weighting the horizontal high-frequency component matrix H, the vertical high-frequency component matrix V and the diagonal high-frequency component matrix D respectively and merging them to obtain a high-frequency feature matrix; wherein the weighting coefficients are ; Extract local fractal features from the high-frequency feature matrix, specifically including: initializing the high-frequency feature matrix, generating a convolution sliding window, the size of the convolution sliding window is N×N, and the sliding step is , the local fractal dimension in the convolution sliding window is calculated based on the box fractal algorithm to obtain a fractal dimension matrix; wherein, the element in the i-th row and j-th column of the fractal dimension matrix represents the fractal dimension of the pixel in the i-th row and j-th column of the water image; the elements in the fractal dimension matrix whose fractal dimension is greater than the preset dimension threshold are replaced with element 1, otherwise they are replaced with element 0 to obtain a water flow matrix; wherein, the elements with element value 1 in the water flow matrix represent the water flow in the water image; the element 1 in the water flow matrix is ​​detected in the eight-neighborhood, and if the element 1 exists in the eight-neighborhood of the element 1, a connection is established for the corresponding element The connected relationship is established to obtain the connected clusters in the water flow matrix; if the number of element 1 in the connected cluster is less than the preset number threshold, the element 1 in the corresponding connected cluster is replaced by element 0 to obtain the updated water flow matrix; wherein the element 1 in the updated water flow matrix corresponds to the water flow in the water outlet image; the covariance matrix of the updated water flow matrix is ​​obtained, and the eigenvalue decomposition of the covariance matrix is ​​performed to extract the main eigenvector of the covariance matrix; the arc cosine value of the ratio of the dot product of the main eigenvector and the preset standard reference vector to the module length product is calculated to obtain the water outlet angle; wherein the preset standard reference vector is expressed as , the water outlet angle refers to the angle between the water flow and the horizontal line.

[0019] Specifically, the collected water outflow image is decomposed by a single layer Haar wavelet to extract high-frequency components in different directions in the image, including horizontal (H), vertical (V) and diagonal (D) components. Thus, the texture information related to the water flow details in the water outflow image is obtained. The horizontal, vertical and diagonal high-frequency components are weighted and merged to obtain a comprehensive high-frequency feature matrix. By weighting (the weights of each direction are the same), the texture information in each direction is balanced so that the constructed feature matrix can better reflect the overall characteristics of the water flow. After obtaining the high-frequency feature matrix, a convolution window of size N×N with a fixed sliding step is used to calculate the fractal dimension in the local area using the box fractal algorithm to generate a fractal dimension matrix. Among them, the fractal dimension is used to describe the complexity of the local area and reflect the texture and morphological characteristics of the water flow in the image. Each element in the matrix represents the fractal dimension of the corresponding pixel area. The elements in the fractal dimension matrix that are greater than the preset threshold are replaced with 1, and those that are lower than the threshold are replaced with 0, thereby obtaining a binary water flow matrix. In the water flow matrix, the area with a value of 1 represents the possible water flow area in the water outflow image. Perform eight-neighborhood detection on the elements with a value of 1 in the water flow matrix, classify the adjacent 1s into the same connected cluster, and filter out the connected clusters with a number less than the preset threshold and set them to 0. In this way, small noise is removed and continuous and representative water flow areas are retained. The covariance matrix is ​​calculated for the updated water flow matrix, and then the eigenvalue decomposition is performed to extract the main eigenvector of the covariance matrix. Among them, the main eigenvector represents the main direction of water flow distribution. The water outflow angle is obtained by calculating the ratio of the dot product of the main eigenvector and the preset standard reference vector (representing the horizontal line direction) to the product of their respective modulus lengths, and taking the inverse cosine value. The water outflow angle reflects the deviation of the water flow from the horizontal line. The larger the value, the more obvious the deviation from the horizontal.

[0020] In one embodiment of the present invention, water outlet smoothness includes: obtaining an updated water flow matrix of the water outlet image and a difference matrix of the water flow matrix; and calculating the ratio of the number of element 1s in the difference matrix to the number of elements in the difference matrix to obtain the water outlet smoothness corresponding to the water outlet image.

