A two-level fusion method for mountain slope survey data based on gradient decision tree

Abstract

The present invention discloses a two-level fusion method for mountain slope survey data based on a gradient decision tree. The method comprises the following steps: S1: processing mountain slope survey data, standardizing the data according to data structure characteristics, and constructing a unified and standardized representation method for survey data features; S2: associating complementary geophysical field attribute characteristics with slope spatial geometric characteristics to achieve shallow initial fusion of survey data; S3: exploring the latent spatial characteristics of high-dimensional data and the latent surface characteristics of low-latitude data; S4: combining the data obtained in S2 to achieve deep feature fusion of mountain slope survey data; S5: optimizing data, improving data fusion efficiency and accuracy, and achieving multi-level hierarchical intelligent fusion of mountain slope survey data. By establishing intelligent analysis and fusion technology for mountain slope survey data, the present invention solves the problem of low generalization of traditional deep learning algorithms in processing multimodal data.

Description

Technical Field

[0001] The invention belongs to the technical field of mountain slope monitoring and data processing, and relates to an intelligent fusion method for mountain slope survey data. Background Art

[0002] Current mountain slope monitoring and early warning technologies and equipment, primarily based on surface deformation surveys, struggle to accurately identify the mechanisms of slope disasters in complex mountain environments, nor can they discern the mapping relationship between the massive amounts of multi-scale electromagnetic detection information and geotechnical physics and mechanics. This is particularly true for mountain slope hazard hazards where pre-disaster deformation and damage signs are not obvious. There is an urgent need to develop refined detection and enhanced display technologies for identifying mountain slope hazard hazards. Developing such technologies requires improving the fusion and interpretation capabilities of mountain slope survey data under different temporal and spatial conditions, achieving the goals of digitizing, three-dimensionally manipulating, and networking interactive services for mountain slope geological information. This will provide data support for the development of a quantified system for mountain slope geological information and intelligent slope hazard assessment methods. Summary of the Invention

[0003] The purpose of this invention is to provide a two-level fusion method for mountain slope survey data based on gradient decision tree, deconstruct the geological information contained in the survey data under different survey scenarios, and innovate the survey data fusion technology in the mountain slope field.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A two-level fusion method for mountain slope survey data based on a gradient decision tree includes the following steps:

[0006] Step S1: Processing the mountain slope survey data, standardizing the data according to the data structure characteristics, and constructing a unified and standardized representation method for survey data characteristics;

[0007] Step S2: Correlate the complementary geophysical field attribute characteristics with the spatial geometric characteristics of the slope body to achieve the initial fusion of the shallow layer of the survey data;

[0008] Step S3: Explore the latent spatial features of high-dimensional data and the latent surface features of low-dimensional data;

[0009] Step S4: Combining the data obtained in step S2, the deep neural network is replaced by the gradient boosting decision tree (GBD) to process the survey information and realize the deep feature fusion of the mountain slope survey data;

[0010] Step S5: Optimize data, improve data fusion efficiency and accuracy, and realize multi-level hierarchical intelligent fusion of mountain slope survey data.

[0011] Compared with the prior art, the present invention has the following advantages:

[0012] 1. Focus on the fine structure of the geophysical field of mountain slope hazard hazards and their differential characteristics, accurately and efficiently extract the geological information contained in survey data such as ground penetrating radar and laser electric cloud, provide data support for the establishment of refined detection equipment for mountain slope hazard hazards, and facilitate the construction of a wide-area slope stability and mountain slope hazard intelligent evaluation model.

[0013] 2. Breakthrough in the fusion and interpretation technology of mountain slope survey data under different time and space conditions, multi-level hierarchical fusion of different survey data, and realization of shallow fusion of low-dimensional homogeneous data, deep fusion of high-dimensional spatial data, and mining of potential spatial features and potential surface features of full-dimensional multi-source heterogeneous data.

[0014] 3. Establish intelligent analysis and fusion technology for mountain slope survey data to solve the problem of low generalization of traditional deep learning algorithms in processing multimodal data. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The figure is a flow chart of the method for standardization and intelligent fusion of mountain slope survey data of the present invention. DETAILED DESCRIPTION

[0016] The technical solution of the present invention is further described below with reference to the accompanying drawings, but is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention that does not depart from the spirit and scope of the technical solution of the present invention should be included in the scope of protection of the present invention.

[0017] The present invention provides a two-level fusion method for mountain slope survey data based on a gradient decision tree. By clustering and normalizing various types of mountain slope monitoring data, a unified standardized characterization method for survey data features at different time and space is constructed; shallow initial fusion of survey data is performed; deep feature fusion of mountain slope survey data is completed; and finally, data is optimized to improve data fusion efficiency and accuracy. Figure 1 As shown, the specific steps include:

[0018] Step S1: Process the mountain slope survey data, standardize the data according to the data structure characteristics, and construct a unified and standardized representation method for the survey data characteristics.

