CN112529084B - Similar landslide recommendation method based on landslide section image classification model - Google Patents
Similar landslide recommendation method based on landslide section image classification model Download PDFInfo
- Publication number
- CN112529084B CN112529084B CN202011483680.9A CN202011483680A CN112529084B CN 112529084 B CN112529084 B CN 112529084B CN 202011483680 A CN202011483680 A CN 202011483680A CN 112529084 B CN112529084 B CN 112529084B
- Authority
- CN
- China
- Prior art keywords
- landslide
- meters
- thickness
- equal
- less
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a similar landslide recommendation method based on a landslide section image classification model, and belongs to the field of image classification and detection and landslide disaster prevention and control. The method aims to solve the problems of automatic classification of landslide section images and similar landslide recommendation in the prior art; the method has the advantages that the landslide section images are classified by adopting a random forest algorithm, similar landslide cases are recommended based on classification results, the classification precision of the landslide section images is high, and the similar landslide cases can be screened from a database and recommended to a user; according to the landslide disaster management method, landslide is described more finely according to the landslide section image, implicit information is mined, and a landslide case which meets requirements better is searched according to the fine description.
Description
Technical Field
The invention belongs to the field of image classification and detection and landslide disaster prevention and treatment, and particularly relates to a similar landslide recommendation method based on a landslide section image classification model.
Background
Sichuan is used as the center of southwest area, takes mountainous regions as main landforms, is one of serious disaster areas with frequent landslides, and accumulates a lot of valuable research data on landslide research and treatment. At present, in the aspect of landslide research, a natural history analysis method and an engineering geology similarity method are taken as main flow directions, but in the aspect of big data + landslide research, the domestic is still in a starting stage.
In a large amount of geological disaster field questionnaires and landslide engineering layout documents, a plurality of valuable landslide geological information is integrated into landslide section images, however, professional personnel are needed for manually reading the data, and the read information is possibly different based on different standards. At present, there are some researches on landslide section, but there is no relevant research result applied in landslide disaster prevention and control based on landslide section image classification.
Disclosure of Invention
The invention aims to provide a similar landslide recommendation method based on a landslide section image classification model, which is used for solving the problems of automatic classification and similar landslide recommendation of landslide section images in the prior art.
In order to achieve the purpose, the technical scheme of the invention is as follows: a similar landslide recommendation method based on a landslide section image classification model comprises the following steps:
step 1: taking out profile image data and expert investigation results from collected landslide geological disaster field questionnaires or landslide engineering layout documents;
step 2: preprocessing a profile image in a field questionnaire or landslide engineering layout document;
step 2.1: because the sectional image of the field questionnaire is not a vector image and can not be directly converted into useful information, the vector image of the landslide section needs to be drawn by one-to-one hand, and is consistent with the sectional image in the table as much as possible; the section image in the landslide engineering layout document is a vector image, and only extraction is needed without hand drawing;
step 2.2: on the landslide section image vector diagram, the following characteristics of different parts of the landslide section are calculated: the length of the landslide, the thickness of the front edge of the landslide, the thickness of the rear edge of the landslide and the thickness of the middle part of the landslide, whether the landslide has artificial excavation traces, the type of strong and weak gliding force and the steep and slow landslide are determined, and a landslide section vector diagram containing different landslide geological information is obtained;
step 2.3: converting the vector diagram into a grid diagram, converting into an ASCII (American Standard code for information interchange) text format, and taking the text as a sample;
and step 3: according to the features of different parts of the landslide section, re-marking all text sample data of the section image, obtaining a marking vector of each landslide section image as training data, wherein each value in the vector represents a marking attribute of one feature;
and 4, step 4: extracting characteristic representative samples from sample data for training;
step 4.1: programming and reading ASCII text of each sample in the training data, and storing the ASCII text into a two-dimensional array;
step 4.2: obtaining the total number of different characteristic values in each row/each column by a method of scanning the rows and columns of the two-dimensional array, obtaining a characteristic array with the length of row number plus column number to represent the sample, and carrying out the subsequent sample classification;
and 5: establishing a random forest model and training, wherein the random forest model is input as a landslide profile map, and output as labels of the length of a landslide, the thickness of the front edge of the landslide, the thickness of the rear edge of the landslide, the thickness of the middle part of the landslide, whether artificial excavation traces exist on the landslide, the type of the strength of the glide force and the type corresponding to the steepness and slowness of the landslide; training a random forest classifier by adopting the sample characteristic array extracted in the step 4 until the verification precision of the classifier basically meets the requirement;
step 6: classifying the landslide section images to be classified by adopting a trained random forest classifier;
and 7: according to the labels obtained by classification, the weights of different geological features are comprehensively considered, and the similarity of the landslide and the landslide case category in the database is calculated in a weighted mode; the calculation formula is as follows:
where X and Y are two samples, XiAnd yiCharacteristic values, alpha, of the i-th characteristic of the two samples, respectivelyiA weight representing the ith feature; and recommending a landslide cases with the highest similarity as reference according to the similarity calculation result.
