Multi-dimensional water turbidity inversion method and system based on intelligent image recognition
The data is acquired through underwater robots and the water turbidity inversion model is constructed using convolutional neural networks, which solves the problem of large-area measurement and time continuity in traditional methods, and realizes high-resolution water turbidity measurement, especially Measurement of deep turbidity.
Patent Information
- Application Number
- CN202411038562.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional underwater turbidity measurement methods cannot meet the needs of large-area measurement and time continuity at the same time, and there are problems of discontinuity and low resolution in remote sensing satellite data, making it difficult to achieve high-resolution turbidity measurement of water bodies, especially inability to measure deep layers. Turbidity of water.
The image data and actual turbidity values are obtained through underwater robots, and the water turbidity inversion model is constructed based on the convolutional neural network to realize multi-dimensional water turbidity inversion and visualization processing. The convolutional neural network is used for feature extraction and classification, and high-quality Resolution turbidity data.
The turbidity measurement of water body with multiple degrees of freedom, continuous time and high resolution is achieved, which can invert the turbidity of water body at different depths, solving the problems of large manpower consumption and low resolution of traditional methods.
Smart Images

Figure CN118968275A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of underwater image processing, and in particular to a multi-dimensional water turbidity inversion method and system based on intelligent image recognition. Background Art
[0002] Water turbidity is one of the important physical indicators of water quality and aquatic ecosystems, and is related to the suspended matter concentration, visibility and pollution level of the water body. Underwater turbidity inversion refers to the direct or indirect acquisition of turbidity data of water bodies by optical or other means. Traditional underwater turbidity measurement methods rely on manpower or remote sensing satellites, and cannot simultaneously meet the needs of large-area measurement and time continuity. Under the dual influence of global climate change and human activities, water pollution problems are becoming increasingly serious, and there is an urgent need for an efficient and accurate underwater turbidity inversion method to support water environment protection and governance.
[0003] The water turbidity identification of the related art is carried out by using an optical backscatter turbidimeter (OBS-3A) for fixed-point measurement and using water color remote sensing satellite data for large-scale water turbidity inversion. However, when using OBS-3A to measure water turbidity identification, it relies on a fixed rack and only supports single-point measurement. It is difficult to achieve turbidity measurement in a large area of water and requires manual installation of the rack required for OBS-3A measurement, which consumes a lot of manpower. The method of inverting large-scale water turbidity using water color remote sensing satellite data has the disadvantage of discontinuous water color remote sensing satellite data. Water color remote sensing satellites move around the earth according to a predetermined trajectory, have a fixed revisit period (1 to 30 days), and are easily affected by rainy days and clouds, making it difficult to obtain continuous measurement data in the same area; in addition, the resolution of remote sensing satellite inversion turbidity depends on the spatial resolution of water color remote sensing satellite data. The spatial resolution of most remote sensing satellite data is 10m-300m, so it is difficult to achieve turbidity measurement with sub-meter resolution in a small range using remote sensing satellite inversion methods; and water color remote sensing satellites can only measure information related to the surface of the water body, and cannot measure the turbidity of deep water bodies. Therefore, related technologies find it difficult to obtain water turbidity measurement results with high resolution. Summary of the invention
[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a multi-dimensional water turbidity inversion method and system based on intelligent image recognition, which can realize multi-degree-of-freedom, time-continuous, high-resolution water turbidity measurement.
[0005] The first technical solution adopted by the present invention is: a multi-dimensional water turbidity inversion method based on intelligent image recognition, comprising the following steps:
[0006] Acquire image data of target water area and actual water turbidity value of target water area through underwater robot;
[0007] Based on convolutional neural network, a water turbidity inversion model is constructed;
[0008] Based on the image data of the target water area and the actual water turbidity value of the target water area, the water turbidity inversion model is trained to obtain a trained water turbidity inversion model;
[0009] Based on the trained water turbidity inversion model, multidimensional water turbidity inversion and visualization are performed to obtain multidimensional water turbidity inversion visualization results.
[0010] Furthermore, the step of obtaining the image data of the target water area and the actual water turbidity value of the target water area by the underwater robot specifically includes:
[0011] The underwater robot is remotely controlled by a remote control handle to move to the target water area and obtain water sample data of the target water area;
[0012] The water sample data of the target water area is photographed and processed by a camera to obtain image data of the target water area;
[0013] The water sample data of the target water area is measured and processed by a turbidity meter to obtain the actual water turbidity value of the target water area.