[0019] In this step, the survey data includes ground penetrating radar, laser point cloud and other survey data.

[0020] In this step, the data structure features include data source, format, resolution, coordinate system and other structural features.

[0021] In this step, data standardization includes data preprocessing, data threshold comparison, and data verification and standardization, among which:

[0022] Data preprocessing is to pre-set a unified standard program for different survey data that best reflects the characteristics of the mountain slope, classify them according to the data structure characteristics, establish a mapping model between different survey data and the standard program, and mark the mapping errors;

[0023] Data threshold comparison is mainly used to process data outside the mapping model. The threshold is set according to the resolution of the data. The threshold size is determined according to the actual processing object. The data within the threshold is included in the mapping model, and the bad data outside the threshold is eliminated.

[0024] Data verification and standardization are recorded based on the mapping results and mapping errors, and the standard program is further updated in combination with the threshold, and the data of the mapping model is further updated in combination with the mapping errors.

[0025] Step S2: Use the pan-sharpening intelligent fusion algorithm to associate the complementary geophysical field attribute characteristics with the spatial geometric characteristics of the slope body to achieve the initial fusion of the shallow layer of the survey data. The specific steps are as follows:

[0026] A fusion framework is established by combining neural networks, and the standardized processing details of the survey data are extracted according to the data format. The fusion factors and evaluation indicators are set, and the sharpening parameters are selected in combination with the resolution. The spatial and spectral information of the high-spatial-resolution panchromatic image and the low-spatial-resolution multispectral image obtained by the survey are fused to achieve a balance between spectral resolution and spatial resolution.

[0027] Step S3: Explore the latent spatial features of high-dimensional data and the latent surface features of low-dimensional data. The specific steps are as follows:

[0028] On the basis of full-color sharpening spectral imaging, based on the data spatial characteristics, coordinate characteristics, and depth data characteristics, we fully explore the high-latitude and low-latitude characteristics and hierarchical characteristics of the fused data, establish a discriminant structure to distinguish the acquired data from the target data, and train and compensate the data to make up for the problems of spatial information loss and spectral distortion of multi-source data.

[0029] Step S4: Combine the data obtained in the preliminary fusion step S2 to achieve deep feature fusion of the mountain slope survey data.

[0030] In this step, the deep feature fusion of survey data is based on spectral pan-sharpening fusion, and the gradient boosting decision tree (GBD) replaces the deep neural network to process survey information, realizing deep fusion of multimodal data.

[0031] In this step, the purpose of the gradient boosting decision tree is to achieve a deep fusion of remote sensing and spectral data. First, the feature vectors of the spectral and remote sensing images are extracted as the input vectors of the decision tree. Different types of data are smoothed in the decision tree, and a unified text vector encoding is specified. The unified vector regional features are extracted to reduce information interference such as additional noise. Secondly, the error value between the input vector and the specified center vector is adjusted to determine the decision weight, fusing the spectral and remote sensing data. Finally, high-confidence reference points are selected at the output of the decision tree to define the data output attributes. The output data is updated based on the confidence reference to improve the robustness of the decision tree.

[0032] Step S5: Optimize data, improve data fusion efficiency and accuracy, and realize multi-level hierarchical intelligent fusion of mountain slope survey data.

[0033] In this step, an optimization framework is designed with geographical, aerial, laser and other survey fusion information as input to achieve multi-level hierarchical intelligent fusion of mountain slope survey data. The specific fusion steps are as follows: first, a probability model is established, and the confidence distance is used to evaluate the fusion effect of existing data. Fusion rules are set, and the optimal fusion parameters are searched within the optimization framework to avoid too many framework training layers and slow efficiency. The fusion model is classified according to the training set and test set; then, an active selection strategy is constructed to eliminate unqualified samples in the probability estimation, and actively select and improve the fusion accuracy when the samples are limited to obtain the optimal benefit function.

[0034] Example:

[0035] A mountain slope survey obtained ground penetrating radar and laser point cloud data. The purpose is to achieve intelligent fusion of the above mountain slope survey data. The specific steps are as follows:

[0036] S1: According to the structural characteristics of GPR data and laser point cloud data, standard procedures are set according to the format, resolution, and coordinate system of the two data. A mapping model is established between different survey data and the standard procedure, and mapping errors are marked. Data outside the mapping model is eliminated, and the data in the mapping library is updated based on the mapping errors.

[0037] S2: Set the fusion factors and evaluation indicators of the geophysical field attribute characteristics and the spatial geometric characteristics of the slope body, and fuse the spatial and spectral information of the high spatial resolution panchromatic image and the low spatial resolution multispectral image obtained by the survey.

