Forest region road network model construction method and device considering terrain constraint and crown shielding

Abstract

The invention discloses a forest road network model construction method and device considering terrain constraints and crown shielding, and belongs to the technical field of road recognition and road network construction. The method comprises the following steps: constructing a data sample set containing typical forest region road features; a machine learning method is adopted, a forest region road preliminary recognition result based on satellite remote sensing visible light data is obtained, an under-forest path recognition result of a high forest stand canopy density area is obtained, the under-forest path recognition result and the forest region road preliminary recognition result are fused, and a forest region fine road network is obtained; high-precision forest region surface elevation information is obtained based on SAR data, DEM data is fused to extract forest region topographic and geomorphic features, and a forest region road network model based on multi-source data is constructed. According to the method, terrain constraint factors are fully considered in the road network construction process, fine surface elevation information is extracted in combination with SAR data, the accuracy of a road network model is guaranteed, and the rationality of road network planning is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of road identification and road network construction, and specifically relates to a method and device for constructing a forest road network model taking into account terrain constraints and tree crown shading. Background Art

[0002] The construction of forest area road network information is an important prerequisite for carrying out forest protection, emergency rescue, nature education and research activities. Especially in the process of forest fire and personnel rescue, high-precision forest area road network information can provide guidance for the rapid response of ground forces in fire rescue activities. Considering the large forest coverage area and the incomplete forest area road network information, the use of remote sensing information to obtain large-scale forest area road network information is currently a common method.

[0003] Currently, when obtaining forest road network information based on remote sensing information, the most commonly used remote sensing data is visible light data. However, due to the complex terrain of the forest area and the high density of the forest stands, the path information obtained only by visible light data is incomplete, and the passability of the path is restricted by the terrain. To address this problem, researchers often use airborne laser radar to detect point cloud data in forest areas, which makes up for the path information of some understory areas and reconstructs the forest terrain. However, the coverage area of ​​airborne laser radar detection is limited and cannot meet the needs of ground activities in large forest areas.

[0004] The strong penetration ability of satellite remote sensing synthetic aperture radar (SAR) data has been used as a supplement to urban road information identification. However, SAR data has not yet been used to identify understory paths in forest areas with high forest canopy density. At the same time, considering that the terrain of forest paths is different from urban roads, there are many terrain features such as deep trenches, cliffs, and swamps, which has brought a major bottleneck to the construction of forest area road networks. Summary of the invention

[0005] To solve the above technical problems, the present invention provides a method and device for constructing a forest road network model taking into account terrain constraints and tree crown shading. Based on multi-source data of forest areas, and facing the characteristics of high forest stand density and complex terrain in forest areas, the method and device combine time-series satellite remote sensing visible light data and SAR data to obtain information on forest paths that break through tree crown shading, further extract forest area terrain features from SAR data and DEM data, integrate the terrain features with the forest road network, and construct a forest road network model that takes into account tree crown shading and terrain constraints.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] A method for constructing a forest road network model considering terrain constraints and tree canopy shading, the method comprising:

[0008] Step 1: Collect multi-source data of typical forest areas, annotate roads in the multi-source data, and construct a data sample set containing road features of typical forest areas, wherein the multi-source data includes satellite remote sensing visible light data, SAR data, and DEM data;

[0009] Step 2: Analyze the differences in forest stand canopy density in different seasons in the forest area based on the data sample set constructed based on satellite remote sensing visible light data, use machine learning methods to train a machine learning network model suitable for forest road recognition, and obtain preliminary forest road recognition results based on satellite remote sensing visible light data;

[0010] Step 3: for the data sample set constructed based on SAR data, the long-wave SAR data and the short-wave SAR data are integrated to construct a forest path recognition model data sample set, and a machine learning network model adapted to the forest path recognition model data sample set is trained by a machine learning method to obtain the forest path recognition result in the high forest canopy density area, and the forest path recognition result is integrated with the preliminary recognition result of the forest road to obtain a fine road network in the forest area;

[0011] Step 4: Obtain high-precision forest area surface elevation information based on SAR data, fuse DEM data to extract forest area topographic features, and combine with the forest area's fine road network to build a forest area road network model based on multi-source data.