Further, the marking method in step 3 is:
the marking method comprises the following steps:
according to the length of the landslide, dividing the landslide with the length less than 200 meters into short landslides and marking the short landslides as 0, dividing the landslide with the length more than or equal to 200 meters and less than 500 meters into short landslides and marking the short landslides as 1, dividing the landslide with the length more than or equal to 500 meters and less than 1500 meters into long landslides and marking the long landslides as 2, and dividing the landslides with the length more than or equal to 1500 meters into long landslides and marking the long landslides as 3;
according to the thickness of the front edge of the landslide, dividing the part with the thickness of less than 5 meters into thin parts and marking the thin parts with the thickness of more than or equal to 5 meters and less than 10 meters into 1, dividing the part with the thickness of more than or equal to 10 meters and less than 20 meters into thick parts and marking the thick parts with the thickness of more than or equal to 20 meters and marking the thick parts with the thickness of 3;
according to the thickness of the rear edge of the landslide, dividing the part with the thickness of less than 5 meters into thin parts and marking the thin parts as 0, dividing the part with the thickness of more than or equal to 5 meters and less than 10 meters into thin parts and marking the thin parts as 1, dividing the part with the thickness of more than or equal to 10 meters and less than 20 meters into thick parts and marking the thick parts as 2, and dividing the part with the thickness of more than or equal to 20 meters into thick parts and marking the thick parts as 3;
according to the thickness of the middle part of the landslide, the part with the thickness of less than 5 meters is divided into a thin part and is marked as 0, the part with the thickness of more than or equal to 5 meters and less than 10 meters is divided into a thin part and is marked as 1, the part with the thickness of more than or equal to 10 meters and less than 20 meters is divided into a thick part and is marked as 2, and the part with the thickness of more than or equal to 20 meters is divided into a thick part and is marked as 3;
according to the fact that whether the landslide has the artificial excavation trace or not, dividing the slope of the slope toe with the angle smaller than 45 degrees into the artificial excavation trace-free trace and marking the artificial excavation trace as 0, dividing the slope toe with the angle larger than or equal to 45 degrees and smaller than 90 degrees into the suspected artificial excavation trace and marking the suspected artificial excavation trace as 1, and dividing the slope toe with the angle larger than or equal to 90 degrees into the artificial excavation trace and marking the artificial excavation trace as 2;
according to the type of the strength of the gliding force, the integral slope of the landslide is less than 10 degrees, the thickness of the front edge, the middle part and the rear edge of the landslide is less than 5 meters, the gliding force is weak and is marked as 0, the integral slope of the landslide is more than or equal to 10 degrees and less than 45 degrees, the thickness of any part of the front edge, the middle part and the rear edge of the landslide is more than or equal to 5 meters and less than 20 meters, the gliding force is weak and is marked as 1, the integral slope of the landslide is more than or equal to 45 degrees, the thickness of any part of the middle part and the rear edge of the landslide is more than or equal to 5 meters and less than 20 meters, the gliding force is strong and is marked as 2, and the integral slope of the landslide is more than or equal to 45 degrees, the thickness of the rear edge of the landslide is more than or equal to 20 meters, the gliding force is strong and is marked as 3;
according to the steepness of the landslide, a score with a gradient of less than 10 degrees is flat and marked as 0, a score with a gradient of 10 degrees or more and less than 20 degrees is gentle and marked as 1, a score with a gradient of 20 degrees or more and less than 45 degrees is steep and marked as 2, and a score with a gradient of more than 45 degrees is steep and marked as 3.
