A Method and System for Extracting Erosion Gullies on the Loess Plateau Based on Deep Learning

Abstract

The present invention relates to a method and system for extracting erosion gullies on the Loess Plateau based on deep learning, and relates to the field of remote sensing image extraction. The method includes: obtaining a first remote sensing image of the Loess Plateau; preprocessing the first remote sensing image to obtain a second remote sensing image; constructing a data set according to the second remote sensing image; constructing a deep learning model with dual-end input, where the deep learning model is used to extract erosion gully information on the Loess Plateau; training the deep learning model with the data set to obtain a trained deep learning model; obtaining a third remote sensing image of the Loess Plateau to be measured; and using the trained deep learning model to extract erosion gullies from the third remote sensing image. The present invention can obtain a high-precision erosion gully information extraction result.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing image extraction, and particularly to a method and system for extracting erosion gullies on the Loess Plateau based on deep learning. Background Art

[0002] The Loess Plateau has complex and unique surface landscape features. Affected by external forces such as wind erosion and water erosion, there is relatively serious surface erosion, and erosion gullies are widely distributed. Traditional methods for extracting erosion gullies based on digital terrain analysis and remote sensing interpretation rely too much on the experience of professionals and high-precision interpretation data, cannot meet the actual production needs, and have low automation and strong limitations. Deep learning methods can use end-to-end extraction methods to intelligently and automatically extract erosion gully information on the Loess Plateau, reducing time-consuming and inefficient work such as field mapping, on-site investigation, and remote sensing interpretation. This not only improves the efficiency of erosion gully extraction but also effectively saves cost expenditures. The existing erosion gully extraction methods have the following deficiencies:

[0003] (1) Traditional remote sensing interpretation methods rely too much on the professional knowledge of interpreters and have low universality.

[0004] (2) The method for extracting erosion gullies based on digital terrain analysis and remote sensing interpretation needs to combine a large amount of field mapping information, cannot give full play to the advantages of historical data, and leads to waste of resources.

[0005] (3) Currently, the existing erosion gully extraction methods based on deep learning models focus on improving accuracy, ignoring extraction efficiency, wasting computing resources, taking a long time, and having low efficiency. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and system for extracting erosion gullies on the Loess Plateau based on deep learning, which can obtain high-precision erosion gully information extraction results.

[0007] To achieve the above purpose, the present invention provides the following solutions:

[0008] A method for extracting erosion gullies on the Loess Plateau based on deep learning includes:

[0009] Obtaining a first remote sensing image of the Loess Plateau;

[0010] Preprocessing the first remote sensing image to obtain a second remote sensing image;

[0011] Constructing a data set according to the second remote sensing image;

[0012] Constructing a deep learning model based on dual-end input, where the deep learning model is used to extract erosion gully information on the Loess Plateau;

[0013] Train the deep learning model using the dataset to obtain a trained deep learning model;

[0014] Obtain the third remote sensing image of the Loess Plateau to be measured;

[0015] Use the trained deep learning model to extract erosion gullies from the third remote sensing image.

[0016] Optionally, the data source of the first remote sensing image uses the 16th-level image in Google Earth.

[0017] Optionally, the dataset includes a second remote sensing image and the DEM, slope, edge information, and texture information of the second remote sensing image.

[0018] Optionally, use the Canny edge detection operator to extract the edge information of the second remote sensing image.

[0019] Optionally, calculate the texture information of the second remote sensing image based on the gray-level co-occurrence matrix; the texture information includes inverse difference, contrast, and energy.

[0020] Optionally, the deep learning model includes a first input end, a second input end, and an output end;

[0021] The first input end and the second input end respectively include five encoder modules connected in sequence, and the output end includes five decoder modules connected in sequence;

[0022] The encoder modules of the first input end and the second input end are respectively connected to the decoder modules of the output end in a corresponding manner;

[0023] The first input end is used to input the second remote sensing image, and the second input end is used to input the DEM, slope, edge information, and texture information.