[0012] In another aspect, the present invention provides a device for constructing a forest road network model taking into account terrain constraints and tree canopy shading, comprising:

[0013] A collection module is used to collect multi-source data of typical forest areas, and after marking the roads in the multi-source data, construct a data sample set containing the characteristics of the roads in the typical forest areas, wherein the multi-source data includes satellite remote sensing visible light data, SAR data and DEM data;

[0014] The forest road recognition module is used to train a machine learning network model suitable for forest road recognition based on a data sample set constructed based on satellite remote sensing visible light data, using a machine learning method, and obtain preliminary forest road recognition results based on satellite remote sensing visible light data;

[0015] A forest area fine road network acquisition module is used to construct a forest path recognition model data sample set for a data sample set constructed based on SAR data by fusing long-wave SAR data and short-wave SAR data, and to train a machine learning network model adapted to the forest path recognition model data sample set using a machine learning method, to obtain forest path recognition results in a high forest canopy density area, and to fuse the forest path recognition results with the preliminary forest road recognition results to obtain a forest area fine road network;

[0016] The forest road network model acquisition module is used to obtain high-precision forest surface elevation information based on SAR data, integrate DEM data to extract forest terrain and geomorphic features, and combine the forest fine road network to build a forest road network model based on multi-source data.

[0017] In a third aspect, the present invention provides an electronic device comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned method for constructing a forest road network model that takes into account terrain constraints and tree canopy shading.

[0018] In a fourth aspect, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the aforementioned method for constructing a forest road network model that takes into account terrain constraints and tree canopy shading.

[0019] The beneficial effects of the present invention are:

[0020] 1. Solve the problem of difficult identification of understory paths: Combine time-series visible light data with SAR data to identify understory paths. On the one hand, by using the difference in forest canopy density in different seasons in the forest area, feature analysis is performed on time-series visible light remote sensing images of different seasons to achieve high-precision identification of understory paths in deciduous forest areas; on the other hand, the penetration characteristics of SAR data in different bands are used to capture the weak scattering signals of understory paths, so as to achieve feature identification of understory paths suitable for different forest areas.

[0021] 2. Terrain constraints and road network integration: Terrain constraints are fully considered in the process of road network construction. DEM data is used to obtain terrain information such as the topography, slope, and slope direction of the road network. At the same time, SAR data is combined to extract detailed surface elevation information to ensure the accuracy of the road network model and improve the rationality of road network planning.

[0022] 3. Engineering application value: The present invention has significant engineering application value, and can provide a feasible technical solution for understory path identification, and provide strong technical support for practical work such as forestry resource management, forest fire prevention and ecological protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flow chart of a method for constructing a forest road network model taking into account terrain constraints and tree canopy shading according to the present invention;

[0024] Figure 2 Construct a flow chart for a sample data set of road characteristics in a typical forest area. DETAILED DESCRIPTION

[0025] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0026] The main purpose of the present invention is to provide a method and device for constructing a forest road network model that takes into account terrain constraints and tree canopy shading. In view of the problems of canopy shading and complex terrain in forest road identification, based on the differential characteristics of time-series visible light data in winter and summer, the penetration ability of SAR data, and the terrain constraint information of DEM data, high-precision identification of understory paths and construction of a fine forest road network are achieved, providing reliable support for forest road extraction and practical application.

[0027] like Figure 1 As shown, the method for constructing a forest road network model considering terrain constraints and tree canopy shading of the present invention includes the following steps:

[0028] Step S1: Collect multi-source data of typical forest areas, annotate the roads in the data, and construct a data sample set containing the characteristics of typical forest area roads. The multi-source data includes satellite remote sensing visible light, synthetic aperture radar (SAR) data, and digital elevation model (DEM) data.

[0029] Step S2: For the data sample set constructed based on satellite remote sensing visible light data, analyze the differences in forest stand canopy density in different seasons in the forest area, use machine learning methods to train a machine learning network model adapted to satellite remote sensing visible light data in the forest area, and achieve preliminary recognition of forest road information based on satellite remote sensing visible light data.

