Method and system for identifying and classifying typical rock mass in western region based on remote sensing image

Abstract

The invention discloses a western region typical rock mass recognition and classification method and system based on remote sensing images, and belongs to the technical field of rock mass recognition and classification. The method comprises the following steps: S1, collecting historical remote sensing data of a target area, and preprocessing the historical remote sensing data to obtain identification data; s2, constructing a rock mass recognition and classification model, and training the rock mass recognition and classification model by adopting the recognition data to obtain a final model; and S3, based on the final model, performing identification and classification on the remote sensing data collected in the target area to obtain an identification and classification result. According to the method, rapid and accurate identification and classification of complex geological structures and rock masses in the western region can be realized, and powerful support is provided for geological survey and mineral resource exploration.

Description

Technical Field

[0001] The invention belongs to the technical field of rock mass identification and classification, and in particular relates to a method and system for identifying and classifying typical rock masses in western regions based on remote sensing images. Background Art

[0002] The western region of my country has a unique regional geological environment and is a resource-rich area in my country. It is rich in minerals, land, water, and other resources, and has great development potential. This is an important foundation and favorable condition for the formation of characteristic economy and advantageous industries in the west. The west has significant mineral resource advantages. Although the development cost of some mineral resources is relatively high, mining development has become an important industry in the west.

[0003] Relying on the unique advantages of remote sensing technology such as large range, high efficiency and non-contact data acquisition capabilities, the identification and classification method of typical rock masses in the western region based on remote sensing images is a complex but extremely valuable research field, especially in geological surveys and mineral resource exploration. Summary of the invention

[0004] The present invention aims to propose a method and system for identifying and classifying typical rock masses in the western region based on remote sensing images. By combining remote sensing data and deep learning technology, the method can realize rapid and accurate identification and classification of complex geological structures and rock masses in the western region, providing strong support for geological surveys and mineral resource exploration.

[0005] To achieve the above object, the present invention provides the following scheme: a method for identifying and classifying typical rock masses in the western region based on remote sensing images, comprising the following steps:

[0006] S1, collecting historical remote sensing data of the target area, and preprocessing the historical remote sensing data to obtain identification data;

[0007] S2, constructing a rock mass identification and classification model, and using the identification data to train the rock mass identification and classification model to obtain a final model;

[0008] S3. Based on the final model, the remote sensing data collected in the target area is identified and classified to obtain identification and classification results.

[0009] Further preferably, the pretreatment method in S1 comprises:

[0010] Performing radiation calibration, geometric correction, atmospheric correction, image fusion, geographic registration and data clipping on the historical remote sensing data to obtain pre-processed data;

[0011] The preprocessed data is interpreted to obtain an optimal band combination of the rock mass, and pseudo-color synthesis is performed based on the optimal band combination to obtain the identification data.

[0012] Further preferably, the final model includes: a feature extraction part, a fusion part, a generation part and a classification and recognition part;

[0013] The feature extraction part is used to extract features from the input to obtain a number of feature maps;

[0014] The fusion part is used to perform fusion processing on the feature map to obtain fusion data;

[0015] The generating part is used to divide the fused data into regions to obtain target data;

[0016] The classification and recognition part is used to perform classification operations and recognition operations on the target data to obtain the classification and recognition results.

[0017] Further preferably, the classification and identification part includes two parallel branches: a classification branch and an identification branch;

[0018] The classification branch is used to normalize the target data, and then use a fully connected network to obtain a classification result;

[0019] The recognition branch is used to perform a convolution operation on the target data to obtain a recognition result.

[0020] The present invention also proposes a typical rock mass identification and classification system in the western region based on remote sensing images, comprising: a data processing module, a model training module and an identification module;

[0021] The data processing module is used to collect historical remote sensing data of the target area and pre-process the historical remote sensing data to obtain identification data;

[0022] The model training module is used to construct a rock mass identification and classification model, and train the rock mass identification and classification model using the identification data to obtain a final model;

[0023] The recognition module is used to identify and classify the remote sensing data collected in the target area based on the final model to obtain recognition and classification results.

