Semantic segmentation-based digital mapping method and system for water network system

Abstract

The invention discloses a water network system digital mapping method and system based on semantic segmentation, and the method employs a deep learning algorithm to construct a semantic segmentation model, so as to accurately recognize the three elements of'outline-view-knot 'in a water network system, and improve the precision and efficiency of digital mapping. According to the method, by combining a multi-scale convolution kernel and an attention mechanism, capturing and enhancing of different scale features in a complex scene are achieved, the segmentation precision of the model can be improved, the adaptability of the model to the complex scene can be enhanced, and powerful support is provided for digital mapping of a water network image. Meanwhile, by designing a specific loss function and introducing an attention mechanism, the model can better understand the structural relationship in the water network image, accurately identify the connection relationship among different elements, and provide reliable technical support for realizing high-precision and high-efficiency water network digital mapping.

Description

Technical Field

[0001] The present invention belongs to the field of digital watershed, and in particular relates to a method and system for digital mapping of a water network system based on semantic segmentation. Background Art

[0002] The water network system refers to the overall water resource transportation and management network composed of rivers, lakes, channels, pipelines, etc. It consists of the main water network structure - outline (for example, natural water systems of major rivers, lakes and lakes, major water diversion and regulation projects and backbone water supply and drainage channels, etc.), specific water network elements - items (for example, regional river and lake water system interconnection projects and water supply channels) and water network connection relationships - nodes (for example, water resource regulation and storage projects with controlling status and functions). It is an important part of water resource scheduling, and its complexity and scale place high demands on management and maintenance.

[0003] Traditional manual management methods are not only time-consuming and labor-intensive, but also difficult to ensure the accuracy and timeliness of data. With the rapid development of computer vision technology, semantic segmentation technology provides a new solution for the digital management of water network systems. However, the existing semantic segmentation technology still faces many difficulties when applied to water network systems, such as low recognition accuracy and slow processing speed. In particular, the challenges are particularly prominent when processing remote sensing images with high resolution and complex scenes.

[0004] In addition, water network images often contain a large number of complex scenes, such as water network elements of different scales and overlapping water network structures, which puts higher requirements on the segmentation model. In addition, existing semantic segmentation technologies are usually oriented to general image tasks, and there are still deficiencies in the digital mapping of the three elements of the water network (i.e., the main water network structure, specific water network elements, and water network connection relationships), resulting in the accuracy and efficiency of digital mapping failing to meet actual needs. Summary of the invention

[0005] The present invention provides a method and system for digital mapping of a water network system based on semantic segmentation, so as to solve the problem that the prior art cannot meet the accuracy and efficiency required for digital mapping of a water network system. In order to solve the above technical problems, the embodiments of the present invention disclose the following technical solutions: One aspect of the present invention provides a method for digital mapping of a water network system based on semantic segmentation, comprising: Acquire a plurality of remote sensing images annotated with water network element categories, wherein the water network element categories include class, order, and knot; Based on the labeled remote sensing image, a semantic segmentation model is obtained by training with a multi-scale convolution kernel and an attention mechanism; The remote sensing image of the target water network area is input into the semantic segmentation model to obtain the water network element category corresponding to each pixel in the remote sensing image, and combined with the pre-acquired basic data of the target water network area to generate a digital map.

[0006] Optionally, the acquiring of a plurality of remote sensing images annotated with water network element categories includes: Use remote sensing satellites and / or drones to obtain multiple remote sensing images of the water network system; Each remote sensing image is preprocessed; The water network feature categories are labeled for the water network system in the preprocessed remote sensing image according to the user input data.

[0007] Optionally, the preprocessing of each remote sensing image includes: Gaussian filtering is used to denoise remote sensing images; Enhance the contrast of remote sensing images through histogram equalization method; The remote sensing images are corrected using geometric transformation methods.

[0008] Optionally, the semantic segmentation model is obtained by training a multi-scale convolution kernel and an attention mechanism based on the labeled remote sensing image, including: The remote sensing images labeled with water network feature categories are used as training data; A multi-scale convolution kernel is introduced to extract features of different scales in the training data, and the attention mechanism is used to identify the connection relationship between water network elements of different categories. The preset loss function is used as the target for iterative training until the semantic segmentation model converges.

[0009] Optionally, the method further includes: Input the pre-set test data into the trained semantic segmentation model; The performance of the semantic segmentation model is verified based on performance evaluation indicators.