[0030] Step S3: For the data sample set constructed based on SAR data, the long-wave SAR data and the short-wave SAR data are integrated to construct the forest path recognition model data sample set, and the machine learning method is used to train the machine learning network model suitable for forest path recognition to obtain the forest path recognition results in the high forest canopy density area. The obtained forest path recognition results are integrated with the preliminary identification results of the forest roads obtained in step S2 to obtain the forest area fine road network containing the forest paths.

[0031] Step S4: Based on satellite remote sensing SAR data, high-precision forest area surface elevation information is obtained, DEM data is integrated to extract forest area topographic features, and combined with forest area fine road network information, a forest area road network model based on multi-source data is constructed.

[0032] Specifically, the step S1 includes:

[0033] Since the multi-source data includes satellite remote sensing visible light data, SAR data and DEM data, and the data acquisition methods and processing procedures of different data sources are similar, the present invention selects visible light data, SAR data and DEM data of typical forest areas for processing in step S1 (the data sources not mentioned can be operated according to similar procedures and will not be repeated here), and uses manual labeling, semi-automatic labeling and deep learning-based labeling methods to label the typical forest area data roads to form a road labeling data sample set. The process is as follows Figure 2 As shown, the specific steps include:

[0034] Step S1.1: Collect satellite remote sensing visible light data, SAR data and DEM data of typical forest areas, including evergreen forests and deciduous forests.

[0035] Step S1.2: Manual labeling and semi-automatic labeling are used to label forest roads in satellite remote sensing visible light and SAR data. The manual labeling method uses ArcGIS or QGIS software to manually draw road lines directly on the image data; the semi-automatic labeling method combines computer assistance and manual correction, and can first preliminarily extract road contours through image processing algorithms, and use LabelMe's polygon tool to adjust and improve these contours, and assign different labels to different roads and other elements.

[0036] Step S1.3: Based on the collected satellite remote sensing visible light, SAR data, DEM data and annotated road information, a data sample set of typical forest area road characteristics is constructed.

[0037] The step S2 comprises:

[0038] Based on the data sample set constructed from satellite remote sensing visible light data, the differences in forest stand canopy density in different seasons are analyzed, and a machine learning network model adapted to satellite remote sensing visible light data in forest areas is trained using machine learning methods to achieve preliminary recognition of forest road information based on satellite remote sensing visible light data. The core algorithm of the machine learning method is convolutional neural network (CNN), including any of the R-CNN, DCNN, FCN, YOLO, ResNet, U-Net, SegNet, DeepLab series, and Transformer-based models.

[0039] The specific steps include:

[0040] Step S2.1: Construct a preliminary forest road identification data sample set based on satellite remote sensing visible light data and perform preprocessing.

[0041] The preliminary identification data sample set of forest roads constructed by satellite remote sensing visible light data is preprocessed by normalizing spatial resolution, trimming data and enhancing data, and the preprocessed data set is divided into three parts: training set, validation set and test set. The specific steps are as follows:

[0042] The preprocessing of the data sample set constructed by satellite remote sensing visible light data includes spatial resolution normalization and data clipping and enhancement. First, the spatial resolution of the data sample set constructed by satellite remote sensing visible light data is normalized, and the spatial resolution is unified to 0.5 meters after processing. Secondly, data clipping and enhancement are performed, and the data samples are processed by geometric transformation methods, including random angle rotation, horizontal and vertical flipping; optical enhancement is also used to adjust the brightness, contrast and saturation of the remote sensing image; the acquired data samples are clipped to 256 The data sample is 256 pixels. The obtained data sample is named ForestRoad_Dateset dataset.

[0043] The ForestRoad_Dateset dataset is divided into training set, validation set and test set in a ratio of 7:2:1 to ensure that each subset contains multi-type samples of different seasons, different terrains and different road conditions.

[0044] Step S2.2: Construction of a machine learning network model suitable for forest area satellite remote sensing visible light data.