[0024] Further preferably, the data processing module comprises: a data acquisition unit, a first processing unit and a second processing unit;

[0025] The data acquisition unit is used to collect the historical remote sensing data and real-time remote sensing data of the target area;

[0026] The first processing unit is used to perform radiation calibration, geometric correction, atmospheric correction, image fusion, geographic registration and data clipping on the historical remote sensing data and the real-time remote sensing data to obtain pre-processed data;

[0027] The second processing unit is used to interpret the preprocessed data, obtain the optimal band combination of the rock mass, perform pseudo-color synthesis based on the optimal band combination, and obtain the identification data.

[0028] Further preferably, the final model includes: a feature extraction part, a fusion part, a generation part and a classification and recognition part;

[0029] The feature extraction part is used to extract features from the input to obtain a number of feature maps;

[0030] The fusion part is used to perform fusion processing on the feature map to obtain fusion data;

[0031] The generating part is used to divide the fused data into regions to obtain target data;

[0032] The classification and recognition part is used to perform classification operations and recognition operations on the target data to obtain the classification and recognition results.

[0033] Further preferably, the classification and identification part includes two parallel branches: a classification branch and an identification branch;

[0034] The classification branch is used to normalize the target data, and then use a fully connected network to obtain a classification result;

[0035] The recognition branch is used to perform a convolution operation on the target data to obtain a recognition result.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention combines the advantages of satellite remote sensing and uses a variety of remote sensing data to achieve comprehensive data collection. By combining the optimal band combination of a variety of remote sensing data, a remote sensing image that is closer to the original ground feature is obtained. At the same time, through the method of data dimensionality reduction, the calculation accuracy and speed of the constructed rock mass identification and classification model are guaranteed, so that the complex geological structure and rock mass in the western region can be quickly and accurately identified and classified, providing strong support for geological surveys and mineral resource exploration. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0039] Figure 1 The present invention is a flowchart of a method for identifying and classifying typical rock masses in western China based on remote sensing images according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Embodiment 1:

[0043] like Figure 1 As shown, this embodiment provides a method for identifying and classifying typical rock masses in the western region based on remote sensing images, comprising the following steps:

[0044] S1. Collect historical remote sensing data of the target area and pre-process the historical remote sensing data to obtain identification data.

[0045] In this embodiment, the remote sensing data include: Landsat8 data, GF-2 data and GF-5 data; among them, the Landsat-8 satellite has a high spatial resolution and rich spectral information, so the collected Landsat8 data retains the image details of the ground object information. The spatial resolution of the Gaofen-2 satellite can reach the sub-meter level, and the GF-2 data can better retain the real texture and color information of the original ground object. The Gaofen-5 hyperspectral data has a total of 330 bands, which can realize full-spectrum observation of the atmosphere and land.

[0046] In this embodiment, the remote sensing data is preprocessed to obtain identification data suitable for model application. Specifically, the preprocessing method in S1 includes: performing radiation calibration, geometric correction, atmospheric correction, image fusion, geo-registration and data clipping on the historical remote sensing data to obtain preprocessed data; interpreting the preprocessed data to obtain the optimal band combination of the rock mass, and performing pseudo-color synthesis based on the optimal band combination to obtain identification data.