Further, in step 7, the weight is [ length of the landslide, thickness of the front edge of the landslide, thickness of the rear edge of the landslide, thickness of the middle part of the landslide, presence or absence of artificial excavation traces of the landslide, type of strong or weak glide force, and steepness and slowness of the landslide ] (0.10, 0.16,0.20,0.18,0.21,0.07, 0.08) ].
According to the method, the landslide section images are classified by adopting a random forest algorithm, similar landslide cases are recommended based on the classification result, the method not only has higher classification precision of the landslide section images, but also can screen the similar landslide cases from a database and recommend the similar landslide cases to a user; according to the landslide disaster management method, landslide is described more finely according to the landslide section image, implicit information is mined, and a landslide case which meets requirements better is searched according to the fine description.
Drawings
FIG. 1 is a flow chart of an embodiment.
Fig. 2 is a landslide section image example of the present invention.
FIG. 3 is an example of a similar landslide cubic recommendation of the present invention; in the three recommendation results in fig. 3, the first row of each recommendation result is the input landslide test sample to be recommended, and the second row is the output 5 similar landslide cases recommended, and the similarity decreases from left to right.
Detailed Description
As shown in fig. 1, a similar landslide recommendation method based on a landslide section image classification model includes the following steps:
s1, preprocessing the collected landslide section images in the field questionnaire or landslide engineering layout document by utilizing ArcGIS to obtain ASCII text type data, reading an image array in the text data and storing the image array in a two-dimensional array to serve as a training data set;
s2, on the basis of the step S1, the data sets are classified into multiple categories according to 7 different landslide geological features, and labels are printed to be used as supervised learning samples;
s3, adopting a row-column scanning method for the image two-dimensional array to obtain a characteristic array with the length of < row number + column number > as sample data;
s4, a random forest classification model is constructed by using a random forest algorithm, 80% of sample data are imported and learned, 20% of the sample data are used as a verification set, and 80% of the sample data are used as a training set. Adjusting parameters of the classification model by a grid search method until the obtained verification precision meets the requirement;
and S5, analyzing the trained model by using the residual 20% of sample data as test samples and evaluating the accuracy.
And S6, classifying the landslide section images to be classified by using the trained random forest classifier.
And S7, according to the labels obtained by classification, comprehensively considering the weights of different geological features, and calculating the similarity of the landslide and the landslide case categories in the database in a weighting manner. Then 5 landslide cases with higher similarity are recommended as reference.
The process of 7 steps is detailed in the following simulation experiment.
Step 1: preprocessing data;
and selecting a landslide section image and part of expert investigation results from the collected field questionnaire or landslide engineering layout document. If the landslide section image of the field questionnaire is available, performing one-to-one fine hand drawing by using ArcGIS to generate a vector diagram; if the landslide section image is in the landslide engineering layout document, the landslide section image is directly extracted in a vector image format. The workload of the step is large, and the requirement on the drawing precision is high. After the vector image of the designated landslide section is extracted from ArcGIS, the vector image is converted into a grid image, resampling is included in the step, then the grid image is converted into an ASCII text format, finally, the text of each designated landslide section image is programmed and read, and the image two-dimensional array of the landslide section image is obtained and serves as a training data set.
Step 2: labeling the image according to landslide geological features;
according to landslide geological features including the length of a landslide, the thickness of the front edge of the landslide, the thickness of the rear edge of the landslide and the thickness of the middle part of the landslide, whether artificial excavation traces exist on the landslide, the type of strong and weak sliding force and the steepness and slowness of the landslide are 7 features, a supervised learning method is adopted to manually classify a training data set, seven different classification tasks respectively correspond to different classification labels, the classification tasks are divided into two classification labels of long landslide and short landslide by taking the classification of the length of the landslide as an example, the classification labels of each landslide section image and corresponding image names are bound to be recorded, and the classification corresponding to the data can be obtained simultaneously when the training data need to be called.
