A rainfall intensity measurement method based on convolutional long short memory neural network and residual convolutional neural network

Abstract

The application provides a rainfall measurement method based on a residual convolutional neural network and a convolutional long short memory neural network, and the rainfall intensity in a video is measured through a machine learning method, and belongs to the technical field of hydrology and meteorology. The method comprises the following steps: establishing a video dataset based on a photographed rainfall video and rainfall intensity information; extracting the space-time features of rainfall data in the rainfall video by using a ResNet machine learning model; and converting the space-time features of the rainfall data in the video into rainfall intensity information based on a Conv-LSTM machine learning model. The model is trained and tested in a research area, the optimal model parameters are obtained, and the rainfall intensity information of the research area is obtained.

Description

Technical Field

[0001] This invention relates to the field of hydrology and meteorology, and in particular to a method for measuring rainfall intensity based on residual convolutional neural networks and convolutional long short-term memory neural networks. Background Technology

[0002] Precipitation is a crucial meteorological variable in many fields, including meteorology, hydrology, and agriculture. Ground observation, radar estimation, and satellite inversion are currently the main methods for acquiring precipitation data. Traditional ground observation is the most direct and effective method, and the precipitation data obtained from it is widely recognized as the most reliable, accurately capturing precipitation near the observation point. However, limitations imposed by factors such as cost and topography restrict the spatial distribution and coverage of ground observation stations, failing to fully reflect the spatial distribution of precipitation. Video observation, as an emerging method for acquiring precipitation data, can cover the observation area at a lower cost, acquire precipitation data in real time, and fully reflect the spatial distribution of precipitation in the observation area.

[0003] Deep learning, a new generation of artificial neural networks, has had a significant impact on industrial production, daily life, and various disciplines in recent years. It represents a major advancement in the ability of neural networks to automatically capture relevant features and complex data distributions, and is considered an indispensable tool for the future. Currently, no research has yet used the coupling of residual convolutional neural networks and convolutional long short-term memory neural networks to estimate rainfall from video observations. Summary of the Invention

[0004] Technical Problem: The purpose of this invention is to propose a rainfall intensity measurement method based on convolutional long short memory neural networks and residual convolutional neural networks, so as to improve the accuracy of rainfall data through machine learning methods.

[0005] Technical Solution: This invention discloses a method for measuring rainfall intensity based on convolutional long short memory neural networks and residual convolutional neural networks. The method includes the following steps:

[0006] Step 1: Obtain rainfall intensity data for a certain period of time in the study area, and preprocess the rainfall intensity data;

[0007] Step 2: Capture rainfall videos of the study area, and then filter and preprocess the rainfall videos;

[0008] Step 3: Set an appropriate video duration, match the rainfall intensity data with the rainfall videos, and build the grayscale video dataset required for deep learning across the entire time series; the dataset is divided into training and validation sets in a 2:1 ratio;

[0009] Step 4: Build a deep learning framework and construct a model for measuring rainfall intensity based on spatiotemporal correlation; use the Residual Convolutional Neural Network (ResNet) method to extract rainfall feature information from the video, and use the Convolutional Long Short Memory Neural Network (Conv-LSTM) method to convert the rainfall feature information into rainfall intensity;

[0010] Step 5: Input the rainfall video data and rainfall intensity data in the dataset into the deep learning model for training. Based on the error between the model output and the true value, the deep learning neural network will update the model parameters through the backpropagation algorithm until the loss error is minimized. By comprehensively comparing the errors during the training period and the validation period, the optimal model is selected.

[0011] Step 6: Input the rainfall video data of the study area for the remaining time into the optimal model in Step 5 to obtain the rainfall intensity information of the study area at other times.

[0012] Furthermore, the preprocessing includes handling missing values ​​and outliers.

[0013] Furthermore, the preprocessing in step 1 includes removing outliers from ground-observed rainfall intensity data and selecting remote sensing rainfall data with appropriate spatial resolution.