[0014] Furthermore, the water turbidity inversion model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a Flatten layer, a fully connected layer and an output layer, wherein the input layer, the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer, the third pooling layer, the Flatten layer, the fully connected layer and the output layer are connected in sequence.
[0015] Further, the step of training the water turbidity inversion model based on the image data of the target water area and the actual water turbidity value of the target water area to obtain the trained water turbidity inversion model specifically includes:
[0016] The image data of the target water area is input into the water turbidity inversion model to perform water turbidity measurement processing to obtain the predicted water turbidity value of the target water area;
[0017] The predicted water turbidity value of the target water area is compared with the actual water turbidity value of the target water area to obtain a comparison result;
[0018] If the comparison result does not meet the preset accuracy requirement, the image data of the target water area is re-input into the water turbidity inversion model for water turbidity measurement and comparison with the actual water turbidity value of the target water area;
[0019] Until the comparison result meets the preset accuracy requirement, the trained water turbidity inversion model is obtained.
[0020] Furthermore, the step of inputting the image data of the target water area into the water turbidity inversion model to perform water turbidity measurement processing to obtain the predicted water turbidity value of the target water area specifically includes:
[0021] Input the image data of the target water area into the water turbidity inversion model;
[0022] Based on the input layer of the water turbidity inversion model, image data of the target water area is obtained;
[0023] Based on the first convolution layer and the first pooling layer of the water turbidity inversion model, feature extraction and pooling processing are performed on the image data of the target water area to obtain image feature data of the first target water area;
[0024] Based on the second convolution layer and the second pooling layer of the water turbidity inversion model, feature extraction and pooling processing are performed on the image feature data of the first target water area to obtain the image feature data of the second target water area;
[0025] Based on the third convolution layer and the third pooling layer of the water turbidity inversion model, feature extraction and pooling processing are performed on the image feature data of the second target water area to obtain image feature data of the third target water area;
[0026] Based on the Flatten layer of the water turbidity inversion model, the image feature data of the third target water area is subjected to dimensionality reduction processing to obtain one-dimensional image feature data of the target water area;
[0027] Based on the fully connected layer of the water turbidity inversion model, the one-dimensional image feature data of the target water area is classified and processed to obtain the predicted water turbidity value of the target water area to be output;
[0028] Based on the output layer of the water turbidity inversion model, the predicted water turbidity value of the target water area to be output is output to obtain the predicted water turbidity value of the target water area.
[0029] Further, the step of performing multidimensional water turbidity inversion and visualization based on the trained water turbidity inversion model to obtain a multidimensional water turbidity inversion visualization result specifically includes:
[0030] The target water area is read frame by frame through the video stream framework of the underwater robot to obtain the video data of the target water area;
[0031] The video data of the target water area is converted into image data and input into the trained water turbidity inversion model to perform multi-dimensional water turbidity inversion to obtain the water inversion turbidity value;
[0032] The inverted turbidity values of water bodies are visualized to obtain multi-dimensional water turbidity inversion visualization results.
[0033] The second technical solution adopted by the present invention is: a multi-dimensional water turbidity inversion system based on intelligent image recognition, comprising:
[0034] An acquisition module is used to acquire image data of a target water area and an actual water turbidity value of the target water area through an underwater robot;
[0035] A construction module for constructing a water turbidity inversion model based on a convolutional neural network;
[0036] A training module is used to train the water turbidity inversion model based on the image data of the target water area and the actual water turbidity value of the target water area to obtain a trained water turbidity inversion model;
[0037] The inversion visualization module is used to perform multi-dimensional water turbidity inversion and visualization based on the trained water turbidity inversion model to obtain multi-dimensional water turbidity inversion visualization results.
[0038] The beneficial effects of the method and system of the present invention are as follows: the present invention obtains image data of the target water area and the actual water turbidity value of the target water area through an underwater robot, and can realize underwater multi-degree-of-freedom large-scale navigation measurement. Further, based on a convolutional neural network, a water turbidity inversion model is constructed, and continuous measurement of turbidity data can be realized, and then multi-dimensional water turbidity inversion and visualization processing are performed, thereby inverting the turbidity of the water body at different depths, and realizing multi-degree-of-freedom, time-continuous, and high-resolution water turbidity measurement. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 It is a flow chart of the steps of a multi-dimensional water turbidity inversion method based on intelligent image recognition of the present invention;
[0040] Figure 2 It is a structural block diagram of a multi-dimensional water turbidity inversion system based on intelligent image recognition of the present invention;
[0041] Figure 3 It is a structural schematic diagram of a water turbidity inversion model provided in a specific embodiment of the present invention. DETAILED DESCRIPTION
[0042] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.
