Satellite remote sensing water body identification method based on deep learning

Abstract

The invention is suitable for the technical field of ecological environment remote sensing monitoring, and particularly relates to a satellite remote sensing water body recognition method based on deep learning, and the method comprises the steps: obtaining multispectral remote sensing data, and constructing a data set; constructing a water body recognition model, wherein the water body recognition model adopts DeepLabV < 3 + > as a basic model; optimizing the water body recognition model by adopting a cross entropy loss function, and training the water body recognition model by adopting an Adam optimizer; evaluating the performance of the water body recognition model through cross validation and performance indexes to obtain a trained water body recognition model; and importing to-be-identified data into the trained water body identification model, carrying out water body identification, and carrying out visualization processing on an identification result. According to the method, the Sentinel-2 satellite remote sensing data is utilized to realize water body recognition, the problem that in the prior art, the effect in a complex environment is often unsatisfactory and is easily influenced by cloud layers, shadows and other interference factors is solved, and the method has the advantages of being simple in operation, rapid in recognition and high in accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of ecological environment remote sensing monitoring, and in particular relates to a satellite remote sensing water body identification method based on deep learning. Background Art

[0002] Water monitoring refers to the process of evaluating water quality, pollution level and its changing trend through regular or continuous measurement and analysis of physical, chemical, biological and microbiological indicators in the water environment. It uses a variety of technologies and methods, including physical monitoring, chemical analysis, biological and microbiological testing, and modern remote sensing and Internet of Things technologies, aiming to timely discover and solve water pollution problems and protect human health and ecological environment safety.

[0003] Water body monitoring is of great significance in global climate change, ecological civilization construction and water resource protection. Traditional water body identification methods usually rely on artificial feature extraction, such as threshold method and water body index method. These methods are often unsatisfactory in complex environments and are easily affected by clouds, shadows and other interference factors. Summary of the invention

[0004] The purpose of the present invention is to provide a satellite remote sensing water body identification method based on deep learning, aiming to solve the problem that the effect of existing technologies in complex environments is often unsatisfactory and is easily affected by clouds, shadows and other interference factors.

[0005] The present invention is implemented as follows: a satellite remote sensing water body identification method based on deep learning, the method comprising: Acquire multispectral remote sensing data and construct a data set, wherein the data set includes labeled samples of water bodies and non-water bodies; Construct a water body recognition model. The water body recognition model uses DeepLabV3+ as the basic model. The water body recognition model includes a decoder and an encoder. The cross entropy loss function is used to optimize the water body recognition model, and the Adam optimizer is used to train the water body recognition model; The performance of the water body recognition model is evaluated through cross-validation and performance indicators to obtain a trained water body recognition model; Import the data to be identified into the trained water body identification model, perform water body identification, and visualize the identification results.

[0006] Preferably, the step of acquiring multispectral remote sensing data and constructing a data set specifically includes: Acquire multispectral remote sensing data from the Sentinel-2 satellite, and obtain images including blue, green, red, near-infrared, and short-wave infrared bands; Use remote sensing image annotation tools to annotate multispectral remote sensing data and identify water and non-water areas; The constructed dataset is divided into training set, validation set and test set.

[0007] Preferably, the water body recognition model is constructed, and the water body recognition model uses DeepLabV3+ as the basic model. The water body recognition model includes the steps of a decoder and an encoder, and uses DeepLabV3+ as the basic model and Xception as the backbone network to extract deep features of the remote sensing image, introduce hole convolutions at different layers, and set different hole rates. The decoder restores the feature map to the same size as the input image through upsampling operations to achieve pixel-level classification.

[0008] Preferably, the network structure of DeepLabV3+ includes: Input layer: used to receive multispectral remote sensing images and input them into the model after preprocessing; Feature extraction layer: Use Xception as the backbone network for feature extraction to extract multi-level features of the image; Dilated convolution layer: Introduce dilated convolution in the feature extraction layer, set different dilation rates, and capture features of different scales; Decoding layer: Through upsampling and feature fusion, the feature map is restored to the size of the original image to achieve pixel-level classification; Output layer: Outputs the classification results of each pixel, identifying water bodies and non-water bodies.