[0014] Furthermore, the filtering in step 2 includes removing videos whose clarity does not meet the requirements and removing videos with more than a preset number of obstructions.

[0015] Furthermore, the preprocessing in step 2 includes converting the selected rainfall videos to grayscale and segmenting them into short videos of a preset duration.

[0016] Furthermore, the dataset in step 3 consists of rainfall videos of the study area at a certain moment and rainfall intensity data corresponding to the time period during which the videos were taken.

[0017] Furthermore, the Residual Convolutional Neural Network (ResNet) method in step 4 extracts rainfall feature information from the video through convolution calculations in the neural network and residual structures.

[0018] Furthermore, the Convolutional Long Short Memory Neural Network (Conv-LSTM) method in step 4 processes the received spatial features of rainfall through convolution calculations in the neural network and processes the received temporal features of rainfall through the gate structure of LSTM.

[0019] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0020] The rainfall intensity measurement method proposed in this invention, based on residual convolutional neural networks and convolutional long short-term memory neural networks, can effectively extract rainfall feature information from relevant rainfall videos. Relying on the ability of deep learning methods to automatically capture relevant features and complex data distributions, the ResNet neural network model can learn rainfall features from different rainfall videos, while the Conv-LSTM neural network model can learn the relationship between rainfall features and rainfall intensity, obtaining reliable rainfall intensity information. Testing has shown that the fusion model of ResNet and Conv-LSTM can effectively acquire rainfall intensity information for the study area. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the technical process of the present invention. Detailed Implementation

[0022] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention may be implemented in many different ways as defined and covered by the claims.

[0023] Reference Figure 1 This invention proposes a method for measuring rainfall intensity based on convolutional long short memory neural networks and residual convolutional neural networks. The method includes the following steps:

[0024] Step 1: Obtain rainfall intensity data for a certain period of time in the study area, and preprocess the rainfall intensity data;

[0025] Step 2: Capture rainfall videos of the study area, and then filter and preprocess the rainfall videos;

[0026] Step 3: Set an appropriate video duration, match the rainfall intensity data with the rainfall videos, and build the grayscale video dataset required for deep learning across the entire time series; the dataset is divided into training and validation sets in a 2:1 ratio;

[0027] Step 4: Build a deep learning framework and construct a model for measuring rainfall intensity based on spatiotemporal correlation; use the Residual Convolutional Neural Network (ResNet) method to extract rainfall feature information from the video, and use the Convolutional Long Short Memory Neural Network (Conv-LSTM) method to convert the rainfall feature information into rainfall intensity;

[0028] Step 5: Input the rainfall video data and rainfall intensity data in the dataset into the deep learning model for training. Based on the error between the model output and the true value, the deep learning neural network will update the model parameters through the backpropagation algorithm until the loss error is minimized. By comprehensively comparing the errors during the training period and the validation period, the optimal model is selected.

[0029] Step 6: Input the rainfall video data of the study area for the remaining time into the optimal model in Step 5 to obtain the rainfall intensity information of the study area at other times.

[0030] Furthermore, the preprocessing includes handling missing values ​​and outliers.

[0031] Furthermore, the preprocessing in step 1 includes removing outliers from ground-observed rainfall intensity data and selecting remote sensing rainfall data with appropriate spatial resolution.

[0032] Furthermore, the filtering in step 2 includes removing videos whose clarity does not meet the requirements and removing videos with more than a preset number of obstructions.

[0033] Furthermore, the preprocessing in step 2 includes converting the selected rainfall videos to grayscale and segmenting them into short videos of a preset duration.

[0034] Furthermore, the dataset in step 3 consists of rainfall videos of the study area at a certain moment and rainfall intensity data corresponding to the time period during which the videos were taken.

[0035] Furthermore, the Residual Convolutional Neural Network (ResNet) method in step 4 extracts rainfall feature information from the video through convolution calculations in the neural network and residual structures.