[0043] In the related technology, for example, an optical backscatter turbidimeter (OBS-3A) is used for fixed-point measurement: OBS-3A is an instrument that uses the infrared scattering principle to measure turbidity. When using OBS-3A to measure water turbidity, it is usually necessary to install a fixed frame in the water area to be measured, and fix OBS-3A on the frame to measure the turbidity of the water body; if turbidity measurements at different depths are to be performed, the OBS-3A must be suspended by a ship-borne winch, and lowered to a certain water depth by shaking the winch; another example is the use of water color remote sensing satellite data for large-scale water turbidity inversion: the use of water color remote sensing satellite data for large-scale water turbidity inversion is a method of inverting water turbidity by analyzing the spectral characteristics of water reflection using multispectral or hyperspectral images taken by remote sensing satellites. When using this method to invert water turbidity, it is first necessary to collect satellite data covering the study area, then pre-process the data through certain methods (such as atmospheric correction, cloud removal, etc.), and finally establish an inversion relationship between satellite data and water turbidity through field measurements of water turbidity, and finally obtain large-area surface water turbidity inversion data.
[0044] But it has some shortcomings, such as:
[0045] 1) The technology of using OBS-3A to measure water turbidity relies on a fixed frame and only supports single-point measurement. It is difficult to measure turbidity in a large area of water. The frame required for OBS-3A measurement needs to be installed manually, which consumes a lot of manpower.
[0046] 2) The technology of using water color remote sensing satellite data for large-scale water turbidity inversion has the problem that water color remote sensing satellite data is discontinuous. Water color remote sensing satellites orbit the earth according to a predetermined trajectory and have a fixed revisit period (1 to 30 days). They are also easily affected by rainy days and clouds, making it difficult to obtain continuous measurement data in the same area. The turbidity resolution of remote sensing satellite inversion depends on the spatial resolution of water color remote sensing satellite data. Most remote sensing satellite data has a spatial resolution of 10m-300m. Therefore, it is difficult to use remote sensing satellite inversion methods for small-scale sub-meter resolution turbidity measurements. In addition, water color remote sensing satellites can only measure information related to the surface of water bodies, and cannot measure turbidity in deep water bodies.
[0047] Based on this, refer to Figure 1 The present invention provides a multi-dimensional water turbidity inversion method based on intelligent image recognition, the method comprising the following steps:
[0048] S100, acquiring image data of a target water area and an actual water turbidity value of the target water area by an underwater robot;
[0049] Specifically, the underwater robot is remotely controlled by a remote control handle to control the underwater robot to move to the target water area and obtain water sample data of the target water area; the water sample data of the target water area is photographed and processed by a camera to obtain image data of the target water area; the water turbidity of the water sample data of the target water area is measured and processed by a turbidity meter to obtain the actual water turbidity value of the target water area.
[0050] In this embodiment, BlueRov is controlled to move to a specified depth position of the water body to be tested; BlueRov is controlled to take underwater photos of the current position; water samples in front of the BlueRov camera at the current position are collected; the turbidity value of the collected water sample is measured using a 2100Q portable turbidity meter; BlueRov is controlled to move to different positions of the water body, and the steps of taking underwater photos and measuring turbidity are repeated to collect underwater photos and turbidity data at least at 100 different positions.
[0051] Among them, it should be noted that BlueRov is an open source underwater robot developed by Blue Robotics. The robot is equipped with a high-definition camera that can transmit underwater video images in real time. At the same time, its open source nature makes it possible for users to customize and expand its data. Users can achieve communication and data transmission with BlueRov based on the MAVLink protocol. The MAVLink protocol is a lightweight communication protocol that is widely used in drone systems to achieve data transmission between drones and ground control stations. Based on the MAVLink protocol, users can send real-time control commands through the ground control station to achieve precise control of BlueRov, including forward and backward, diving and floating, etc. In addition, the MAVLink protocol can also be used to achieve real-time transmission of BlueRov data, including the navigation depth and speed of BlueRov. The Gstreamer video streaming framework is an open source multimedia framework for building various multimedia applications, such as video editing, audio processing, streaming media transmission, etc. The Gstreamer video streaming framework and the MAVLink protocol can also be used to achieve real-time transmission, frame-by-frame reading and processing of BlueRov underwater videos.