[0024] Optionally, the number of channels of the depthwise separable convolution in the five encoder modules are 32, 64, 128, 256, and 256 respectively, the convolution kernel size of the per-channel convolution is 3*3*C i-1 , and the convolution kernel size of the pointwise convolution is 1*1*C i , where C i-1 is the number of channels of the depthwise separable convolution output by the upper layer, and C i is the number of channels of the depthwise separable convolution output by this layer.

[0025] Optionally, the number of convolution kernel channels of the convolutional layers in the five decoder modules are 32, 64, 128, 256, and 256 respectively.

[0026] A system for extracting erosion gullies on the Loess Plateau based on deep learning, comprising:

[0027] The first remote sensing image acquisition module is used to acquire the first remote sensing image of the Loess Plateau;

[0028] The second remote sensing image acquisition module is used to preprocess the first remote sensing image to obtain a second remote sensing image;

[0029] The dataset construction module is used to construct a dataset based on the second remote sensing image;

[0030] The deep learning model construction module is used to construct a deep learning model with dual - end input, and the deep learning model is used to extract erosion gully information of the Loess Plateau;

[0031] The training module is used to train the deep learning model with the dataset to obtain a trained deep learning model;

[0032] The third remote sensing image acquisition module is used to acquire the third remote sensing image of the Loess Plateau to be measured;

[0033] The erosion gully extraction module is used to extract erosion gullies from the third remote sensing image with the trained deep learning model.

[0034] A computer - readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the method for extracting erosion gullies of the Loess Plateau based on deep learning.

[0035] According to the specific embodiments provided by the present invention, the following technical effects are disclosed:

[0036] The present invention uses a depth - separable convolutional layer to replace the traditional convolutional layer and combines residual connection processing to establish an erosion gully information extraction model, which not only effectively reduces the model complexity, but also improves the model convergence speed and extraction accuracy. At the same time, the deep learning model with dual - end input can also combine shallow features such as texture and shape extracted from remote sensing images to achieve information fusion at the dimension level, integrating the semantic information of deep features and the texture edge information of shallow features, and finally obtaining a high - precision erosion gully information extraction result. Description of the Drawings

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the following - described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained without creative efforts based on these drawings.

[0038] Figure 1 It is a flowchart of the method for extracting erosion gullies of the Loess Plateau based on deep learning of the present invention;

[0039] Figure 2 Schematic diagram of the deep learning model structure based on dual - end input of the present invention;

[0040] Figure 3 Schematic diagram of the encoder structure of the present invention;

[0041] Figure 4 Schematic diagram of the decoder structure of the present invention;

[0042] Figure 5 Schematic diagram of the gully extraction result of the present invention. Specific embodiments

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0044] Since the method for gully extraction based on remote sensing interpretation cannot achieve the automatic extraction of gully information, this method cannot give full play to the role of historical data, resulting in waste of resources such as manpower and material resources. In addition, although the deep learning model can achieve gully information extraction in an end - to - end manner, the pursuit of high - precision extraction results ignores the impact of model complexity and convergence speed on the model, thus unable to fully utilize the advantages of automation and intelligence of the deep learning model.

[0045] Therefore, by using depth - separable convolutional layers to replace traditional convolutional layers and combining residual connection processing to establish a gully information extraction model, not only is the model complexity effectively reduced, but also the model convergence speed and extraction accuracy are improved. At the same time, the deep learning model based on dual - end input can also combine shallow features such as texture and shape extracted from remote sensing images to achieve information fusion at the dimension level, integrating the semantic information of deep features and the texture edge information of shallow features, and finally being able to obtain high - precision gully information extraction results.

[0046] To sum up, the purpose of the present invention is to provide a method and system for gully extraction on the Loess Plateau based on deep learning, which can obtain high - precision gully information extraction results.