[0047] Among them, the radiation calibration process is used to eliminate the system error of the sensor itself; the principle is to convert the acquired original image pixel value into the absolute radiation brightness value of the real object spectrum. This embodiment is completed using the radiation calibration tool of ENVI software. Geometric correction is to eliminate the geometric distortion of the image caused by factors such as camera lens and ground undulation through a certain mathematical model. This embodiment uses SRTM with a resolution of 90m for correction. Since the air sensor is affected by atmospheric components such as atmospheric molecules, aerosols and cloud particles during data collection, the data accuracy may not meet the requirements of quantitative analysis, so it is necessary to eliminate the influence of atmospheric components in remote sensing images. This embodiment uses the sen2cor tool to complete atmospheric correction. Multiple remote sensing images are fused according to the vector contour of the target area to eliminate the influence of the image boundary on its quality. By performing geographic registration, the object can be located more accurately. The ROI cropping tool of ENVI is used to select high-quality image segments, and the data features are integrated to obtain the cropping area.

[0048] In order to improve the calculation accuracy and speed of the subsequent rock mass identification and classification model, the preprocessed data obtained after the above processing is first subjected to dimensionality reduction processing. The dimensionality reduction processing method includes: calculating the preprocessed data for feature decomposition to obtain the first matrix, the second matrix and the eigenvalue matrix; extracting the first k eigenvectors in the second matrix, and mapping the preprocessed data to a new matrix form to complete dimensionality reduction.

[0049] After that, the optimal band combination method is used for the preprocessed data that has completed the dimension reduction, the standard deviation and correlation between the bands are calculated, and the optimal band combination is adopted to retain more ground objects and spectral information. Then, pseudo-color synthesis is performed to obtain the recognition data.

[0050] S2. Construct a rock mass identification and classification model, and use the identification data to train the rock mass identification and classification model to obtain the final model.

[0051] In this embodiment, the final model includes four parts: feature extraction part, fusion part, generation part and classification and recognition part; specifically, the feature extraction part is used to extract features from the input to obtain several feature maps; the fusion part is used to fuse the feature maps to obtain fused data; the generation part is used to divide the fused data into regions to obtain target data; the classification and recognition part is used to perform classification and recognition operations on the target data to obtain classification and recognition results. Among them, the classification and recognition part includes two parallel branches: a classification branch and a recognition branch; the classification branch is used to normalize the target data, and then use a fully connected network to obtain the classification result; the recognition branch is used to perform a convolution operation on the target data to obtain the recognition result.

[0052] In this embodiment, the feature extraction part uses the residual network ResNet-50 for feature extraction. For the input recognition data, the feature extraction part performs layer-by-layer feature extraction to obtain the corresponding feature vector. The fusion part scales the feature vector obtained above and the feature vector of the later layer and then performs an addition operation, so that the output feature vector of each layer of the neural network obtained integrates the features of different resolutions and semantic strengths. The fusion part includes a convolutional neural network module, an upsampling module and a fusion module respectively; the convolutional application network module reduces the feature vector size of the feature extraction part to construct a pyramid structure of the output feature vector, and activates the output with the feature of the last residual structure of each stage. The upsampling module uses the difference method to interpolate between pixels on the feature vector and expand the feature vector to its original size. The fusion part is used to radially fuse the feature vector obtained by 1*1 convolution of the feature vector of the convolutional neural network module with the feature vector obtained by the upsampling module to obtain fused data. The generation part uses a regional generation network, which uses an anchor point to scan with a sliding window to define generation boxes of different sizes; the intersection of the label data in the fusion data and the generation box coordinates is calculated, and the intersection threshold range is set. When the intersection is greater than the threshold range, the generation box is defined as the first generation target; when it is within the threshold range, the type is set to the second generation target; when it is less than the threshold range, the type is set to the third generation target; non-maximum suppression processing is used to obtain the final target generation box, and then the target data is obtained. Finally, the classification and recognition part is used to perform classification and recognition processing on it to obtain the classification and recognition results. Among them, the classification branch is a fully connected network plus a classifier softmax; the fully connected network is used to regress the target generation box. Since the fully connected network requires the input size to be consistent, in the process of mapping the generation target generation boxes of different sizes to the feature map of fixed size, the bilinear interpolation method is used to calculate the pixel value at the floating point, thereby eliminating the error between the actual target. The classifier is used to classify the target data and obtain the category probability distribution:

[0053]

[0054] Where p(y=i|x) represents the predicted probability that the input x belongs to category i; c represents the number of categories.