The label marking method comprises the following steps:
according to the length of the landslide, dividing the landslide with the length less than 200 meters into short landslides and marking the short landslides as 0, dividing the landslide with the length more than or equal to 200 meters and less than 500 meters into short landslides and marking the short landslides as 1, dividing the landslide with the length more than or equal to 500 meters and less than 1500 meters into long landslides and marking the long landslides as 2, and dividing the landslides with the length more than or equal to 1500 meters into long landslides and marking the long landslides as 3;
according to the thickness of the front edge of the landslide, dividing the part with the thickness of less than 5 meters into thin parts and marking the thin parts with the thickness of more than or equal to 5 meters and less than 10 meters into 1, dividing the part with the thickness of more than or equal to 10 meters and less than 20 meters into thick parts and marking the thick parts with the thickness of more than or equal to 20 meters and marking the thick parts with the thickness of 3;
according to the thickness of the rear edge of the landslide, dividing the part with the thickness of less than 5 meters into thin parts and marking the thin parts as 0, dividing the part with the thickness of more than or equal to 5 meters and less than 10 meters into thin parts and marking the thin parts as 1, dividing the part with the thickness of more than or equal to 10 meters and less than 20 meters into thick parts and marking the thick parts as 2, and dividing the part with the thickness of more than or equal to 20 meters into thick parts and marking the thick parts as 3;
according to the thickness of the middle part of the landslide, the part with the thickness of less than 5 meters is divided into a thin part and is marked as 0, the part with the thickness of more than or equal to 5 meters and less than 10 meters is divided into a thin part and is marked as 1, the part with the thickness of more than or equal to 10 meters and less than 20 meters is divided into a thick part and is marked as 2, and the part with the thickness of more than or equal to 20 meters is divided into a thick part and is marked as 3;
according to the existence of artificial excavation traces on the landslide, dividing the slope of the slope toe with the slope angle less than 45 degrees into the artificial excavation traces without the slope toe and marking the artificial excavation traces as 0, dividing the slope toe with the slope angle more than or equal to 45 degrees and less than 90 degrees into suspected artificial excavation and marking the suspected artificial excavation traces as 1, and dividing the slope toe with the slope angle more than or equal to 90 degrees into the artificial excavation traces with the slope toe with the slope angle more than or equal to 2;
according to the type of the downward sliding force, dividing the integral slope of the landslide into the downward sliding force weak part and the mark 0, wherein the slope of the integral slope of the landslide is less than 10 degrees, the thickness of the front edge, the middle part and the rear edge of the landslide is less than 5 meters, the slope of the integral slope of the landslide is more than or equal to 10 degrees and less than 45 degrees, the thickness of any part of the front edge, the middle part and the rear edge of the landslide is more than or equal to 5 meters and less than 20 meters, the slope of the integral slope of the landslide is stronger and the mark 2, and the slope of the integral slope of the landslide is more than or equal to 45 degrees, the thickness of the rear edge of the landslide is more than or equal to 20 meters, the slope of the integral slope of the landslide is stronger and the mark 3;
according to the steepness of the landslide, a score with a gradient of less than 10 degrees is flat and marked as 0, a score with a gradient of 10 degrees or more and less than 20 degrees is gentle and marked as 1, a score with a gradient of 20 degrees or more and less than 45 degrees is steep and marked as 2, and a score with a gradient of more than 45 degrees is steep and marked as 3.
And step 3: extracting characteristics and training samples;
and after classification is finished, acquiring a feature array of the landslide section image according to a method of a two-dimensional array of a row-column scanning image, wherein the feature array represents the image features and serves as a training sample.
The specific operation is as follows: through the previous steps, each landslide section image already acquires a corresponding image two-dimensional array and a corresponding label category in each classification task. Taking landslide length classification as an example, if the number 0 is a short landslide and the number 1 is a long landslide, in the landslide length classification task, the label of a certain landslide section image corresponds to 0 or 1, the training data is an image feature array generated by adopting a row-column scanning method, and the size of the image is not very large because the steps such as resampling and the like are already carried out in the preprocessing stage, so that the array size is not compressed by adopting methods such as PCA and the like. In each classification task, the label category of each landslide section image needs to be stored in a label array at the same time, the characteristic array corresponding to the landslide image is stored in a training sample array, and-9999 representing a blank space in the array is replaced by 0.
And 4, step 4: establishing a random forest classification model;
in order to improve the classification accuracy, 7 classification tasks can be divided into a plurality of classifiers, and currently, the 7 classification tasks are divided into 4 classifiers for training. Establishing a random forest classification model, inputting 80% of sample sets of image characteristics (the rest 20% of sample sets are used for later model tests and are not input into the training model) into the training model, splitting the data set input into the training model, wherein 80% of the data set is used as the training set, and 20% of the data set is used as a verification set, and training the classification model.