[0021] Specifically, the updated water flow matrix is ​​subtracted from the water flow matrix. The area with the element value of 1 in the difference matrix represents the water flow that is not connected to the main water flow and is eliminated during the update process, indicating the instability of the water flow. It can be understood that the more branch flows there are, the more unstable the interior is.

[0022] In one embodiment of the present invention, the water outlet atomization area includes: filtering the water outlet image to obtain a standard image; calculating the difference matrix between the water outlet image and the standard image; threshold classifying the elements in the difference matrix; if the element value is greater than a preset pixel threshold, it is replaced with 1, otherwise it is replaced with 0 to obtain a binary matrix; calculate the ratio of the number of element 1s in the binary matrix to the number of elements in the binary matrix to obtain the water outlet atomization area.

[0023] Specifically, the collected water outlet image is filtered to obtain a standard image. The original water outlet image is compared with the standard image, and the difference matrix between the two is calculated. The difference matrix reflects the part of the original image that is inconsistent with the standard image. The inconsistent part may correspond to the atomization phenomenon, that is, the water mist dispersion area. A threshold judgment is performed on each element in the difference matrix, and in the obtained binary matrix, the area with a value of 1 represents an obvious atomization area in the water outlet image. The number of elements with a value of 1 in the binary matrix is ​​counted, and then the ratio is calculated with the total number of elements in the binary matrix. The ratio quantitatively reflects the size of the water outlet atomization area, that is, the proportion of the atomization area in the entire image.

[0024] In one embodiment of the present invention, a regression mapping model of turbidity and water outlet characteristics is obtained, including: constructing a feature vector of the corresponding water outlet image based on the normalized water outlet angle, water outlet stability and water outlet atomization area; initializing a training matrix of G rows for G moments of the interval operation time of a filter dust removal device, and establishing training samples for each moment based on the training matrix, specifically including: obtaining the feature vectors of the 1st to gth moments, and filling them into the 1st to gth rows of the training matrix respectively, and filling the gth to Gth rows of the training matrix based on element 0 to obtain the training sample of the gth moment; wherein 1≤g≤G; establishing sample labels for the training samples of the G moments through the turbidity at the corresponding moments, and obtaining the regression mapping model through training based on the training samples and the sample labels; wherein the regression mapping model is constructed based on a long short-term memory network and iteratively updated based on a mean square error loss function.

[0025] Specifically, a regression mapping model is constructed so that the system can predict the turbidity of the water body in the target pipe based on the features extracted from the water discharge image (water discharge angle, water discharge stability and water discharge atomization area). First, the extracted water discharge angle, water discharge stability and water discharge atomization area are normalized to ensure that the features are in the same order of magnitude, so as to avoid the influence of model training due to the difference in feature scales. The normalized features are combined into a multidimensional feature vector, which can fully describe the water flow characteristics of each water discharge image. For G moments in the interval operation time of a filter dust removal device, a training matrix with G rows is initialized. Each row is expected to store the feature data of one moment. For any moment g (1≤g≤G), the normalized feature vectors from moment 1 to moment g are first filled into the 1st to gth rows of the training matrix in sequence. The remaining rows from g+1th to Gth are filled with elements 0 to form training samples of uniform size. A corresponding relationship is established between each training sample and the water turbidity value at the corresponding moment as the label of the sample. Using the above training samples and the corresponding turbidity labels, a regression mapping model is constructed through a long short-term memory network (LSTM). LSTM is suitable for processing training data with time series characteristics and can capture dynamic changes in the filter dust removal device during its operation cycle. The mean square error (MSE) is used as the loss function, and the model is gradually optimized by iteratively updating the network weights and biases so that it can accurately map the relationship between the outlet water characteristics and turbidity. By constructing a training matrix, the outlet water characteristics at multiple times are combined, making full use of the time series characteristics of the data and better reflecting the water quality change process. Normalization processing and a unified training sample format improve the stability of model training, and the LSTM network effectively captures the time dependency in the data, thereby improving the accuracy of turbidity prediction.