[0009] Preferably, the step of optimizing the water body recognition model by using a cross entropy loss function and training the water body recognition model by using an Adam optimizer comprises: Normalize multispectral remote sensing data and adjust image size; Set training parameters, use the Adam optimization algorithm for training, regularly evaluate model performance during training, and adjust hyperparameters based on verification results; During the training process, random rotation, flipping and scaling operations are performed to reduce the risk of overfitting through data augmentation.

[0010] Preferably, the performance of the water body recognition model is evaluated through cross-validation and performance indicators to obtain a trained water body recognition model, and the performance is evaluated through a test set, the accuracy, recall rate and F1-score indicators are calculated, and the classification effect of the model is analyzed through a confusion matrix to identify misclassification situations.

[0011] Preferably, in the step of calculating the accuracy, recall and F1-score indicators, the evaluation indicators are: Accuracy: the ratio of correctly classified pixels to the total number of pixels; Recall rate: the ratio of the number of correctly identified water pixels to the number of actual water pixels; F1-score: An indicator that comprehensively considers accuracy and recall. The calculation formula is: F1-Score = 2 * (Precision * Recall) / (Precision + Recall).

[0012] Preferably, in the step of importing the data to be identified into the trained water body identification model, the trained water body identification model is deployed to a server, the server is provided with a human-computer interaction interface, and the user uploads the multispectral remote sensing data through the server.

[0013] Preferably, the architecture of the server includes an upload module, a model reasoning module and a result display module.

[0014] The satellite remote sensing water body identification method based on deep learning provided by the present invention uses Sentinel-2 satellite remote sensing data to realize water body identification, which solves the problem that the effect of the prior art in complex environments is often unsatisfactory and is easily affected by clouds, shadows and other interference factors. It has the advantages of simple operation, fast identification and high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of a satellite remote sensing water body identification method based on deep learning of the present invention; Figure 2 It is a deep learning DeepLabV3+ model structure diagram of a satellite remote sensing water body identification method based on deep learning of the present invention; Figure 3 It is a first output result graph of water body identification obtained by a satellite remote sensing water body identification method based on deep learning of the present invention; Figure 4 This is a second output result diagram of water body identification obtained by a satellite remote sensing water body identification method based on deep learning in the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] like Figure 1 FIG. 1 is a flow chart of a method for identifying water bodies using satellite remote sensing based on deep learning according to the present invention, wherein the method comprises: Multispectral remote sensing data is acquired and a data set is constructed, wherein the data set includes labeled samples of water bodies and non-water bodies.

[0018] In this step, multispectral remote sensing data is obtained from the Sentinel-2 satellite, and the data is preprocessed, including multispectral images (such as blue, green, red, and near-infrared bands) and combined with ground-based measured data to build a high-quality water body identification dataset. The dataset includes labeled samples of water bodies and non-water bodies, and the images are cropped, scaled, and normalized. Data enhancement is performed to improve the generalization ability of the model; the data source can be the open data platform of the European Space Agency (ESA); the acquired images are labeled using a professional remote sensing image annotation tool (LabelMe) to identify water bodies and non-water body areas. Ensure the accuracy and consistency of the annotations to improve the effect of model training, and divide the constructed dataset into training set, validation set, and test set, usually with a ratio of 70% for training, 15% for validation, and 15% for testing to ensure the generalization ability of the model.

[0019] Construct a water body recognition model. The water body recognition model uses DeepLabV3+ as the basic model. The water body recognition model includes a decoder and an encoder.

[0020] In this step, DeepLabV3+ is used as the basic model, and its hole convolution characteristics are used to enhance the model's ability to recognize water bodies of different scales. The model structure includes an encoder and a decoder. The encoder extracts features and the decoder performs segmentation. Xception is used as the backbone network to extract deep features of remote sensing images. Hollow convolutions are introduced at different layers, and different hole rates are set to enhance the model's ability to capture multi-scale information. The decoder restores the feature map to the same size as the input image through upsampling operations to achieve pixel-level classification.

