A full-process risk warning system for pure water production

Abstract

The present invention discloses a full-process risk early warning system for pure water production, which belongs to the technical field of water body detection, and specifically comprises: a database, which is used to store water body images of various water body impurities; a data acquisition module, which is used to collect real-time water body images and water body samples in different areas; an image recognition module, which is used to build a water body impurity recognition model based on deep learning and the database, input the real-time water body image into the water body impurity recognition model after preprocessing, and identify impurities and impurity types contained in the real-time water body image; an accuracy improvement module, which is used to select real-time water body images that cannot be recognized by the water body impurity recognition model, select water body samples in corresponding areas, detect the water body samples, and judge the impurities and impurity types in the corresponding areas according to the detection data of the water body samples; the present invention detects impurities in pure water production based on big data and real-time detection.

Description

Technical Field

[0001] The present invention relates to the technical field of water body monitoring, and in particular to a full-process risk early warning system for pure water production. Background Art

[0002] In the pure water production process, impurities in the water have a significant impact on the water quality. These impurities may include suspended particles, microorganisms, chemicals, etc. Their presence not only affects the taste and safety of pure water, but may also cause blockage and corrosion of production equipment. Therefore, it is crucial to detect impurities in water in a timely and accurate manner.

[0003] Traditional methods for detecting impurities in water bodies mainly rely on manual sampling and laboratory analysis. This method requires professionals to go to the site to collect samples and then send the samples to the laboratory for analysis. The whole process is time-consuming and labor-intensive, and real-time monitoring cannot be achieved. In addition, since there may be errors in the sampling process, the final test results may not be accurate enough.

[0004] With the development of computer vision and deep learning technology, water impurity detection methods based on image recognition have gradually attracted attention. This method uses a camera to collect water images, and then uses a deep learning algorithm to identify and classify impurities in the image. Compared with traditional methods, image recognition methods have the advantages of strong real-time performance and high degree of automation.

[0005] However, existing image recognition methods still have certain limitations when dealing with complex water environments. First, since there are many types and shapes of impurities in water, it is very challenging to build a model that can accurately identify various impurities. Secondly, existing image recognition methods often have difficulty processing large amounts of data. When faced with large-scale water monitoring tasks, it may take a long time and a lot of computing resources. In addition, some methods also need to be improved in recognition accuracy and may miss or misdetect some impurities. Summary of the invention

[0006] The purpose of the present invention is to provide a full-process risk early warning system for pure water production to solve the following technical problems:

[0007] Existing image recognition methods still have certain limitations when dealing with complex water environments.

[0008] The purpose of the present invention can be achieved through the following technical solutions:

[0009] A full-process risk early warning system for pure water production, comprising:

[0010] A database for storing water images of various water impurities;

[0011] Data acquisition module, used to collect real-time water images and water samples in different areas;

[0012] An image recognition module, used to construct a water body impurity recognition model based on deep learning and the database, input the real-time water body image into the water body impurity recognition model after preprocessing, and recognize impurities and impurity types contained in the real-time water body image;

[0013] The accuracy improvement module is used to select real-time water images that cannot be identified by the water impurity recognition model, select water samples in the corresponding area, detect the water samples, and determine the impurities and impurity types in the corresponding area based on the detection data of the water samples.

[0014] As a further solution of the present invention: the process of preprocessing of the image recognition module is:

[0015] The real-time water body image is gray-processed to obtain a gray-scale image, and the gray-scale image is denoised. The Sobel edge detection operator is used to perform edge detection on the gray-scale image after denoising, and the gray-scale image after edge detection is binarized to obtain a binary image. After the binarization, the gray-scale value of the impurity is 255, and the gray-scale value of the water body is 0. The binary image is denoised again.

[0016] As a further solution of the present invention: in the accuracy improvement module, impurities and impurity types are identified based on the detection data, and the impurity types and corresponding real-time water body images are input into the water body impurity recognition model for training and reverse updating.