[0036] Furthermore, the Convolutional Long Short Memory Neural Network (Conv-LSTM) method in step 4 processes the received spatial features of rainfall through convolution calculations in the neural network and processes the received temporal features of rainfall through the gate structure of LSTM.

[0037] In summary, the rainfall intensity measurement method based on Residual Convolutional Neural Network (ResNet) and Convolutional Long Short-Term Memory Neural Network (Conv-LSTM) described above can effectively extract rainfall features from relevant rainfall videos. It leverages the ability of deep learning methods to automatically design relevant features and capture complex data distributions. Residual Convolutional Neural Network (ResNet) can learn rainfall features from different rainfall videos, while Conv-LSTM can learn the relationship between rainfall features and rainfall intensity, thus obtaining reliable rainfall intensity information. Verification shows that the fusion model of ResNet and Conv-LSTM can accurately measure rainfall intensity information in the study area.

[0038] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for measuring rainfall intensity based on convolutional long short-term memory neural networks and residual convolutional neural networks, characterized in that, The method includes the following steps: Step 1: Obtain rainfall intensity data for a certain period of time in the study area, and preprocess the rainfall intensity data; Step 2: Capture rainfall videos of the study area, and then filter and preprocess the rainfall videos; Step 3: Set an appropriate video duration, match the rainfall intensity data with the rainfall videos, and build the grayscale video dataset required for deep learning across the entire time series; the dataset is divided into training and validation sets in a 2:1 ratio; Step 4: Build a deep learning framework and construct a model for measuring rainfall intensity based on spatiotemporal correlation; use the ResNet method to extract rainfall feature information from the video, and use the Conv-LSTM method to convert the rainfall feature information into rainfall intensity; Step 5: Input the rainfall video data and rainfall intensity data in the dataset into the deep learning model for training. Based on the error between the model output and the true value, the deep learning neural network will update the model parameters through the backpropagation algorithm until the loss error is minimized. By comprehensively comparing the errors during the training period and the validation period, the optimal model is selected. Step 6: Input the rainfall video data of the study area at other times into the optimal model in Step 5 to obtain the rainfall intensity information of the study area at other times.

2. The rainfall intensity measurement method based on convolutional long short memory neural network and residual convolutional neural network according to claim 1, characterized in that, The preprocessing includes handling missing and outlier values.

3. The rainfall intensity measurement method based on convolutional long short memory neural network and residual convolutional neural network according to claim 1, characterized in that, The preprocessing in step 1 includes removing outliers from ground-observed rainfall intensity data and selecting remote sensing rainfall data with appropriate spatial resolution.

4. The rainfall intensity measurement method based on convolutional long short memory neural network and residual convolutional neural network according to claim 1, characterized in that, The filtering in step 2 includes removing videos whose clarity does not meet the requirements and removing videos with more than a preset number of obstructions.

5. The rainfall intensity measurement method based on convolutional long short memory neural network and residual convolutional neural network according to claim 1, characterized in that, The preprocessing in step 2 includes converting the selected rainfall videos to grayscale and dividing them into short videos of a preset duration.

6. The rainfall intensity measurement method based on convolutional long short memory neural network and residual convolutional neural network according to claim 1, characterized in that, The dataset in step 3 consists of rainfall videos of the study area at a certain moment and rainfall intensity data corresponding to the time period of the video shooting.

7. The rainfall intensity measurement method based on convolutional long short memory neural network and residual convolutional neural network according to claim 1, characterized in that, The ResNet method in step 4 extracts rainfall feature information from the video through convolution calculations and residual structures in the neural network.

8. The rainfall intensity measurement method based on convolutional long short memory neural network and residual convolutional neural network according to claim 1, characterized in that, In step 4, the Conv-LSTM method uses convolutional long short memory neural network to process the received spatial features of rainfall through convolutional calculations in the neural network and uses the gating structure of LSTM to process the received temporal features of rainfall.

Images