[0010] Optionally, the remote sensing image of the target water network area is input into a semantic segmentation model to obtain the water network element category corresponding to each pixel in the remote sensing image, and combined with the pre-acquired basic data of the target water network area to generate a digital map, including: Acquire a remote sensing image of a target water network area, and preprocess the remote sensing image; The preprocessed remote sensing image is input into the trained semantic segmentation model to identify the water network feature category of each pixel; The remote sensing image is matched with basic data of the target water network area, and a digital map is constructed based on the spatial position of each pixel and the category of the water network element.

[0011] Optionally, matching the remote sensing image with basic data of the target water network area and constructing a digital map based on the spatial position of each pixel and the category of the water network element includes: Use spatial overlay analysis technology to match remote sensing images with basic data of the target water network area; Generate a digital image of the target water network area according to the spatial position of each pixel in the remote sensing image and the category of water network elements; The digital image is combined with basic data to establish a digital map of the target water network area.

[0012] Optionally, after generating the digital map, the method further includes: Based on the historical monitoring data of the target water network area, a prediction model is constructed using a machine learning algorithm; Input the real-time monitoring data of the target water network area into the prediction model to obtain the prediction data for the future period; The predicted data is matched with the digital map to provide a data basis for users to optimize water resource scheduling strategies.

[0013] Optionally, the method further includes: Integrate the digital map and user-defined water resource scheduling strategy into the existing management system and provide users with a visual display interface.

[0014] Another aspect of the present invention provides a digital mapping system for a water network system based on semantic segmentation, comprising: A remote sensing image acquisition module is configured to acquire a plurality of remote sensing images annotated with water network element categories, wherein the water network element categories include class categories, order categories, and knot categories; A semantic segmentation model module is configured to obtain a semantic segmentation model based on the labeled remote sensing image by using a multi-scale convolution kernel and an attention mechanism for training; The digital mapping module is configured to input the remote sensing image of the target water network area into the semantic segmentation model, obtain the water network element category corresponding to each pixel in the remote sensing image, and generate a digital mapping map in combination with the pre-acquired basic data of the target water network area.

[0015] The present invention discloses a method and system for digital mapping of a water network system based on semantic segmentation, which uses a deep learning algorithm to construct a semantic segmentation model to accurately identify the three elements of "structure-mesh-knot" in the water network system, thereby improving the accuracy and efficiency of digital mapping. Among them, by combining multi-scale convolution kernels and attention mechanisms, the capture and enhancement of features of different scales in complex scenes can be achieved, which can not only improve the segmentation accuracy of the model, but also enhance the adaptability of the model to complex scenes, providing strong support for the digital mapping of water network images. At the same time, by designing a specific loss function and introducing an attention mechanism, the model can better understand the structural relationship in the water network image, accurately identify the connection relationship between different elements, and provide a reliable technical guarantee for achieving high-precision and high-efficiency digital mapping of water networks.

[0016] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the disclosure, nor is it intended to limit the scope of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects, features and advantages of the present disclosure will become more apparent through a more detailed description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present disclosure.

[0018] Figure 1 A schematic diagram of a flow chart of a digital mapping method for a water network system based on semantic segmentation provided by an embodiment of the present invention; Figure 2 An implementation provided for an embodiment of the present invention Figure 1 Schematic diagram of the process of step S100; Figure 3 An implementation provided for an embodiment of the present invention Figure 1 Schematic diagram of the process of step S200; Figure 4 An implementation provided for an embodiment of the present invention Figure 1 Schematic diagram of the process of step S300; Figure 5 An implementation provided for an embodiment of the present invention Figure 4 Schematic diagram of the process of step S303; Figure 6 A schematic structural diagram of a water network system digital mapping system based on semantic segmentation provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0020] As used herein, the term "including" and its variations mean open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "based at least in part on". The terms "an example embodiment" and "an embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0021] Figure 1 A flow chart of a method for digital mapping of a water network system based on semantic segmentation provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method comprises the following steps: Step S100: Acquire a plurality of remote sensing images marked with water network element categories.

[0022] The categories of water network elements include outline, item and knot. Among them, the outline category is the main water network structure, such as the Yangtze River, Yellow River, Pearl River and other major rivers, Dongting Lake, Poyang Lake, Taihu Lake and other large lakes; the item category is the specific water network elements, which are secondary elements in the water network system that supplement or support the operation of the outline functions, such as regional river and lake water system connection projects and water supply channels; the knot category is the element connecting the outline category and the item category, which is a water resource storage project with a controlling status and function, such as reservoirs, flood storage areas, water conservancy hubs, gate systems, etc.