The specific operation is as follows: and loading the training data and the corresponding label data to a random forest classification model, and adjusting the parameters of the classification model. And adjusting model parameters by adopting a grid search method, wherein n _ estimators is the maximum iteration number of the learner, max _ depth is the maximum depth of the decision tree, and max _ features is the maximum feature number, the three parameters are the most important parameters of the whole classifier, and other parameters also play a great role. And continuously adjusting the parameters until the verification precision reaches a satisfactory result.
And 5: analyzing the model and evaluating the accuracy;
and (3) comparing the y _ predicted (predicted value label obtained by model learning of the x _ test set through S4) obtained by importing the rest 20% of samples into the model as the test set with the actual y _ test (actual accuracy label of the test set) to obtain corresponding accuracy, and evaluating the model according to the accuracy.
Step 6: classifying the given landslide section image by adopting a trained classifier;
inputting a landslide section image, and classifying by a plurality of classifiers to obtain 7 label results which respectively represent category information of different geological features.
And 7: and according to the labels obtained by classification, comprehensively considering the weights of different geological features, and weighting and calculating the similarity of the landslide and the landslide case category in the database. Then 5 landslide cases with higher similarity are recommended as reference
The specific operation is as follows: according to the geological feature category to which the landslide belongs, obtained through classification in the last step, different weight arrays are set according to expert experience and data analysis results, the weight arrays can reflect the importance degree of different geological features under different scenes through parameter adjustment, and the method has a good application value. Then, a similarity calculation method is adopted to calculate the similarity between the landslide geological feature information embodied by the landslide profile image input by the user and the landslide cases in the database, a weighted Euclidean distance algorithm is adopted to calculate the similarity, a weight array is used, and a calculation formula is as follows:
where X and Y are two samples, XiAnd yiCharacteristic values, alpha, of the i-th characteristic of the two samples, respectivelyiThe weight representing the ith characteristic is obtained by analyzing the domain knowledge and factors, and the weight is corresponding to the length of the landslide, the thickness of the front edge of the landslide, the thickness of the rear edge of the landslide, the thickness of the middle part of the landslide, the existence of artificial excavation traces on the landslide, the type of the strong or weak glide force, the steepness and slowness of the landslide]=[0.10,0.16,0.20,0.18,0.21,0.07,0.08]。
After the similarity between the category of the landslide section image input by the user and the landslide cases in the database is obtained through calculation, the landslide section images of the first 5 landslide cases with higher similarity are recommended, and other specific treatment measures can be provided as references according to needs.
Claims (2)
1. A similar landslide recommendation method based on a landslide section image classification model comprises the following steps:
step 1: taking out profile image data and expert investigation results from collected landslide geological disaster field questionnaires or landslide engineering layout documents;
step 2: preprocessing a profile image in a field questionnaire or landslide engineering layout document;
step 2.1: because the sectional image of the field questionnaire is not a vector image and can not be directly converted into useful information, the vector image of the landslide section needs to be drawn by one-to-one hand, and is consistent with the sectional image in the table as much as possible; the section image in the landslide engineering layout document is a vector image, and only extraction is needed without hand drawing;
step 2.2: on the landslide section image vector diagram, the following characteristics of different parts of the landslide section are calculated: the length of the landslide, the thickness of the front edge of the landslide, the thickness of the rear edge of the landslide and the thickness of the middle part of the landslide, wherein the landslide has or does not have artificial excavation traces, and the type of the strong and weak gliding force and the steepness and slowness of the landslide are used for obtaining a landslide section vector diagram containing different landslide geological information;
step 2.3: converting the vector diagram into a grid diagram, converting into an ASCII text format, and taking the text as a sample;
and step 3: according to the features of different parts of the landslide section, re-marking all text sample data of the section image, obtaining a marking vector of each landslide section image as training data, wherein each value in the vector represents a marking attribute of one feature;
the marking method comprises the following steps:
according to the length of the landslide, dividing the landslide with the length less than 200 meters into short landslides and marking the short landslides as 0, dividing the landslide with the length more than or equal to 200 meters and less than 500 meters into short landslides and marking the short landslides as 1, dividing the landslide with the length more than or equal to 500 meters and less than 1500 meters into long landslides and marking the long landslides as 2, and dividing the landslides with the length more than or equal to 1500 meters into long landslides and marking the long landslides as 3;