[0047] To make the above - mentioned objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0048] Figure 1 Flowchart of the method for gully extraction on the Loess Plateau based on deep learning of the present invention, as Figure 1As shown in the figure, a method for extracting erosion gullies on the Loess Plateau based on deep learning includes:

[0049] Step 1: Obtain the first remote sensing image of the Loess Plateau.

[0050] Step 2: Preprocess the first remote sensing image to obtain a second remote sensing image.

[0051] Step 3: Construct a dataset based on the second remote sensing image.

[0052] Among them, in order to obtain a high-precision erosion gully extraction result, select the 16th-level image in Google Earth as the remote sensing data source of the first remote sensing image and preprocess it. This image has a spatial resolution of 1.92m.

[0053] Furthermore, combining the DEM, slope, edge information, and texture information of the image, establish an erosion gully information extraction dataset. Among them, the edge information uses the Canny edge detection operator to extract the edge information of the remote sensing image. The texture features select the inverse difference, contrast, and energy of the eigenvalues calculated based on the gray-level co-occurrence matrix (GLCM). Among them, the inverse difference is used to measure the degree of local change in texture, and the larger its value, the smaller the local texture change in the image (Equation 1); the contrast reflects the clarity of the image and the depth of texture grooves (Equation 2); the energy is used to describe the trend of image texture change and is a standard for measuring the uniformity of image gray distribution and the thickness of texture (Equation 3);

[0054]

[0055] Con = ∑ i ∑ j (i - j) 2 g(i, j) (2)

[0056] Ene = ∑ i ∑ j g(i, j) 2 (3)

[0057] In the formula, Hom represents the inverse difference; Con represents the contrast; Ene represents the energy; i and j represent the row and column numbers of the gray-level co-occurrence matrix; g(i, j) represents the value of the i-th row and the j-th column.

[0058] Step 4: Construct a deep learning model based on dual-end input, and the deep learning model is used to extract the erosion gully information on the Loess Plateau.

[0059] Specifically, it includes: constructing an erosion gully information extraction model based on a dual-end input convolutional neural network.

[0060] Depthwise separable convolution is a new basic module for neural networks. Different from traditional convolutional layers, depthwise separable convolution (DSC) combines depthwise convolution and pointwise convolution to use fewer model parameters to improve the feature extraction ability of deep learning models. The traditional U-Net model is optimized using depthwise separable convolution. Combining Dem, slope, edge information, and texture features, a dual-input structure is adopted to build a network model. At the same time, residual connections are introduced to solve the problem of gradient disappearance in the model. The model structure is as Figure 2 shown. In the constructed dual-input convolutional neural network erosion gully information extraction model, through dual-input, integrating the spectral information of the image, DEM, slope, edge information, and texture features, the accuracy of the model for extracting erosion gullies is improved; by using depthwise separable convolution, the number of model parameters is greatly reduced, significantly reducing the model complexity. While ensuring the model accuracy, the model training efficiency is improved; in the feature extraction module, residual connections are used. On the one hand, the generalization ability of the model is improved, enhancing the ability to extract erosion gullies of different forms. On the other hand, the convergence speed of the model is increased, making the optimization of the model more stable.

[0061] Figure 2 In [figure], E1, E2, E3, E4, and E5 represent 5 encoder modules, and D1, D2, D3, D4, and D5 represent 5 decoder modules.

[0062] The encoder structure is as Figure 3 shown. The input data is processed through two depthwise separable convolutions and residual connections. One end is used as the input information for the next encoder, and the other end uses skip connections to fuse context information as the input information for the decoder. For model parameters, the number of channels C of the depthwise separable convolution in the five encoders E1, E2, E3, E4, and E5 is 32, 64, 128, 256, and 256 respectively. The kernel size of the depthwise convolution is 3*3*C i-1 , and the kernel size of the pointwise convolution is 1*1*C i , where C i-1 is the number of channels of the depthwise separable convolution output by the upper layer, and C i is the number of channels of the depthwise separable convolution output by this layer.