[0021] like Figure 2 As shown in the figure, the network structure of DeepLabV3+ mainly includes the following parts: Input layer: accepts multispectral remote sensing images and inputs them into the model after preprocessing.

[0022] Feature extraction layer: Use Xception as the backbone network for feature extraction to extract multi-level features of the image.

[0023] Dilated convolution layer: Introduce dilated convolution in the feature extraction layer, set different dilation rates, and capture features of different scales.

[0024] Decoding layer: Through upsampling and feature fusion, the feature map is restored to the size of the original image to achieve pixel-level classification.

[0025] Output layer: Outputs the classification results of each pixel, identifying water bodies and non-water bodies.

[0026] The cross entropy loss function is used to optimize the water body recognition model, and the Adam optimizer is used to train the water body recognition model.

[0027] In this step, the constructed data set is used to train the water body recognition model. The specific steps are as follows: Data preprocessing: Normalize the image and adjust the image size to ensure the consistency of the input data. Image enhancement techniques (such as random cropping, rotation, flipping, etc.) can be used to increase the diversity of the data.

[0028] Model training: Set the training parameters, including learning rate, batch size, and number of iterations, and use the Adam optimization algorithm for training. During the training process, the model (i.e.) performance is regularly evaluated and the hyperparameters are adjusted based on the verification results.

[0029] Data Augmentation: During the training process, random rotation, flipping, and scaling operations are performed to enhance the robustness of the model. Through data enhancement technology, the generalization ability of the model can be effectively improved and the risk of overfitting can be reduced.

[0030] The performance of the water body recognition model is evaluated through cross-validation and performance indicators to obtain a trained water body recognition model.

[0031] In this step, the model is verified, the model performance is evaluated on the test set, and indicators such as accuracy, recall, and F1-score are calculated to ensure the effectiveness of the model. At the same time, the confusion matrix is ​​used to analyze the classification effect of the model, identify misclassification, and provide a basis for subsequent optimization.

[0032] Evaluation Metrics: Precision: The ratio of the number of correctly classified pixels to the total number of pixels.

[0033] Recall: The ratio of the number of correctly identified water pixels to the actual number of water pixels.

[0034] F1-score: An indicator that comprehensively considers accuracy and recall. The calculation formula is: F1-Score = 2 * (Precision * Recall) / (Precision + Recall).

[0035] Confusion Matrix: The confusion matrix is ​​used to analyze the classification effect of the model, identify misclassification situations, and understand the performance of the model between different categories to facilitate subsequent optimization.

[0036] Import the data to be identified into the trained water body identification model, perform water body identification, and visualize the identification results.

[0037] In this step, application and deployment, the trained model is deployed to the server, and a user-friendly interface is developed. Users can upload Sentinel-2 satellite remote sensing images, and the system automatically identifies water bodies and returns the results in the form of images and data.

[0038] The server system architecture includes a data upload module, a model inference module, and a result display module. Users upload images through the front-end interface, and the back-end server calls the trained model for inference, and finally displays the recognition results to the user in a visual form.

[0039] User interface design: Design a simple and easy-to-use user interface so that users can easily upload images, view recognition results, and download analysis reports.

[0040] Experiments were conducted on Sentinel-2 satellite remote sensing images from different time periods. The performance of the model in the water body recognition task is shown in Table 1: Table 1 Training and verification results of satellite remote sensing water body recognition method based on deep learning area time Accuracy Recall F1-Score Tianjin Fall 2023 93.3% 91.1% 92.2% Tianjin Spring 2024 93.7% 91.7% 92.7% Tianjin Summer 2024 94.5% 92.1% 93.3% The above results show that the water body recognition algorithm of the present invention shows good performance in different time periods, with high accuracy and robustness. Figure 3 , Figure 4 This is the output result of water body identification. The blue area is the water area.

[0041] Through efficient data processing and model design, the present invention can input the area of ​​the water body to be identified with one click and obtain the corresponding output result map of water body identification, which significantly improves the accuracy and efficiency of water body identification and has good application prospects and market value.