[0055] The recognition branch includes several convolutional networks, pooling layers, and deconvolutional layers. After the target data passes through the convolutional network and pooling layer, the output size is reduced to one-half; after passing through the second convolutional network and pooling layer again, the output size is reduced to one-quarter; and after passing through the fifth convolutional network and pooling layer, the output size is reduced to one-thirty-second. The deconvolutional layer is used to upsample the feature vector obtained by the last convolutional layer to obtain a prediction map with the same size as the input image. After the recognition branch, a feature map with the same size as the original input feature vector can be obtained, realizing pixel-level recognition of remote sensing images.

[0056] S3. Based on the final model, the remote sensing data collected in the target area is identified and classified to obtain the identification and classification results.

[0057] Embodiment 2:

[0058] The present invention also proposes a typical rock mass identification and classification system in the western region based on remote sensing images, comprising: a data processing module, a model training module and a recognition module; the data processing module is used to collect historical remote sensing data of the target area, and pre-process the historical remote sensing data to obtain recognition data; the model training module is used to construct a rock mass identification and classification model, and use the recognition data to train the rock mass identification and classification model to obtain a final model; the recognition module is used to identify and classify the remote sensing data collected in the target area based on the final model to obtain recognition and classification results.

[0059] In this embodiment, the data processing module includes: a data acquisition unit, a first processing unit and a second processing unit; the data acquisition unit is used to collect historical remote sensing data and real-time remote sensing data of the target area; the first processing unit is used to perform radiation calibration, geometric correction, atmospheric correction, image fusion, geographic registration and data cropping on the historical remote sensing data and the real-time remote sensing data to obtain pre-processed data; the second processing unit is used to interpret the pre-processed data, obtain the optimal band combination of the rock mass, perform pseudo-color synthesis based on the optimal band combination, and obtain identification data.

[0060] In this embodiment, the remote sensing data collected by the data acquisition unit include: Landsat8 data, GF-2 data and GF-5 data; among them, the Landsat-8 satellite has a high spatial resolution and rich spectral information, so the collected Landsat8 data retains the image details of the ground object information. The spatial resolution of the Gaofen-2 satellite can reach the sub-meter level, and the GF-2 data can better retain the real texture and color information of the original ground object. The Gaofen-5 hyperspectral data has a total of 330 bands, which can realize full-spectrum observation of the atmosphere and land.

[0061] In this embodiment, the method for preprocessing by the first processing unit includes: performing radiation calibration, geometric correction, atmospheric correction, image fusion, geographic registration and data cropping on historical remote sensing data to obtain preprocessed data; interpreting the preprocessed data to obtain the optimal band combination of the rock mass, and performing pseudo-color synthesis based on the optimal band combination to obtain identification data.

[0062] Among them, the radiation calibration process is used to eliminate the system error of the sensor itself; the principle is to convert the acquired original image pixel value into the absolute radiation brightness value of the real object spectrum. This embodiment is completed using the radiation calibration tool of ENVI software. Geometric correction is to eliminate the geometric distortion of the image caused by factors such as camera lens and ground undulation through a certain mathematical model. This embodiment uses SRTM with a resolution of 90m for correction. Since the air sensor is affected by atmospheric components such as atmospheric molecules, aerosols and cloud particles during data collection, the data accuracy may not meet the requirements of quantitative analysis, so it is necessary to eliminate the influence of atmospheric components in remote sensing images. This embodiment uses the sen2cor tool to complete atmospheric correction. Multiple remote sensing images are fused according to the vector contour of the target area to eliminate the influence of the image boundary on its quality. By performing geographic registration, the object can be located more accurately. The ROI cropping tool of ENVI is used to select high-quality image segments, and the data features are integrated to obtain the cropping area.