[0063] The decoder structure is as Figure 4 shown. The decoder receives three input information, which come from the encoder at the Input1 input end, the encoder at the Input2 input end, and the previous decoder layer respectively. The number of channels of the convolution kernels in the convolutional layers of the decoder is the same as that of the corresponding encoder, which are 32, 64, 128, 256, and 256 respectively.

[0064] Step 5: Use the dataset to train the deep learning model to obtain a trained deep learning model.

[0065] Establish a model training set and parameter settings. In this network model, remote sensing images are input at the Input1 end, and slope, DEM, Canny edge detection, energy, inverse difference, and contrast are input at the Input2 end. The training images and test images are randomly divided according to a 7:3 ratio. During the model training process, the initial learning rate is 0.001, the optimizer is Adaptive Moment Estimation (Adam), and the batch-size is 4.

[0066] In addition, in order to verify the effectiveness of this network model, the present invention also conducts model comparison experiments and accuracy evaluation. Select U-Net with a similar structure and FCN8s with excellent performance (the backbone feature extraction network is Resnet50), and the DeepLabv3 network for the comparison experiment. The comparison results are as Figure 5 shown, where (a) is the remote sensing image, (b) is the true label, (c) is the result extracted by the U-Net network, (d) is the result extracted by the FCN8s (Resnet50) network, (e) is the result extracted by the Deeplabv3 network, and (f) is the result extracted by the dual-input network.

[0067] The experimental results show that using the dual-input model to extract the abstract semantic information of remote sensing images and shallow feature data can effectively improve the accuracy of gully extraction, and fully utilize the rich information contained in the gully in texture features, edge information, and terrain factors. Table 1 shows the accuracy comparison results of different models and the number of model parameters. The quantitative analysis results show that the network model constructed by the present invention has a high extraction accuracy, and the IOU and F1-score reach 93.52%, 63.78%, and 0.7379 respectively, and the number of model parameters is also greatly reduced, less than other deep learning models.

[0068] Table 1 Comparison table of network model accuracy

[0069]

[0070] Step 6: Obtain the third remote sensing image of the Loess Plateau to be measured.

[0071] Step 7: Use the trained deep learning model to extract gullies from the third remote sensing image.

[0072] Based on the above method, the present invention also discloses a gully extraction system for the Loess Plateau based on deep learning, including:

[0073] The first remote sensing image acquisition module is used to acquire the first remote sensing image of the Loess Plateau.

[0074] The second remote sensing image acquisition module is used to preprocess the first remote sensing image to obtain the second remote sensing image.

[0075] The dataset construction module is used to construct a dataset based on the second remote sensing image.

[0076] The deep learning model construction module is used to construct a deep learning model with dual - end input, and the deep learning model is used to extract gully information of the Loess Plateau.

[0077] The training module is used to train the deep learning model using the dataset to obtain the trained deep learning model.

[0078] The third remote sensing image acquisition module is used to acquire the third remote sensing image of the Loess Plateau to be measured.

[0079] The gully extraction module is used to extract gullies from the third remote sensing image using the trained deep learning model.

[0080] The present invention also discloses the following technical effects:

[0081] 1. Using depth - separable convolution instead of traditional convolution processing can improve the generalization ability of the model while reducing the model complexity.

[0082] 2. Combining the deep - layer feature information extracted by the deep learning model and the shallow - layer features such as edges and textures extracted from remote sensing images, giving play to the advantages of the dual - end input model, and improving the accuracy of gully extraction.

[0083] 3. Using the deep learning model to extract gully information can effectively utilize historical data, train a network model based on existing data, initially obtain gully information, and effectively avoid resource waste.

[0084] 4. Using depth - separable convolution instead of traditional convolution processing can improve the feature extraction ability of the model while reducing the model complexity, use fewer parameters to fit complex gully feature information, and improve the efficiency of gully information extraction.