[0023] In one embodiment disclosed in the present invention, Figure 2 As shown, the following sub-steps are used to implement S100: Step S101: Acquire multiple remote sensing images of a water network system using remote sensing satellites and / or drones.

[0024] The use of high-resolution remote sensing satellites (such as the Gaofen series, Landsat, and Sentinel) to obtain large-scale water network images is suitable for comprehensive observation of wide-area water systems.

[0025] Low-altitude drones (equipped with high-definition cameras or multispectral sensors) are used to collect high-precision images of local water network areas to compensate for the insufficient resolution or lack of details in satellite images.

[0026] Step S102: pre-process each remote sensing image.

[0027] In one embodiment disclosed in the present invention, preprocessing can be completed in the following manner: (1) Denoising: Gaussian filtering or median filtering is used to remove noise caused by sensor or environmental interference in remote sensing images. At the same time, for satellite remote sensing images, multispectral cloud detection algorithms can be used to remove areas blocked by clouds.

[0028] (2) Image enhancement: The histogram equalization method is used to highlight the boundary between the water network and the surrounding area. The water body index method (such as NDWI: Normalized Difference Water Index) can also be applied to distinguish between water bodies and non-water areas.

[0029] (3) Geometric correction: The geometric transformation method is used to correct the geometric distortion of the image caused by the earth's curvature, imaging angle and sensor characteristics to ensure the accuracy of the geographic location in the image.

[0030] Step S103: labeling water network element categories for the water network system in the pre-processed remote sensing image according to the user input data.

[0031] Users can use geographic information systems or other software to label the water network element categories for each processed remote sensing image.

[0032] Step S200: Based on the labeled remote sensing image, a semantic segmentation model is obtained by training with a multi-scale convolution kernel and an attention mechanism.

[0033] In one embodiment disclosed in the present invention, Figure 3 As shown, the following sub-steps may be used to implement step S200: Step S201: Use remote sensing images marked with water network element categories as training data.

[0034] In one embodiment disclosed in the present invention, remote sensing images labeled with water network element categories are divided into training data, verification data and test data, wherein the category labels at least include three element classifications of the water network system: class, order and knot.

[0035] Step S202: introduce a multi-scale convolution kernel to extract features of different scales in the training data, use an attention mechanism to identify the connection relationship between water network elements of different categories, and perform iterative training with a preset loss function as the target until the semantic segmentation model converges.

[0036] (1) Multi-scale convolution kernel Convolution kernels of different sizes (such as 3×3, 5×5, and 7×7) are used in the convolutional layer of the model to capture features of different scales in remote sensing images. Small convolution kernels are used to extract local fine features, such as tributaries and irrigation canals, while large convolution kernels are used to extract global features, such as the overall morphology of main rivers or large lakes.

[0037] The features extracted by convolution kernels of different scales are fused, combining fine-grained and coarse-grained information to improve the model's adaptability to complex scenes.

[0038] (2) Attention Mechanism In one embodiment disclosed in the present invention, a self-attention mechanism can be used to assign weights to each pixel, dynamically adjust its feature representation based on its correlation with other pixels, emphasize the characteristics of key areas in the water network system (such as river intersections or gate locations), and make the model pay more attention to these key points.

[0039] In one embodiment disclosed in the present invention, a channel attention mechanism may also be used to assign weights to each channel of the feature map to highlight important category information, such as the characteristics of major water systems or controlling nodes.

[0040] In one embodiment disclosed in the present invention, a spatial attention mechanism can also be used to learn the dependencies of spatial positions to help the model understand the connection between water network elements, such as the distribution of main rivers and tributaries or the connection between reservoirs and irrigation channels.

[0041] (3) Preset loss function In one embodiment disclosed in the present invention, cross entropy loss can be used as a basic loss function to measure the difference between the predicted category and the true label. Context loss can also be introduced to combine context information to constrain the connection relationship between water network elements and improve the accuracy of overall segmentation.

[0042] The Adam or SGD optimization algorithm is used to iteratively adjust the model parameters and gradually reduce the loss function value. When the loss of the model on the validation set no longer decreases significantly or reaches the set number of iterations, the training is stopped.

[0043] In one embodiment disclosed in the present invention, after the semantic segmentation model is trained, the model is verified in the following manner: (1) Input the pre-set test data into the trained semantic segmentation model, where the test data is a remote sensing image with labeled water network feature categories.

[0044] (2) The performance of the semantic segmentation model is verified according to the performance evaluation indicators. The performance of the semantic segmentation model can be evaluated by using indicators such as pixel accuracy, average intersection-over-union ratio, category intersection-over-union ratio, confusion matrix, etc.