[0026] It should be noted that the activation function of the hidden layer of the regression mapping model is ReLU, which is used to eliminate the influence of element 0 in the training matrix.

[0027] In one embodiment of the present invention, the regression mapping model is configured with an initialization unit, and the initialization unit is used to initialize the memory unit of the regression mapping model based on the interval operation time of the filter dust removal device.

[0028] Specifically, the initialization unit uses the prior information of the interval operation time of the filter dust removal device to set the initial state of the LSTM memory unit. By initializing the memory unit according to the device operation cycle, the LSTM can more accurately capture the periodic characteristics of water flow changes.

[0029] In one embodiment of the present invention, the turbidity of the water body in the target pipeline is obtained, including: obtaining the water outlet images from the 1st to the kth moment in the monitoring time slice to obtain the corresponding water outlet features; filling the training matrix based on the normalized water outlet features from the 1st to the kth moment, and inputting it into the regression mapping model to obtain the turbidity at the kth moment in the monitoring time slice.

[0030] Specifically, the entire monitoring cycle is divided into several "monitoring time slices". In each time slice, the system continuously obtains water outflow images at multiple moments to form a time series data. In a certain monitoring time slice, water outflow images are collected from the 1st moment to the kth moment, and the corresponding water outflow features are extracted using the aforementioned method (based on Haar wavelet decomposition, local fractal feature extraction and other technologies). In order to eliminate the influence of different dimensions between the features, the water outflow features extracted from the 1st to the kth moment are normalized. Subsequently, these normalized features are sequentially filled into a pre-initialized training matrix. The filled training matrix is ​​input into the regression mapping model based on the long short-term memory network (LSTM). The unit uses the trained model and the initial memory state set by the initialization unit to capture the time series changes of the water outflow features and output the water turbidity prediction value corresponding to the kth moment in the monitoring time slice. In this way, the system can use the accumulated water outflow feature information in the previous period in each monitoring time slice to output the water quality turbidity detection results in real time. This dynamic monitoring method based on time series data not only improves the detection accuracy, but also enables the system to have strong real-time response capabilities and can promptly reflect changes in water turbidity.

[0031] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A water quality detection system for water conservancy projects based on water conservancy information technology, characterized in that: include: A data acquisition module, used to obtain characteristic data of the drainage filter unit and the target pipe within a preset time period; wherein the drainage filter unit includes a filter screen and a filter screen dust removal device, and the filter screen and the filter screen dust removal device are both installed at the water outlet of the target pipe; The characteristic data include: the water outflow image of the water outlet of the target pipeline, the interval operation time of the filter dust removal device, the water pressure of the target pipeline, and the turbidity of the water body in the target pipeline; A data processing module is used to identify the water-out images at fixed time intervals based on physical properties to obtain water-out features in each water-out image; wherein the water-out features include: water-out angle, water-out stability and water-out atomization area; The correlation matching module is used to obtain the turbidity of the water body in the target pipeline at fixed time intervals; and to correlate the turbidity and the water discharge characteristics at the same time to obtain a regression mapping model of the turbidity and the water discharge characteristics; The dynamic monitoring module is used to divide the monitoring time period into several monitoring time slices based on the interval operation time of the filter dust removal device within the monitoring time period; in each monitoring time slice, the turbidity of the water body in the target pipeline is obtained based on the corresponding water outlet image and the regression mapping model.

2. A water quality detection system for water conservancy projects based on water conservancy informationization according to claim 1, characterized in that: The out-of-water images are acquired in the horizontal direction at a fixed angle.