[0052] S200, based on convolutional neural network, build a water turbidity inversion model;
[0053] Specifically, Figure 3As shown, the water turbidity inversion model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a Flatten layer, a fully connected layer and an output layer, wherein the input layer, the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer, the third pooling layer, the Flatten layer, the fully connected layer and the output layer are connected in sequence.
[0054] In an embodiment of the present invention, a convolutional neural network is constructed using Python programming software, and the architecture of the convolutional neural network is as follows: the input layer processes underwater images of a size of 128x128x3. The network comprises three convolutional layers, each of which is followed by a pooling layer. The number of convolution kernels of the convolutional layers is 32, 64, and 128, respectively, and the convolution kernel size is 3x3, using the ReLU activation function. The window size of the pooling layer is 2x2. After being processed by three convolutional layers and the pooling layer, the image features are flattened into a one-dimensional vector through the Flatten layer, and then enter a fully connected layer containing 128 neurons, which also uses the ReLU activation function. Finally, the network generates an output representing the turbidity value of the water body through an output layer containing 1 neuron, and the output layer uses a linear activation function.
[0055] S300, training a water turbidity inversion model based on the image data of the target water area and the actual water turbidity value of the target water area to obtain a trained water turbidity inversion model;
[0056] Specifically, the image data of the target water area is input into the water turbidity inversion model for water turbidity measurement and processing to obtain the predicted water turbidity value of the target water area; the predicted water turbidity value of the target water area is compared with the actual water turbidity value of the target water area to obtain a comparison result; if the comparison result does not meet the preset accuracy requirement, the image data of the target water area is re-input into the water turbidity inversion model for water turbidity measurement and processing and compared with the actual water turbidity value of the target water area; until the comparison result meets the preset accuracy requirement, the trained water turbidity inversion model is obtained.
[0057] Among them, it should be noted that the image data of the target water area is input into the water turbidity inversion model; based on the input layer of the water turbidity inversion model, the image data of the target water area is obtained; based on the first convolution layer and the first pooling layer of the water turbidity inversion model, the image data of the target water area is subjected to feature extraction and pooling processing to obtain the image feature data of the first target water area; based on the second convolution layer and the second pooling layer of the water turbidity inversion model, the image feature data of the first target water area is subjected to feature extraction and pooling processing to obtain the image feature data of the second target water area; based on the third convolution layer and the third pooling layer of the water turbidity inversion model, the image feature data of the first target water area is subjected to feature extraction and pooling processing to obtain the image feature data of the second target water area. , perform feature extraction and pooling processing on the image feature data of the second target water area to obtain the image feature data of the third target water area; based on the Flatten layer of the water turbidity inversion model, perform dimensionality reduction processing on the image feature data of the third target water area to obtain the one-dimensional image feature data of the target water area; based on the fully connected layer of the water turbidity inversion model, perform classification processing on the one-dimensional image feature data of the target water area to obtain the predicted water turbidity value of the target water area to be output; based on the output layer of the water turbidity inversion model, output the predicted water turbidity value of the target water area to be output to obtain the predicted water turbidity value of the target water area.
[0058] In an embodiment of the present invention, the collected underwater photos and turbidity data are matched one by one at the same position, and are divided into a training set and a validation set at a ratio of 7:3. The underwater photos of the training set are used as the input layer, and the turbidity of the training set is used as the output layer to train the convolutional neural network; the underwater photos of the validation set are used as the input layer and input into the trained convolutional neural network, and the output turbidity value is compared with the turbidity value corresponding to the validation set to evaluate the prediction accuracy of the convolutional neural network. If the accuracy of the convolutional neural network is low, it is necessary to increase or collect training samples again; when the prediction accuracy of the convolutional neural network meets the user's needs, the turbidity inversion model is constructed. At this time, the model weights, architecture, and configuration information are stored in the HDF5 file.