[0045] Step S300: Input the remote sensing image of the target water network area into the semantic segmentation model to obtain the water network element category corresponding to each pixel in the remote sensing image, and generate a digital map in combination with the pre-acquired basic data of the target water network area.

[0046] In one embodiment disclosed in the present invention, Figure 4 As shown, the following sub-steps may be used to implement step S300: Step S301: Acquire a remote sensing image of a target water network area and preprocess the remote sensing image.

[0047] High-resolution images covering the target water network area are acquired by remote sensing satellites or drones, and the remote sensing images are preprocessed in a manner similar to that in the above-mentioned embodiment.

[0048] Step S302: input the preprocessed remote sensing image into the trained semantic segmentation model to identify the water network element category of each pixel.

[0049] For pixels related to the water network system, the characteristic values ​​of the water network element categories can be marked as the corresponding categories; for pixels unrelated to the water network system, such as buildings, roads, etc., the characteristic values ​​of the water network element categories can be marked as none or other.

[0050] Step S303: Match the remote sensing image with the basic data of the target water network area, and construct a digital map based on the spatial position of each pixel and the category of the water network element.

[0051] Combined with the geographic information of remote sensing images, using geographic information systems or other tools, the spatial coordinates of each pixel are spatially superimposed with the basic geographic data of the target water network area, and a digital map is generated in combination with the pixel classification results.

[0052] In one embodiment disclosed in the present invention, Figure 5 As shown, the following sub-steps may be used to implement step S300: Step S3031: Use spatial overlay analysis technology to match the remote sensing image with the basic data of the target water network area.

[0053] The classification results of remote sensing images are superimposed with the basic geographic data of the target area (such as administrative boundaries and water resource utilization data) using spatial overlay analysis technology.

[0054] Step S3032: Generate a digital image of the target water network area according to the spatial position of each pixel in the remote sensing image and the water network element category.

[0055] In a specific embodiment disclosed in the present invention, the pixel values ​​can be re-encoded according to the semantic segmentation results. For example, each water network element category is assigned a unique color or number, and a raster image is output to intuitively display the classification results of the target water network area.

[0056] Step S3033: Combine the digital image with the basic data to create a digital map of the target water network area.

[0057] Basic data include but are not limited to the following: terrain elevation data (DEM), river distribution map, location of water conservancy projects, water resources utilization and scheduling planning data, administrative boundaries or geographical division information, etc.

[0058] The obtained digital image is integrated with the basic data to obtain a digital map of the target water network area, and the integrated information is saved in a standard geographic information format (such as GeoJSON, Shapefile) for easy application in the management system.

[0059] In one embodiment disclosed in the present invention, after generating the digital map, the following steps are also included: (1) Based on the historical monitoring data of the target water network area, a prediction model is constructed using a machine learning algorithm.

[0060] Collect historical data of the target water network area, including but not limited to: dynamic hydrological data such as water level, flow velocity, and flow; meteorological data such as rainfall and evaporation; water resource utilization (such as water supply and water consumption), etc.

[0061] After processing the outliers and missing values ​​of the historical data, the prediction model is trained using the machine learning algorithm.

[0062] (2) Input the real-time monitoring data of the target water network area into the prediction model to obtain the prediction data for the future period.

[0063] (3) Match the predicted data with the digital map to provide a data basis for users to optimize water resource scheduling strategies.

[0064] Overlay forecast data onto digital maps to give the forecast results a spatial reference. For example, mark areas of high water levels or water shortages in the future on the digital map.

[0065] Users define scheduling strategies based on the overlay map and water resource management goals (such as flood control and water supply priority). For example, they can adjust the opening and closing time of reservoirs or sluice gates, optimize the flow distribution of water supply channels, etc.

[0066] In an embodiment disclosed in the present invention, the digital map and the water resource scheduling strategy set by the user can also be integrated into the existing management system, and a visual display interface can be provided for the user.

[0067] The generated digital map file is loaded into the management system, and the map can be linked with the real-time monitoring data to regularly update the operation status of the water network.

[0068] The user-defined scheduling strategy and the digital map are integrated into the management system, and the execution results of the scheduling strategy (such as the sluice switch status and the water supply channel flow) can be marked on the map.

[0069] Figure 6 A schematic diagram of a water network system digital mapping system based on semantic segmentation provided by an embodiment of the present invention is shown in FIG. Figure 6 As shown, the system includes the following modules: The remote sensing image acquisition module 1 is configured to acquire a plurality of remote sensing images annotated with water network element categories, where the water network element categories include class, order and knot.