[0038] S3: Establish a discriminant structure to distinguish acquired data from target data, and train and compensate data to explore the latent spatial characteristics of high-dimensional data and the latent surface characteristics of low-dimensional data.

[0039] S4: Extract the feature vectors of laser point cloud spectral image and ground penetrating radar electromagnetic data as the input vector of the decision tree, select high-confidence benchmark points at the output of the decision tree to define the data output attributes, and update the output data based on the confidence benchmark.

[0040] S5: Use confidence distance to evaluate the fusion effect of existing data, search for the optimal fusion parameters within the optimization framework, eliminate unqualified samples in probability estimation, and actively select and improve fusion accuracy when samples are limited.

Claims

1. A two-level fusion method for mountain slope survey data based on gradient decision tree, characterized by The method comprises the following steps: Step S1: Processing the mountain slope survey data, standardizing the data according to the data structure characteristics, and constructing a unified and standardized representation method for survey data characteristics; Step S2: Correlate the complementary geophysical field attribute characteristics with the spatial geometric characteristics of the slope body to achieve the initial fusion of the shallow layer of the survey data. The specific steps are as follows: A fusion framework is established by combining neural networks. The standardized processing details of the survey data are extracted according to the data format. The fusion factors and evaluation indicators are set. The sharpening parameters are selected based on the resolution. The spatial and spectral information of the high-spatial-resolution panchromatic image obtained from the survey and the low-spatial-resolution multispectral image are fused to achieve a balance between spectral and spatial resolution. Step S3: Explore the latent spatial features of high-dimensional data and the latent surface features of low-dimensional data. The specific steps are as follows: On the basis of full-color sharpening spectral imaging, based on the data spatial characteristics, coordinate characteristics, and depth data characteristics, the high-latitude and low-latitude characteristics and hierarchical characteristics of the fused data are fully explored, a discriminant structure is established to distinguish the acquired data from the target data, and training and compensation data are used to make up for the spatial information loss and spectral distortion of multi-source data; Step S4: Combined with the data obtained in step S2, the deep neural network is replaced by the gradient boosting decision tree to process the survey information, and the deep feature fusion of the mountain slope survey data is realized. The specific steps are as follows: First, the feature vectors of spectral images and remote sensing images are extracted as the input vectors of the decision tree. Different types of data are smoothed in the decision tree, a unified text vector encoding is specified, and unified vector regional features are extracted to reduce the interference of additional noise information. Secondly, the error between the input vector and the specified center vector is adjusted to determine the decision weight, thus fusing the spectral and remote sensing data. Finally, a high-confidence benchmark is selected at the output of the decision tree to define the data output attributes, and the output data is updated based on the confidence benchmark to improve the robustness of the decision tree. Step S5: Optimize data, improve data fusion efficiency and accuracy, and realize multi-level hierarchical intelligent fusion of mountain slope survey data.

2. The two-level fusion method for mountain slope survey data based on gradient decision tree according to claim 1 is characterized in that In step S1, the survey data includes ground penetrating radar and laser point cloud survey data.

3. The two-level fusion method for mountain slope survey data based on gradient decision tree according to claim 1 is characterized in that In step S1, the data structure characteristics include data source, format, resolution, and coordinate system structure characteristics.

4. The two-level fusion method for mountain slope survey data based on gradient decision tree according to claim 1 is characterized in that In step S1, data standardization includes data preprocessing, data threshold comparison, and data verification and standardization, wherein: Data preprocessing is to pre-set a unified standard program for different survey data that best reflects the characteristics of the mountain slope, classify them according to the data structure characteristics, establish a mapping model between different survey data and the standard program, and mark the mapping errors; Data threshold comparison is mainly used to process data outside the mapping model. The threshold is set according to the resolution of the data. The threshold size is determined according to the actual processing object. The data within the threshold is included in the mapping model, and the bad data outside the threshold is eliminated. Data verification and standardization are recorded based on the mapping results and mapping errors, and the standard procedures are further updated in combination with the thresholds, and the data in the mapping model are further updated in combination with the mapping errors.

5. The two-level fusion method for mountain slope survey data based on gradient decision tree according to claim 1 is characterized in that In step S5, an optimization framework is designed with geographical, aerial, and laser survey fusion information as input to achieve multi-level hierarchical intelligent fusion of mountain slope survey data. The specific fusion steps are as follows: first, a probability model is established, and the confidence distance is used to evaluate the fusion effect of existing data. Fusion rules are set, and the optimal fusion parameters are searched within the optimization framework to avoid too many framework training layers and slow efficiency. The fusion model is classified according to the training set and the test set; then, an active selection strategy is constructed to eliminate unqualified samples in the probability estimation, and actively select and improve the fusion accuracy when the samples are limited to obtain the optimal benefit function.

Images