[0059] Among them, convolutional neural network (CNN) is a deep learning model that is widely used in the fields of image recognition and computer vision. CNN is mainly composed of convolutional layers, pooling layers, and fully connected layers. The convolutional layer extracts local features of the image through convolution operations, and the pooling layer reduces the size of the feature map by downsampling, thereby reducing computational complexity and overfitting risks. The fully connected layer integrates the extracted features and outputs the final classification results. The advantage of CNN is that it can capture information of different scales and different levels of abstraction in the image, thus performing well in image classification and monitoring tasks.
[0060] S400, performing multi-dimensional water turbidity inversion and visualization based on the trained water turbidity inversion model to obtain a multi-dimensional water turbidity inversion visualization result.
[0061] Specifically, the target water area is read frame by frame through the video stream framework of the underwater robot to obtain the video data of the target water area; the video data of the target water area is converted into image data and input into the trained water turbidity inversion model for multi-dimensional water turbidity inversion to obtain the water inversion turbidity value; the water inversion turbidity value is visualized to obtain the multi-dimensional water turbidity inversion visualization result.
[0062] In an embodiment of the present invention, firstly, based on the Mavlink unmanned aerial vehicle communication protocol, the pymavlink library and pygame library provided by Python are used, and Python programming software is used to realize the remote control handle to control the multi-degree-of-freedom movement of BlueRov; BlueRov is placed in the water area to be tested, and the remote control handle is used to control BlueRov to dive and move to a specified position at a specified depth. Secondly, the Gstreamer video stream framework is used to transmit and read the BlueRov video data frame by frame in real time; the BlueRov video frame is input as a picture into the turbidity inversion model to obtain the inverted turbidity value. Finally, the user interface is designed using Python programming software; the current depth of BlueRov is read in real time based on the Mavlink communication protocol, and displayed in the user interface; the Gstreamer video stream framework is used to transmit the BlueRov video data in real time and display it in the user interface; the inverted turbidity value is displayed in real time in the user interface.
[0063] In summary, the embodiments of the present invention have the following advantages:
[0064] 1) The present invention uses the BlueRov underwater robot as a measurement carrier, which can realize underwater multi-degree-of-freedom large-range navigation measurement, and solves the shortcoming of only single-point measurement when using OBS-3A measurement; the present invention only needs to use a control handle to control the movement of BlueRov in the water, gets rid of the dependence on a fixed frame, and greatly saves manpower.
[0065] 2) The present invention transmits BlueRov video data in real time through the Gstreamer video stream framework and reads it frame by frame, and uses a convolutional neural network to invert turbidity data, thereby realizing continuous measurement of turbidity data; the present invention uses BlueRov camera video as an inversion basis, and the inverted area is located in the water mass area in front of the camera. The data resolution depends on the camera frame rate, which is much higher than the turbidity data obtained by inverting remote sensing data; the measurement carrier BlueRov of the present invention can dive to a specified depth of the water area according to user requirements, thereby inverting the turbidity of the water body at different depths.
[0066] Reference Figure 2 , a multi-dimensional water turbidity inversion system based on intelligent image recognition, comprising:
[0067] An acquisition module 201 is used to acquire image data of a target water area and an actual water turbidity value of the target water area through an underwater robot;
[0068] A construction module 202 is used to construct a water turbidity inversion model based on a convolutional neural network;
[0069] The training module 203 is used to train the water turbidity inversion model based on the image data of the target water area and the actual water turbidity value of the target water area to obtain a trained water turbidity inversion model;
[0070] The inversion visualization module 204 is used to perform multi-dimensional water turbidity inversion and visualization based on the trained water turbidity inversion model to obtain a multi-dimensional water turbidity inversion visualization result.
[0071] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0072] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.
Claims
1. A multi-dimensional water turbidity inversion method based on intelligent image recognition, characterized in that: The following steps are involved: Acquire image data of target water area and actual water turbidity value of target water area through underwater robot; Based on convolutional neural network, a water turbidity inversion model is constructed; Based on the image data of the target water area and the actual water turbidity value of the target water area, the water turbidity inversion model is trained to obtain a trained water turbidity inversion model; Based on the trained water turbidity inversion model, multidimensional water turbidity inversion and visualization are performed to obtain multidimensional water turbidity inversion visualization results.
2. According to claim 1, a multi-dimensional water turbidity inversion method based on intelligent image recognition is characterized in that: The step of obtaining the image data of the target water area and the actual water turbidity value of the target water area by the underwater robot specifically includes: The underwater robot is remotely controlled by a remote control handle to move to the target water area and obtain water sample data of the target water area; The water sample data of the target water area is photographed and processed by a camera to obtain image data of the target water area; The water sample data of the target water area is measured and processed by a turbidity meter to obtain the actual water turbidity value of the target water area.