[0063] In order to improve the calculation accuracy and speed of the subsequent rock identification and classification model, the second processing unit first performs dimensionality reduction processing on the preprocessed data obtained after the above processing. The dimensionality reduction processing method includes: calculating the preprocessed data for feature decomposition to obtain the first matrix, the second matrix and the eigenvalue matrix; extracting the first k eigenvectors in the second matrix, and mapping the preprocessed data to a new matrix form to complete dimensionality reduction.

[0064] After that, the optimal band combination method is used for the preprocessed data that has completed the dimension reduction, the standard deviation and correlation between the bands are calculated, and the optimal band combination is adopted to retain more ground objects and spectral information. Then, pseudo-color synthesis is performed to obtain the recognition data.

[0065] In this embodiment, the final model includes four parts: feature extraction part, fusion part, generation part and classification and recognition part; specifically, the feature extraction part is used to extract features from the input to obtain several feature maps; the fusion part is used to fuse the feature maps to obtain fused data; the generation part is used to divide the fused data into regions to obtain target data; the classification and recognition part is used to perform classification and recognition operations on the target data to obtain classification and recognition results. Among them, the classification and recognition part includes two parallel branches: a classification branch and a recognition branch; the classification branch is used to normalize the target data, and then use a fully connected network to obtain the classification result; the recognition branch is used to perform a convolution operation on the target data to obtain the recognition result.

[0066] In this embodiment, the feature extraction part uses the residual network ResNet-50 for feature extraction. For the input recognition data, the feature extraction part performs layer-by-layer feature extraction to obtain the corresponding feature vector. The fusion part scales the feature vector obtained above and the feature vector of the later layer and then performs an addition operation, so that the output feature vector of each layer of the neural network obtained integrates the features of different resolutions and semantic strengths. The fusion part includes a convolutional neural network module, an upsampling module and a fusion module respectively; the convolutional application network module reduces the feature vector size of the feature extraction part to construct a pyramid structure of the output feature vector, and activates the output with the feature of the last residual structure of each stage. The upsampling module uses the difference method to interpolate between pixels on the feature vector and expand the feature vector to its original size. The fusion part is used to radially fuse the feature vector obtained by 1*1 convolution of the feature vector of the convolutional neural network module with the feature vector obtained by the upsampling module to obtain fused data. The generation part uses a regional generation network, which uses an anchor point to scan with a sliding window to define generation boxes of different sizes; the intersection of the label data in the fusion data and the generation box coordinates is calculated, and the intersection threshold range is set. When the intersection is greater than the threshold range, the generation box is defined as the first generation target; when it is within the threshold range, the type is set to the second generation target; when it is less than the threshold range, the type is set to the third generation target; non-maximum suppression processing is used to obtain the final target generation box, and then the target data is obtained. Finally, the classification and recognition part is used to perform classification and recognition processing on it to obtain the classification and recognition results. Among them, the classification branch is a fully connected network plus a classifier softmax; the fully connected network is used to regress the target generation box. Since the fully connected network requires the input size to be consistent, in the process of mapping the generation target generation boxes of different sizes to the feature map of fixed size, the bilinear interpolation method is used to calculate the pixel value at the floating point, thereby eliminating the error between the actual target. The classifier is used to classify the target data and obtain the category probability distribution:

[0067]

[0068] Where p(y=i|x) represents the predicted probability that the input x belongs to category i; c represents the number of categories.

[0069] The recognition branch includes several convolutional networks, pooling layers, and deconvolutional layers. After the target data passes through the convolutional network and pooling layer, the output size is reduced to one-half; after passing through the second convolutional network and pooling layer again, the output size is reduced to one-quarter; and after passing through the fifth convolutional network and pooling layer, the output size is reduced to one-thirty-second. The deconvolutional layer is used to upsample the feature vector obtained by the last convolutional layer to obtain a prediction map with the same size as the input image. After the recognition branch, a feature map with the same size as the original input feature vector can be obtained, realizing pixel-level recognition of remote sensing images.