[0070] The semantic segmentation model module 2 is configured to obtain a semantic segmentation model based on the labeled remote sensing images using multi-scale convolution kernels and attention mechanism training.

[0071] The digital mapping module 3 is configured to input the remote sensing image of the target water network area into the semantic segmentation model, obtain the water network element category corresponding to each pixel in the remote sensing image, and generate a digital mapping map in combination with the pre-acquired basic data of the target water network area.

[0072] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A digital mapping method for a water network system based on semantic segmentation, characterized in that: include: Acquire a plurality of remote sensing images annotated with water network element categories, wherein the water network element categories include class, order, and knot; Based on the labeled remote sensing image, a semantic segmentation model is obtained by training with a multi-scale convolution kernel and an attention mechanism; The remote sensing image of the target water network area is input into the semantic segmentation model to obtain the water network element category corresponding to each pixel in the remote sensing image, and combined with the pre-acquired basic data of the target water network area to generate a digital map.

2. The digital mapping method according to claim 1, characterized in that: The step of obtaining a plurality of remote sensing images marked with water network element categories includes: Use remote sensing satellites and / or drones to obtain multiple remote sensing images of the water network system; Each remote sensing image is preprocessed; According to the user input data, the water network feature categories are labeled for the water network system in the pre-processed remote sensing image.

3. The digital mapping method according to claim 2, characterized in that: The preprocessing of each remote sensing image includes: Gaussian filtering is used to denoise remote sensing images; Enhance the contrast of remote sensing images through histogram equalization method; The remote sensing images are corrected using geometric transformation methods.

4. The digital mapping method according to claim 1, characterized in that: The semantic segmentation model is obtained by training a multi-scale convolution kernel and an attention mechanism based on the labeled remote sensing image, including: The remote sensing images labeled with water network feature categories are used as training data; A multi-scale convolution kernel is introduced to extract features of different scales in the training data, and the attention mechanism is used to identify the connection relationship between water network elements of different categories. The preset loss function is used as the target for iterative training until the semantic segmentation model converges.

5. The digital mapping method according to claim 1, characterized in that: The method further comprises: Input the pre-set test data into the trained semantic segmentation model; The performance of the semantic segmentation model is verified based on performance evaluation indicators.

6. The digital mapping method according to claim 1, characterized in that: The remote sensing image of the target water network area is input into the semantic segmentation model to obtain the water network element category corresponding to each pixel in the remote sensing image, and combined with the pre-acquired basic data of the target water network area to generate a digital map, including: Acquire a remote sensing image of a target water network area, and preprocess the remote sensing image; The preprocessed remote sensing image is input into the trained semantic segmentation model to identify the water network feature category of each pixel; The remote sensing image is matched with basic data of the target water network area, and a digital map is constructed based on the spatial position of each pixel and the category of the water network element.

7. The digital mapping method according to claim 6, characterized in that: The remote sensing image is matched with the basic data of the target water network area, and a digital map is constructed based on the spatial position of each pixel and the category of the water network elements, including: Use spatial overlay analysis technology to match remote sensing images with basic data of the target water network area; Generate a digital image of the target water network area according to the spatial position of each pixel in the remote sensing image and the category of water network elements; The digital image is combined with basic data to establish a digital map of the target water network area.

8. The digital mapping method according to claim 1, characterized in that: After generating the digital map, the method further includes: Based on the historical monitoring data of the target water network area, a prediction model is constructed using a machine learning algorithm; Input the real-time monitoring data of the target water network area into the prediction model to obtain the prediction data for the future period; The predicted data is matched with the digital map to provide a data basis for users to optimize water resource scheduling strategies.

9. The digital mapping method according to claim 8, characterized in that: The method further comprises: Integrate the digital map and user-defined water resource scheduling strategy into the existing management system and provide users with a visual display interface.

10. A digital mapping system for water network system based on semantic segmentation, characterized in that: include: A remote sensing image acquisition module is configured to acquire a plurality of remote sensing images annotated with water network element categories, wherein the water network element categories include class categories, order categories, and knot categories; A semantic segmentation model module is configured to obtain a semantic segmentation model based on the labeled remote sensing image by using a multi-scale convolution kernel and an attention mechanism for training; The digital mapping module is configured to input the remote sensing image of the target water network area into the semantic segmentation model, obtain the water network element category corresponding to each pixel in the remote sensing image, and generate a digital mapping map in combination with the pre-acquired basic data of the target water network area.

Images