3. A water quality detection system for water conservancy projects based on water conservancy informationization according to claim 2, characterized in that: The method comprises: performing a single-layer Haar wavelet decomposition on the outflow image to extract the horizontal high-frequency component matrix H, the vertical high-frequency component matrix V and the diagonal high-frequency component matrix D of the outflow image; weighting the horizontal high-frequency component matrix H, the vertical high-frequency component matrix V and the diagonal high-frequency component matrix D respectively and merging them to obtain a high-frequency feature matrix; wherein the weighting coefficients are ; Extract local fractal features from the high-frequency feature matrix, specifically including: initializing the high-frequency feature matrix, generating a convolution sliding window, the size of the convolution sliding window is N×N, and the sliding step is , the local fractal dimension in the convolution sliding window is calculated based on the box fractal algorithm to obtain a fractal dimension matrix; wherein, the element in the i-th row and j-th column of the fractal dimension matrix represents the fractal dimension of the pixel in the i-th row and j-th column of the water image; the elements in the fractal dimension matrix whose fractal dimension is greater than the preset dimension threshold are replaced with element 1, otherwise they are replaced with element 0 to obtain a water flow matrix; wherein, the elements with element value 1 in the water flow matrix represent the water flow in the water image; the element 1 in the water flow matrix is ​​detected in the eight-neighborhood, and if the element 1 exists in the eight-neighborhood of the element 1, a connection is established for the corresponding element The connected relationship is established to obtain the connected clusters in the water flow matrix; if the number of element 1 in the connected cluster is less than the preset number threshold, the element 1 in the corresponding connected cluster is replaced by element 0 to obtain the updated water flow matrix; wherein the element 1 in the updated water flow matrix corresponds to the water flow in the water outlet image; the covariance matrix of the updated water flow matrix is ​​obtained, and the eigenvalue decomposition of the covariance matrix is ​​performed to extract the main eigenvector of the covariance matrix; the arc cosine value of the ratio of the dot product of the main eigenvector and the preset standard reference vector to the module length product is calculated to obtain the water outlet angle; wherein the preset standard reference vector is expressed as , the water outlet angle refers to the angle between the water flow and the horizontal line.

4. A water quality detection system for water conservancy projects based on water conservancy informationization according to claim 3, characterized in that: The water outlet stability includes: obtaining an updated water flow matrix of the water outlet image and a difference matrix of the water flow matrix; and calculating the ratio of the number of element 1s in the difference matrix to the number of elements in the difference matrix to obtain the water outlet stability corresponding to the water outlet image.

5. A water quality detection system for water conservancy projects based on water conservancy informationization according to claim 4, characterized in that: The water outlet atomization area comprises: filtering the water outlet image to obtain a standard image; calculating a difference matrix between the water outlet image and the standard image; threshold classification of elements in the difference matrix; replacing the element value with 1 if it is greater than a preset pixel threshold, otherwise replacing it with 0 to obtain a binary matrix; calculating the ratio of the number of element 1s in the binary matrix to the number of elements in the binary matrix to obtain the water outlet atomization area.

6. A water quality detection system for water conservancy projects based on water conservancy informationization according to claim 5, characterized in that: A regression mapping model of turbidity and outlet water characteristics is obtained, including: constructing a feature vector of a corresponding outlet water image based on a normalized outlet water angle, outlet water stability and outlet water atomization area; initializing a training matrix of G rows for G moments of the interval operation time of a filter dust removal device, and establishing training samples for each moment based on the training matrix, specifically including: obtaining feature vectors from the 1st to the gth moment, and filling them into the 1st to the gth rows of the training matrix respectively, and filling the gth to the Gth rows of the training matrix based on element 0 to obtain the training sample at the gth moment; wherein 1≤g≤G; establishing sample labels for the training samples at the G moments through the turbidity at the corresponding moments, and obtaining a regression mapping model through training based on the training samples and the sample labels; wherein the regression mapping model is constructed based on a long short-term memory network and iteratively updated based on a mean square error loss function.

7. A water quality detection system for water conservancy projects based on water conservancy informationization according to claim 6, characterized in that: The regression mapping model is configured with an initialization unit, and the initialization unit is used to initialize the memory unit of the regression mapping model based on the interval operation time of the filter dust removal device.

8. A water quality detection system for water conservancy projects based on water conservancy informationization according to claim 7, characterized in that: The turbidity of the water body in the target pipeline is obtained, including: obtaining the water outlet images from the 1st to the kth moment in the monitoring time slice to obtain the corresponding water outlet features; filling the training matrix based on the normalized water outlet features from the 1st to the kth moment, and inputting it into the regression mapping model to obtain the turbidity at the kth moment in the monitoring time slice.

Images