[0017] As a further solution of the present invention: the process of constructing the model by the image recognition module is:

[0018] The water impurity recognition model is based on an RNN network and a CNN network. The RNN network includes an LSTM layer. Several fully connected layers are arranged at the back end of the RNN network. When each node of the previous connection layer is fully forward connected to each node of the connection layer, the two connection layers are fully connected. The CNN network includes several CNN layers. Each CNN layer generates a multidimensional array. The dimension of the image of the matrix is ​​reduced, but a new dimension is generated at the same time. The size of the new dimension is equal to the number of filters applied to the image. Successive CNN layers reduce the image size and increase the size of the newly generated size. The tensor in the CNN network is converted into a vector, and the vector is used as the input of the final fully connected layer.

[0019] As a further solution of the present invention: the water impurity identification model also includes:

[0020] Keeping the dimensions associated with the filters unchanged, flattening is performed on the time dimension and the feature vector, reshaping the final tensor of the chained CNN layer into a matrix, and using the matrix as the input to the LSTM layer of the RNN network. The values ​​produced by the filters of the last CNN layer are equivalent to the feature vectors, and the flattened direction produced by the reshaping operation is used as the time dimension required by the LSTM layer.

[0021] As a further solution of the present invention: the accuracy improvement module also includes: marking the area containing impurities exceeding a preset threshold as an abnormal area, and when the proportion of the number of abnormal areas to the overall water area is greater than the preset threshold, a prompt is given.

[0022] Beneficial effects of the present invention:

[0023] By introducing a full-process risk warning system that combines image processing and actual water sample data analysis, the present invention can significantly improve the efficiency and real-time performance of impurity monitoring in the pure water production industry, enhance recognition accuracy, reduce operating costs, and optimize decision support through more comprehensive water quality monitoring, while improving product quality and safety, and promoting sustainable management of water resources and environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The present invention will be further described below in conjunction with the accompanying drawings.

[0025] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0026] 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.

[0027] See also Figure 1 As shown, the present invention is a full-process risk early warning system for pure water production, comprising:

[0028] A database for storing water images of various water impurities;

[0029] Data acquisition module, used to collect real-time water images and water samples in different areas;

[0030] An image recognition module, used to construct a water body impurity recognition model based on deep learning and the database, input the real-time water body image into the water body impurity recognition model after preprocessing, and recognize impurities and impurity types contained in the real-time water body image;

[0031] The accuracy improvement module is used to select real-time water images that cannot be identified by the water impurity recognition model, select water samples in the corresponding area, detect the water samples, and determine the impurities and impurity types in the corresponding area based on the detection data of the water samples.

[0032] Through the image recognition module, the system can monitor the water status in real time and automatically identify the type of impurities, significantly improving detection efficiency and reducing reliance on manual sampling and laboratory analysis.

[0033] Combined with deep learning technology, especially the application of RNN and CNN networks, the system's accuracy in identifying water impurities has been significantly improved, and it can handle more complex and changeable water environments.

[0034] The precision improvement module selects difficult-to-identify water images and samples for detection, and uses actual data to continuously train and optimize the water impurity recognition model, thereby further improving the system's recognition accuracy.

[0035] The system design allows for the effective processing of large amounts of data, adapting to large-scale water monitoring tasks and ensuring monitoring and detection capabilities when facing large-flow water bodies.

[0036] The system can not only identify impurities, but also evaluate the quantity and distribution of impurities, mark and prompt abnormal areas, thus providing an effective risk warning for pure water production.

[0037] In another preferred embodiment of the present invention, the image recognition module preprocessing process is:

[0038] The real-time water body image is gray-processed to obtain a gray-scale image, and the gray-scale image is denoised. The Sobel edge detection operator is used to perform edge detection on the gray-scale image after denoising, and the gray-scale image after edge detection is binarized to obtain a binary image. After the binarization, the gray-scale value of the impurity is 255, and the gray-scale value of the water body is 0. The binary image is denoised again.

[0039] In the accuracy improvement module in another preferred embodiment of the present invention, impurities and impurity types are identified based on detection data, and the impurity types and corresponding real-time water body images are input into the water body impurity recognition model for training and reverse updating.