[0045] The differences in forest canopy density in different seasons in the forest area were analyzed, and a machine learning network model suitable for satellite remote sensing visible light data in the forest area was constructed based on a typical convolutional neural network (CNN). First, the convolutional layer, pooling layer, activation layer and fully connected layer structure of the machine learning network model suitable for forest road recognition were designed, and the update step size, batch size, number of training rounds and initialization model parameters of the model parameters were set. The training set in the ForestRoad_Dateset dataset was input into the machine learning network model, and the model performance was evaluated on the validation set. The loss function was used to calculate the loss between the predicted forest road results and the true label, and the gradient of the loss to the model parameters was calculated. The optimizer was used to update the model weights according to the gradient. The above steps were iterated until all training rounds were completed, and the construction of the machine learning network model for forest road recognition was realized.

[0046] The accuracy, precision, and recall indicators are used to evaluate the performance of the model on the test set. The accuracy indicator evaluation model is:

[0047] ,

[0048] The precision index evaluation model is:

[0049] ,

[0050] The recall rate indicator evaluation model is:

[0051] , where TP is the correct positive prediction, FP is the wrong positive prediction, TN is the correct negative prediction, and FN is the wrong negative prediction.

[0052] Step S2.3: Preliminary identification of forest road information based on satellite remote sensing visible light data.

[0053] After obtaining a machine learning network model suitable for satellite remote sensing visible light data in forest areas, the visible light data to be identified is input to obtain preliminary identification results of forest surface vegetation information and forest road information.

[0054] The step S3 comprises:

[0055] Based on satellite remote sensing SAR data, long-wave SAR data and short-wave SAR data are integrated to construct a sample set of forest path recognition model data. The machine learning method is used to train the machine learning network model for forest path recognition to obtain forest road information in high forest canopy density areas. The obtained forest road information is integrated with the preliminary identification results of the forest road information obtained in step S2 to obtain a fine forest road network containing forest paths. Specifically, the following steps are included:

[0056] Step S3.1: constructing a training data sample set for the understory path recognition model;

[0057] Based on the data sample set constructed by satellite remote sensing SAR data, the long-wave and short-wave SAR data are spatially matched, and polarization correction, geometric correction and data enhancement are carried out. The long-wave band includes P and L bands, and the short-wave band includes X and C bands. The satellite remote sensing long-wave SAR data and short-wave SAR data are fused to comprehensively obtain the crown information and understory information of forest vegetation. At the same time, the path information obtained by the winter phase satellite remote sensing visible light data is used to construct the data sample set of the understory path recognition model.

[0058] The data sample set of the forest path recognition model is enhanced by rotating, flipping, scaling and other operations to expand the size of the data set. The resulting data sample set is named ForestRoad_SAR_Dateset dataset.

[0059] The ForestRoad_SAR_Dateset dataset was divided into training set, validation set and test set in a ratio of 7:2:1 to ensure that each subset contained multi-type samples with different forest density, different terrains and different road conditions.

[0060] Step S3.2: Construction of machine learning network model for understory path identification;

[0061] Based on the typical convolutional neural network (CNN), the machine learning network model for forest path identification is constructed. First, the convolutional layer, pooling layer, activation layer and fully connected layer structure of the machine learning network model for forest path identification are designed, and the update step size, batch size, training round number and initialization model parameters of the model parameters are set. The training data sample set of the forest path identification model is input into the machine learning network model, and the model performance is evaluated on the validation set. The loss function is used to calculate the loss between the predicted forest path result and the true label, and the gradient of the loss to the model parameters is calculated. The optimizer is used to update the model weight according to the gradient. The above steps are iterated until all training rounds are completed, and the construction of the machine learning network model for forest path identification can be realized.

[0062] Use accuracy, precision, and recall to evaluate the performance of the model on the test dataset. See step S2.2 for the specific equation.

[0063] Step S3.3: Identification of understory paths based on satellite remote sensing SAR data;

[0064] Obtain satellite remote sensing SAR data of the area where understory path identification is needed, input the data into the machine learning network model of understory path identification, and obtain the understory path identification results in areas with high forest density.