[0085] 5. Using the dual - end input model can make full use of the texture features, edge information and terrain information of gullies, respectively perform semantic information abstraction on remote sensing images and shallow - layer feature information, fuse deep - layer features and shallow - layer features, improve the generalization ability of the model, and enhance the accuracy of the model for gully information extraction.

[0086] 6. Using residual connections in the feature extraction module of the double-ended input model, on the one hand, it improves the generalization ability of the model, enabling the model to adapt to erosion ditches with different morphological features and further enhancing the accuracy of erosion ditch information extraction; on the other hand, it enables the integration of context information, making the optimization process of the model more stable.

[0087] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same and similar parts among the various embodiments, reference can be made to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple. For related parts, reference can be made to the description in the method section.

[0088] Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. A method for extracting erosion gullies on the Loess Plateau based on deep learning, characterized in that, The method for extracting erosion gullies on the Loess Plateau based on deep learning includes: Obtaining a first remote sensing image of the Loess Plateau; Preprocessing the first remote sensing image to obtain a second remote sensing image; Constructing a data set according to the second remote sensing image; Constructing a deep learning model with dual-end input, which is used to extract erosion gully information on the Loess Plateau; Training the deep learning model with the data set to obtain a trained deep learning model; Obtaining a third remote sensing image of the Loess Plateau to be measured; Extracting erosion gullies from the third remote sensing image using the trained deep learning model; The data set includes the second remote sensing image and the DEM, slope, edge information, and texture information of the second remote sensing image; Extracting the edge information of the second remote sensing image using the Canny edge detection operator; Calculating the texture information of the second remote sensing image based on the gray-level co-occurrence matrix; the texture information includes inverse difference, contrast, and energy; The deep learning model includes a first input end, a second input end, and an output end; The first input end and the second input end respectively include five encoder modules connected in sequence, and the output end includes five decoder modules connected in sequence; The encoder modules of the first input end and the second input end are respectively connected to the decoder modules of the output end in a corresponding manner; The first input end is used to input the second remote sensing image, and the second input end is used to input DEM, slope, edge information, and texture information.

2. The method for extracting gully erosion in the Loess Plateau based on deep learning according to claim 1, wherein The data source of the first remote sensing image uses the 16th-level image in Google Earth.

3. The method for extracting erosion gullies on the Loess Plateau based on deep learning according to claim 1, wherein The number of channels of depthwise separable convolution in the five encoder modules are 32, 64, 128, 256, and 256 respectively. The kernel size of per-channel convolution is 3*3*C i-1 , and the kernel size of pointwise convolution is 1*1*C i , where, C i-1 is the number of channels of depthwise separable convolution output by the upper layer, and C i is the number of channels of depthwise separable convolution output by this layer.

4. The method for extracting gully erosion on the Loess Plateau based on deep learning according to claim 1, characterized in that The number of convolution kernel channels of the convolution layer in the five decoder modules is 32, 64, 128, 256, and 256 respectively.

5. A loess plateau gully extraction system based on deep learning is applied to the loess plateau gully extraction method based on deep learning according to any one of claims 1-4, characterized in that, The system for extracting erosion gullies on the Loess Plateau based on deep learning includes: A first remote sensing image acquisition module, which is used to obtain a first remote sensing image of the Loess Plateau; A second remote sensing image acquisition module, which is used to preprocess the first remote sensing image to obtain a second remote sensing image; A data set construction module, which is used to construct a data set according to the second remote sensing image; A deep learning model construction module, which is used to construct a deep learning model with dual-end input, and the deep learning model is used to extract erosion gully information on the Loess Plateau; A training module, which is used to train the deep learning model with the data set to obtain a trained deep learning model; A third remote sensing image acquisition module, which is used to obtain a third remote sensing image of the Loess Plateau to be measured; An erosion gully extraction module, which is used to extract erosion gullies from the third remote sensing image using the trained deep learning model.

6. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by a processor, it implements the method for extracting erosion gullies on the Loess Plateau based on deep learning according to any one of claims 1-4.

Images