3. According to claim 2, a multi-dimensional water turbidity inversion method based on intelligent image recognition is characterized in that: The water turbidity inversion model includes an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a Flatten layer, a fully connected layer and an output layer, wherein the input layer, the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer, the third pooling layer, the Flatten layer, the fully connected layer and the output layer are connected in sequence.
4. According to claim 3, a multi-dimensional water turbidity inversion method based on intelligent image recognition is characterized in that: The step of training the water turbidity inversion model based on the image data of the target water area and the actual water turbidity value of the target water area to obtain the trained water turbidity inversion model specifically includes: The image data of the target water area is input into the water turbidity inversion model to perform water turbidity measurement processing to obtain the predicted water turbidity value of the target water area; The predicted water turbidity value of the target water area is compared with the actual water turbidity value of the target water area to obtain a comparison result; If the comparison result does not meet the preset accuracy requirement, the image data of the target water area is re-input into the water turbidity inversion model for water turbidity measurement and comparison with the actual water turbidity value of the target water area; Until the comparison result meets the preset accuracy requirement, the trained water turbidity inversion model is obtained.
5. According to claim 4, a multi-dimensional water turbidity inversion method based on intelligent image recognition is characterized in that: The step of inputting the image data of the target water area into the water turbidity inversion model to perform water turbidity measurement processing to obtain the predicted water turbidity value of the target water area specifically includes: Input the image data of the target water area into the water turbidity inversion model; Based on the input layer of the water turbidity inversion model, image data of the target water area is obtained; Based on the first convolution layer and the first pooling layer of the water turbidity inversion model, feature extraction and pooling processing are performed on the image data of the target water area to obtain image feature data of the first target water area; Based on the second convolution layer and the second pooling layer of the water turbidity inversion model, feature extraction and pooling processing are performed on the image feature data of the first target water area to obtain the image feature data of the second target water area; Based on the third convolution layer and the third pooling layer of the water turbidity inversion model, feature extraction and pooling processing are performed on the image feature data of the second target water area to obtain image feature data of the third target water area; Based on the Flatten layer of the water turbidity inversion model, the image feature data of the third target water area is subjected to dimensionality reduction processing to obtain one-dimensional image feature data of the target water area; Based on the fully connected layer of the water turbidity inversion model, the one-dimensional image feature data of the target water area is classified and processed to obtain the predicted water turbidity value of the target water area to be output; Based on the output layer of the water turbidity inversion model, the predicted water turbidity value of the target water area to be output is output to obtain the predicted water turbidity value of the target water area.
6. According to claim 5, a multi-dimensional water turbidity inversion method based on intelligent image recognition is characterized in that: The step of performing multidimensional water turbidity inversion and visualization based on the trained water turbidity inversion model to obtain a multidimensional water turbidity inversion visualization result specifically includes: The target water area is read frame by frame through the video stream framework of the underwater robot to obtain the video data of the target water area; The video data of the target water area is converted into image data and input into the trained water turbidity inversion model to perform multi-dimensional water turbidity inversion to obtain the water inversion turbidity value; The inverted turbidity values of water bodies are visualized to obtain multi-dimensional water turbidity inversion visualization results.
7. A multi-dimensional water turbidity inversion system based on intelligent image recognition, characterized in that: Includes the following modules: An acquisition module is used to acquire image data of a target water area and an actual water turbidity value of the target water area through an underwater robot; A construction module for constructing a water turbidity inversion model based on a convolutional neural network; A training module is used to train the water turbidity inversion model based on the image data of the target water area and the actual water turbidity value of the target water area to obtain a trained water turbidity inversion model; The inversion visualization module is used to perform multi-dimensional water turbidity inversion and visualization based on the trained water turbidity inversion model to obtain multi-dimensional water turbidity inversion visualization results.
Citation Information
Patent Citations
River and lake water body intelligent monitoring method, device and system and readable storage medium
CN112067517A
Water turbidity detection method based on deep regression network
CN115063596A
Cosmetic Composition for Atopic Dermatitis Improvement and Treatment and Method the Same
KR1020250083251A
Cited By
Water turbidity identification method and system based on deep learning
CN120182778A