[0065] Step S3.4: Construction of forest area fine road network model;

[0066] The obtained understory path recognition result is combined with the preliminary recognition result of the forest road information obtained in step S2 to obtain a fine forest road network including understory paths. The specific steps are as follows:

[0067] Based on the obtained preliminary identification results of forest road information and forest path identification results, the geographic coordinate systems of the two are converted to the WGS-84 geographic coordinate system, and the preliminary identification results of forest road information and forest path identification results are registered using the control point matching method or automatic image registration algorithm, and road intersections, endpoints or other highly recognizable feature points are selected as control points. The buffer matching method is used to match the center line and optimize the matching of road coordinate information.

[0068] After the preliminary identification results of forest road information and the identification results of forest paths are matched, the roads are vectorized, the coordinates of all breakpoints and endpoints of discontinuous areas are extracted, the distance matching method is used to determine the possible connection paths, and the morphological methods of expansion and closing operations are applied to fill the gaps between roads to generate continuous candidate connection paths. The algorithm searches for the optimal path among the candidate connection paths and completes the connection between the endpoints. Through the geometric constraints and attribute characteristics of the road in terms of direction, length, and angle, the wrong connections are eliminated, and the construction of a fine road network in the forest area including understory paths is realized.

[0069] The step S4 comprises:

[0070] Based on satellite remote sensing SAR data, high-precision forest surface elevation information is obtained, DEM data is integrated to extract topographic features, and combined with forest road network information, a forest road network information model based on multi-source data is constructed.

[0071] First, extract the surface elevation information of the forest area from the SAR data, and combine it with the surface elevation information of the DEM data to obtain the topographical features of the forest area. Spatially align and fuse the topographical features of the forest area with the fine road network of the forest area to generate a comprehensive road network feature containing terrain constraints; finally, construct a forest road network information model. The topographical features of the forest area include terrain slope, slope direction, undulation, relative height, terrain roughness, and curvature, and specifically include the following steps:

[0072] Step S4.1: Extracting forest area topographic features.

[0073] The InSAR technology is used to extract high-precision forest ground elevation data from SAR data. The DEM data is preprocessed to obtain reference DEM data; the forest ground elevation data and the reference DEM data are registered and fused using the registration function, including: resampling the DEM data using bilinear or cubic convolution interpolation, and performing coordinate system conversion so that it has the same spatial resolution and projection system as the high-precision forest ground elevation data. The two types of data are accurately registered using the registration function to obtain registered forest ground elevation data. The registration function can be expressed as:

[0074] ,

[0075] In the formula is mutual information, is the marginal entropy, is the joint entropy.

[0076] Based on the registered forest area ground elevation data, the topographic and geomorphic characteristics of the forest area are obtained, including terrain slope, slope aspect, undulation, relative height, terrain roughness and curvature.

[0077] Step S4.2: Constructing a forest road network information model based on multi-source data.

[0078] Based on the acquired forest area fine road network and forest area topography and geomorphic features, the geographic coordinate systems of the two are converted into the WGS-84 geographic coordinate system, and the control point matching method is used to align the forest area fine road network and forest area topography and geomorphic features, and road intersections, endpoints or other highly recognizable feature points are selected as control points. After alignment, the forest area fine road network and forest area topography and geomorphic feature data are fused to realize the construction of a forest area road network model based on multi-source data.

[0079] In summary, the method for constructing a forest road network model taking into account tree crown shading and terrain constraints provided by the present invention can achieve high-precision identification of understory paths, extraction of terrain and geomorphic features, and construction of a fine road network in the forest area, providing important data support and technical reference for forest fire fighting and rescue.

[0080] On the other hand, the present invention provides a device for constructing a forest road network model taking into account terrain constraints and tree canopy shading, wherein the modules included in the device can implement the various steps of the aforementioned method, specifically including:

[0081] A collection module is used to collect multi-source data of typical forest areas, and after marking the roads in the multi-source data, construct a data sample set containing the characteristics of the roads in the typical forest areas, wherein the multi-source data includes satellite remote sensing visible light data, SAR data and DEM data;

[0082] The forest road recognition module is used to train a machine learning network model suitable for forest road recognition based on a data sample set constructed based on satellite remote sensing visible light data, using a machine learning method, and obtain preliminary forest road recognition results based on satellite remote sensing visible light data;

[0083] A forest area fine road network acquisition module is used to construct a forest path recognition model data sample set for a data sample set constructed based on SAR data by fusing long-wave SAR data and short-wave SAR data, and to train a machine learning network model adapted to the forest path recognition model data sample set using a machine learning method, to obtain forest path recognition results in a high forest canopy density area, and to fuse the forest path recognition results with the preliminary forest road recognition results to obtain a forest area fine road network;

[0084] The forest road network model acquisition module is used to obtain high-precision forest surface elevation information based on SAR data, integrate DEM data to extract forest terrain and geomorphic features, and combine the forest fine road network to build a forest road network model based on multi-source data.