[0070] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A typical rock mass identification and classification method in western China based on remote sensing images, characterized in that: The following steps are involved: S1, collecting historical remote sensing data of the target area, and preprocessing the historical remote sensing data to obtain identification data; S2, constructing a rock mass identification and classification model, and using the identification data to train the rock mass identification and classification model to obtain a final model; S3. Based on the final model, the remote sensing data collected in the target area is identified and classified to obtain identification and classification results.

2. The method for identifying and classifying typical rock masses in western China based on remote sensing images according to claim 1 is characterized in that: The pre-processing method described in S1 includes: Performing radiation calibration, geometric correction, atmospheric correction, image fusion, geographic registration and data clipping on the historical remote sensing data to obtain pre-processed data; The preprocessed data is interpreted to obtain an optimal band combination of the rock mass, and pseudo-color synthesis is performed based on the optimal band combination to obtain the identification data.

3. The method for identifying and classifying typical rock masses in western China based on remote sensing images according to claim 1 is characterized in that: The final model includes: a feature extraction part, a fusion part, a generation part and a classification and recognition part; The feature extraction part is used to extract features from the input to obtain a number of feature maps; The fusion part is used to perform fusion processing on the feature map to obtain fusion data; The generating part is used to divide the fused data into regions to obtain target data; The classification and recognition part is used to perform classification operations and recognition operations on the target data to obtain the classification and recognition results.

4. The method for identifying and classifying typical rock masses in western China based on remote sensing images according to claim 3 is characterized in that: The classification and recognition part includes two parallel branches: a classification branch and a recognition branch; The classification branch is used to normalize the target data, and then use a fully connected network to obtain a classification result; The recognition branch is used to perform a convolution operation on the target data to obtain a recognition result.

5. A typical rock mass identification and classification system in western China based on remote sensing images, the system being used to implement the method described in any one of claims 1 to 4, characterized in that: include: Data processing module, model training module and recognition module; The data processing module is used to collect historical remote sensing data of the target area and pre-process the historical remote sensing data to obtain identification data; The model training module is used to construct a rock mass identification and classification model, and train the rock mass identification and classification model using the identification data to obtain a final model; The recognition module is used to identify and classify the remote sensing data collected in the target area based on the final model to obtain recognition and classification results.

6. The typical rock mass identification and classification system in western China based on remote sensing images according to claim 5 is characterized in that: The data processing module includes: a data acquisition unit, a first processing unit and a second processing unit; The data acquisition unit is used to collect the historical remote sensing data and real-time remote sensing data of the target area; The first processing unit is used to perform radiation calibration, geometric correction, atmospheric correction, image fusion, geographic registration and data clipping on the historical remote sensing data and the real-time remote sensing data to obtain pre-processed data; The second processing unit is used to interpret the preprocessed data, obtain the optimal band combination of the rock mass, perform pseudo-color synthesis based on the optimal band combination, and obtain the identification data.

7. The typical rock mass identification and classification system in western China based on remote sensing images according to claim 5 is characterized in that: The final model includes: a feature extraction part, a fusion part, a generation part and a classification and recognition part; The feature extraction part is used to extract features from the input to obtain a number of feature maps; The fusion part is used to perform fusion processing on the feature map to obtain fusion data; The generating part is used to divide the fused data into regions to obtain target data; The classification and recognition part is used to perform classification operations and recognition operations on the target data to obtain the classification and recognition results.

8. The typical rock mass identification and classification system in western China based on remote sensing images according to claim 7 is characterized in that: The classification and recognition part includes two parallel branches: a classification branch and a recognition branch; The classification branch is used to normalize the target data, and then use a fully connected network to obtain a classification result; The recognition branch is used to perform a convolution operation on the target data to obtain a recognition result.

Images