[0040] In another preferred embodiment of the present invention, the process of constructing the model by the image recognition module is:

[0041] The water impurity recognition model is based on an RNN network and a CNN network. The RNN network includes an LSTM layer. Several fully connected layers are arranged at the back end of the RNN network. When each node of the previous connection layer is fully forward connected to each node of the connection layer, the two connection layers are fully connected. The CNN network includes several CNN layers. Each CNN layer generates a multidimensional array. The dimension of the image of the matrix is ​​reduced, but a new dimension is generated at the same time. The size of the new dimension is equal to the number of filters applied to the image. Successive CNN layers reduce the image size and increase the size of the newly generated size. The tensor in the CNN network is converted into a vector, and the vector is used as the input of the final fully connected layer.

[0042] In another preferred embodiment of the present invention, the water impurity identification model further includes:

[0043] Keeping the dimensions associated with the filters unchanged, flattening is performed on the time dimension and the feature vector, reshaping the final tensor of the chained CNN layer into a matrix, and using the matrix as the input to the LSTM layer of the RNN network. The values ​​produced by the filters of the last CNN layer are equivalent to the feature vectors, and the flattened direction produced by the reshaping operation is used as the time dimension required by the LSTM layer.

[0044] In another preferred embodiment of the present invention, the accuracy improvement module also includes: marking the area containing impurities exceeding a preset threshold as an abnormal area, and giving a prompt when the proportion of the number of abnormal areas to the overall water area is greater than the preset threshold.

[0045] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A full-process risk warning system for pure water production, characterized in that: include: A database for storing water images of various water impurities; Data acquisition module, used to collect real-time water images and water samples in different areas; An image recognition module is used to construct a water body impurity recognition model based on deep learning and the database, input the real-time water body images of different regions into the water body impurity recognition model after preprocessing, and recognize impurities and impurity types contained in the real-time water body images of different regions; The precision improvement module is used to select real-time water images of different areas that cannot be identified by the water impurity recognition model, select water samples of the corresponding areas, detect the water samples, and determine the impurities and impurity types of the corresponding areas based on the detection data of the water samples; The process of constructing the model of the image recognition module is as follows: The water impurity recognition model is based on an RNN network and a CNN network. The RNN network includes an LSTM layer. Several fully connected layers are arranged at the back end of the RNN network. When each node of the previous connection layer is fully forward connected to each node of the connection layer, the two connection layers are fully connected. The CNN network includes several CNN layers. Each CNN layer generates a multidimensional array. The dimension of the image of the matrix is ​​reduced, but a new dimension is generated at the same time. The size of the new dimension is equal to the number of filters applied to the image. Successive CNN layers reduce the image size and increase the size of the newly generated size. The tensor in the CNN network is converted into a vector, and the vector is used as the input of the final fully connected layer. The water impurity identification model also includes: Keeping the dimensions associated with the filters unchanged, flattening is performed on the time dimension and the feature vector, reshaping the final tensor of the chained CNN layer into a matrix, and using the matrix as the input to the LSTM layer of the RNN network. The values ​​produced by the filters of the last CNN layer are equivalent to the feature vectors, and the flattened direction produced by the reshaping operation is used as the time dimension required by the LSTM layer.

2. A full-process risk warning system for pure water production according to claim 1, characterized in that: The image recognition module preprocessing process is as follows: The real-time water body images in different areas are gray-processed to obtain gray-scale images, and the gray-scale images are denoised. The Sobel edge detection operator is used to perform edge detection on the denoised gray-scale images, and the gray-scale images after edge detection are binarized to obtain a binary image. After the binarization, the gray-scale value of the impurity is 255, and the gray-scale value of the water body is 0. The binarized image is denoised again.

3. A full-process risk warning system for pure water production according to claim 1, characterized in that: In the accuracy improvement module, impurities and impurity types are identified based on the detection data, and the impurity types and corresponding real-time water body images of different areas are input into the water body impurity recognition model for training and reverse updating.

4. A full-process risk warning system for pure water production according to claim 1, characterized in that: The accuracy improvement module also includes: marking the area containing impurities exceeding a preset threshold as an abnormal area, and giving a prompt when the proportion of the number of the abnormal area to the overall water area is greater than the preset threshold.

Images