[0085] In a third aspect, the present invention provides an electronic device comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned method for constructing a forest road network model that takes into account terrain constraints and tree canopy shading.

[0086] In a fourth aspect, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the aforementioned method for constructing a forest road network model that takes into account terrain constraints and tree canopy shading.

[0087] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing a forest road network model considering terrain constraints and tree canopy shading, characterized in that: The method comprises: Step 1: Collect multi-source data of typical forest areas, annotate roads in the multi-source data, and construct a data sample set containing road features of typical forest areas, wherein the multi-source data includes satellite remote sensing visible light data, SAR data, and DEM data; Step 2: Analyze the differences in forest stand canopy density in different seasons in the forest area based on the data sample set constructed based on satellite remote sensing visible light data, use machine learning methods to train a machine learning network model suitable for forest road recognition, and obtain preliminary forest road recognition results based on satellite remote sensing visible light data; Step 3: for the data sample set constructed based on SAR data, the long-wave SAR data and the short-wave SAR data are integrated to construct a forest path recognition model data sample set, and a machine learning network model adapted to the forest path recognition model data sample set is trained by a machine learning method to obtain the forest path recognition result in the high forest canopy density area, and the forest path recognition result is integrated with the preliminary recognition result of the forest road to obtain a fine road network in the forest area; Step 4: Obtain high-precision forest area surface elevation information based on SAR data, fuse DEM data to extract forest area topographic features, and combine with the forest area's fine road network to build a forest area road network model based on multi-source data.

2. A method for constructing a forest road network model considering terrain constraints and tree canopy shading according to claim 1, characterized in that: The step 1 comprises: Step 1.1, collecting satellite remote sensing visible light data, SAR data and DEM data of typical forest areas, wherein the typical forest areas include evergreen forests and deciduous forests; Step 1.2, using manual labeling or semi-automatic labeling to label the roads in the satellite remote sensing visible light data and SAR data; Step 1.3: Construct the corresponding data sample set based on the collected satellite remote sensing visible light data, SAR data, DEM data and annotated road information.

3. The method for constructing a forest road network model considering terrain constraints and tree canopy shading according to claim 1, characterized in that: The step 2 comprises: Step 2.1: Preprocess the data sample set constructed based on satellite remote sensing visible light data, including spatial resolution normalization, data clipping and data enhancement, and divide the preprocessed data set into a training set, a validation set and a test set; Step 2.2, analyzing the differences in forest stand canopy density in different seasons in the forest area, constructing a machine learning network model suitable for forest road recognition based on a convolutional neural network, inputting the training set into the machine learning network model, evaluating the model performance on the validation set, using a loss function to calculate the loss between the predicted forest road result and the true label, calculating the gradient of the loss to the model parameters, using an optimizer to update the model weights according to the gradient, looping and iterating until all training rounds are completed, and obtaining a machine learning network model for forest road recognition; Step 2.3: Input the visible light data to be identified into the forest road identification machine learning network model to obtain preliminary identification results including forest surface vegetation information and forest road information.

4. The method for constructing a forest road network model considering terrain constraints and tree canopy shading according to claim 1, characterized in that: The step 3 comprises: Step 3.1: For the data sample set constructed based on SAR data, spatial matching is performed on the long-wave and short-wave SAR data, and polarization correction, geometric correction and data enhancement are performed. At the same time, the path information obtained from the winter phase satellite remote sensing visible light data is used to construct a data sample set for the understory path recognition model. After preprocessing, the data sample set is divided into a training set, a validation set and a test set in proportion. Step 3.2, construct a machine learning network model suitable for understory path recognition based on a convolutional neural network, input the training set of the understory path recognition model into the machine learning network model, evaluate the model performance on the validation set, use the loss function to calculate the loss between the predicted understory path result and the true label, calculate the gradient of the loss to the model parameters, use the optimizer to update the model weight according to the gradient, iterate until all training rounds are completed, and obtain the understory path recognition machine learning network model; Step 3.3, obtain the SAR data of the area where understory path identification is required and input it into the understory path identification machine learning network model to obtain the understory path identification results in the high forest canopy density area; Step 3.4: merge the understory path recognition result with the preliminary forest road recognition result to obtain a fine forest road network including the understory paths.