[0042] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A satellite remote sensing water body identification method based on deep learning, characterized in that: The method comprises: Acquire multispectral remote sensing data and construct a data set, wherein the data set includes labeled samples of water bodies and non-water bodies; Construct a water body recognition model. The water body recognition model uses DeepLabV3+ as the basic model. The water body recognition model includes a decoder and an encoder. The cross entropy loss function is used to optimize the water body recognition model, and the Adam optimizer is used to train the water body recognition model; The performance of the water body recognition model is evaluated through cross-validation and performance indicators to obtain a trained water body recognition model; Import the data to be identified into the trained water body identification model, perform water body identification, and visualize the identification results; The water body recognition model is constructed, and the water body recognition model uses DeepLabV3+ as a basic model. The water body recognition model includes the steps of a decoder and an encoder. DeepLabV3+ is used as the basic model, Xception is used as the backbone network, deep features of the remote sensing image are extracted, and hole convolutions are introduced at different layers. Different hole rates are set. The decoder restores the feature map to the same size as the input image through an upsampling operation to achieve pixel-level classification.

2. The satellite remote sensing water body identification method based on deep learning according to claim 1 is characterized in that: The step of acquiring multispectral remote sensing data and constructing a data set specifically includes: Acquire multispectral remote sensing data from the Sentinel-2 satellite, and obtain images including blue, green, red, near-infrared, and short-wave infrared bands; Use remote sensing image annotation tools to annotate multispectral remote sensing data and identify water and non-water areas; The constructed dataset is divided into training set, validation set and test set.

3. The satellite remote sensing water body identification method based on deep learning according to claim 1 is characterized in that: The network structure of DeepLabV3+ includes: Input layer: used to receive multispectral remote sensing images and input them into the model after preprocessing; Feature extraction layer: Use Xception as the backbone network for feature extraction to extract multi-level features of the image; Dilated convolution layer: Introduce dilated convolution in the feature extraction layer, set different dilation rates, and capture features of different scales; Decoding layer: Through upsampling and feature fusion, the feature map is restored to the size of the original image to achieve pixel-level classification; Output layer: Outputs the classification results of each pixel, identifying water bodies and non-water bodies.

4. The satellite remote sensing water body identification method based on deep learning according to claim 1 is characterized in that: The step of optimizing the water body recognition model by using the cross entropy loss function and training the water body recognition model by using the Adam optimizer includes: Normalize multispectral remote sensing data and adjust image size; Set training parameters, use the Adam optimization algorithm for training, regularly evaluate model performance during training, and adjust hyperparameters based on verification results; During the training process, random rotation, flipping and scaling operations are performed to reduce the risk of overfitting through data augmentation.

5. The satellite remote sensing water body identification method based on deep learning according to claim 1 is characterized in that: The performance of the water body recognition model is evaluated through cross-validation and performance indicators to obtain a trained water body recognition model, and the performance is evaluated through a test set, the accuracy, recall rate and F1-score indicators are calculated, and the classification effect of the model is analyzed through a confusion matrix to identify misclassification.

6. The satellite remote sensing water body identification method based on deep learning according to claim 5 is characterized in that: In the steps of calculating the accuracy, recall and F1-score indicators, the evaluation indicators are: Accuracy: the ratio of correctly classified pixels to the total number of pixels; Recall rate: the ratio of the number of correctly identified water pixels to the number of actual water pixels; F1-score: An indicator that comprehensively considers accuracy and recall. The calculation formula is: F1-Score = 2 * (Precision * Recall) / (Precision + Recall).

7. The satellite remote sensing water body identification method based on deep learning according to claim 1 is characterized in that: In the step of importing the data to be identified into the trained water body identification model, the trained water body identification model is deployed to a server, the server is provided with a human-computer interaction interface, and the user uploads the multispectral remote sensing data through the server.

8. The method for identifying water bodies using satellite remote sensing based on deep learning according to claim 7, characterized in that: The architecture of the server includes an upload module, a model reasoning module and a result display module.

Images