5. The method for constructing a forest road network model considering terrain constraints and tree canopy shading according to claim 4, characterized in that: The step 3.4 comprises: The geographic coordinate system of the obtained preliminary identification results of forest roads and forest paths is converted into the WGS-84 geographic coordinate system, and the preliminary identification results of forest roads and forest paths are registered by using the control point matching method or the automatic image registration algorithm; After completing the preliminary identification results of forest roads and the registration of forest path identification results, the roads are vectorized, the coordinates of all breakpoints and endpoints of discontinuous areas are extracted, the distance matching method is used to determine the possible connection paths, and the morphological methods of expansion and closing operations are applied to fill the gaps between roads to generate continuous candidate connection paths; Dijkstra algorithm or A The algorithm searches for the optimal path among the candidate connection paths and completes the connection between the endpoints. Through the geometric constraints and attribute characteristics of the road in direction, length, and angle, it eliminates incorrect connections and realizes the construction of a fine road network in the forest area including understory paths.

6. The method for constructing a forest road network model considering terrain constraints and tree canopy shading according to claim 1, characterized in that: The step 4 comprises: Step 4.1, using InSAR technology to extract high-precision forest area ground elevation data from SAR data, preprocessing the DEM data to obtain reference DEM data; using a registration function to register and fuse the forest area ground elevation data and the reference DEM data to obtain registered forest area ground elevation data; based on the registered forest area ground elevation data, obtaining the topographic and geomorphic features of the forest area; Step 4.2: Based on the acquired forest area fine road network and forest area topographic and geomorphic features, the control point matching method is used to align the forest area fine road network and forest area topographic and geomorphic features. After the alignment, the forest area fine road network and forest area ground elevation data are fused to realize the construction of the forest area road network model based on multi-source data.

7. A method for constructing a forest road network model considering terrain constraints and tree canopy shading according to claim 3 or 4, characterized in that: The performance of the constructed machine network learning model on the test set is evaluated using accuracy, precision, and recall indicators.

8. A device for constructing a forest road network model taking into account terrain constraints and tree canopy shading, characterized in that: include: A collection module is used to collect multi-source data of typical forest areas, and after marking the roads in the multi-source data, construct a data sample set containing the characteristics of the roads in the typical forest areas, wherein the multi-source data includes satellite remote sensing visible light data, SAR data and DEM data; The forest road recognition module is used to train a machine learning network model suitable for forest road recognition based on a data sample set constructed based on satellite remote sensing visible light data, using a machine learning method, and obtain preliminary forest road recognition results based on satellite remote sensing visible light data; A forest area fine road network acquisition module is used to construct a forest path recognition model data sample set for a data sample set constructed based on SAR data by fusing long-wave SAR data and short-wave SAR data, and to train a machine learning network model adapted to the forest path recognition model data sample set using a machine learning method, to obtain forest path recognition results in a high forest canopy density area, and to fuse the forest path recognition results with the preliminary forest road recognition results to obtain a forest area fine road network; The forest road network model acquisition module is used to obtain high-precision forest surface elevation information based on SAR data, integrate DEM data to extract forest terrain and geomorphic features, and combine the forest fine road network to build a forest road network model based on multi-source data.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; Wherein, when one or more programs are executed by the one or more processors, the one or more processors implement a method for constructing a forest road network model taking into account terrain constraints and tree canopy shading as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: Executable instructions are stored thereon, which, when executed by a processor, enable the processor to implement a method for constructing a forest road network model that takes into account terrain constraints and tree canopy shading as described in any one of claims 1-7.

Images