according to the thickness of the front edge of the landslide, dividing the part with the thickness of less than 5 meters into thin parts and marking the thin parts with the thickness of more than or equal to 5 meters and less than 10 meters into 1, dividing the part with the thickness of more than or equal to 10 meters and less than 20 meters into thick parts and marking the thick parts with the thickness of more than or equal to 20 meters and marking the thick parts with the thickness of 3;
according to the thickness of the rear edge of the landslide, dividing the part with the thickness of less than 5 meters into thin parts and marking the thin parts as 0, dividing the part with the thickness of more than or equal to 5 meters and less than 10 meters into thin parts and marking the thin parts as 1, dividing the part with the thickness of more than or equal to 10 meters and less than 20 meters into thick parts and marking the thick parts as 2, and dividing the part with the thickness of more than or equal to 20 meters into thick parts and marking the thick parts as 3;
according to the thickness of the middle part of the landslide, the part with the thickness of less than 5 meters is divided into a thin part and is marked as 0, the part with the thickness of more than or equal to 5 meters and less than 10 meters is divided into a thin part and is marked as 1, the part with the thickness of more than or equal to 10 meters and less than 20 meters is divided into a thick part and is marked as 2, and the part with the thickness of more than or equal to 20 meters is divided into a thick part and is marked as 3;
according to the existence of artificial excavation traces on the landslide, dividing the slope of the slope toe with the slope angle less than 45 degrees into the artificial excavation traces without the slope toe and marking the artificial excavation traces as 0, dividing the slope toe with the slope angle more than or equal to 45 degrees and less than 90 degrees into suspected artificial excavation and marking the suspected artificial excavation traces as 1, and dividing the slope toe with the slope angle more than or equal to 90 degrees into the artificial excavation traces with the slope toe with the slope angle more than or equal to 2;
according to the type of the downward sliding force, dividing the integral slope of the landslide into the downward sliding force weak part and the mark 0, wherein the slope of the integral slope of the landslide is less than 10 degrees, the thickness of the front edge, the middle part and the rear edge of the landslide is less than 5 meters, the slope of the integral slope of the landslide is more than or equal to 10 degrees and less than 45 degrees, the thickness of any part of the front edge, the middle part and the rear edge of the landslide is more than or equal to 5 meters and less than 20 meters, the slope of the integral slope of the landslide is stronger and the mark 2, and the slope of the integral slope of the landslide is more than or equal to 45 degrees, the thickness of the rear edge of the landslide is more than or equal to 20 meters, the slope of the integral slope of the landslide is stronger and the mark 3;
according to the steepness of the landslide, the part with the gradient less than 10 degrees is divided into gentle parts and marked as 0, the part with the gradient more than or equal to 10 degrees and less than 20 degrees is divided into gentle parts and marked as 1, the part with the gradient more than or equal to 20 degrees and less than 45 degrees is divided into steep parts and marked as 2, and the part with the gradient more than 45 degrees is divided into steep parts and marked as 3;
and 4, step 4: extracting characteristic representative samples from sample data for training;
step 4.1: programming and reading ASCII text of each sample in the training data, and storing the ASCII text into a two-dimensional array;
step 4.2: obtaining the total number of different characteristic values in each row/each column by a method of scanning the rows and columns of the two-dimensional array, obtaining a characteristic array with the length of row number plus column number to represent the sample, and carrying out the subsequent sample classification;
and 5: establishing a random forest model and training, wherein the random forest model is input as a landslide section map, and output as the length of a landslide, the thickness of the front edge of the landslide, the thickness of the rear edge of the landslide, the thickness of the middle part of the landslide, whether artificial excavation traces exist on the landslide, and the type of the glide force and the type of the corresponding grade of the steepness and slowness of the landslide; training a random forest classifier by adopting the sample characteristic array extracted in the step 4 until the verification precision of the classifier basically meets the requirement;
step 6: classifying the landslide section images to be classified by adopting a trained random forest classifier;
and 7: according to the labels obtained by classification, the weights of different geological features are comprehensively considered, and the similarity of the landslide and the landslide case category in the database is calculated in a weighted mode; the calculation formula is as follows:
where X and Y are two samples, XiAnd yiCharacteristic values, alpha, of the i-th characteristic of the two samples, respectivelyiA weight representing the ith feature; and recommending a landslide cases with the highest similarity as reference according to the similarity calculation result.
2. The similar landslide recommendation method based on landslide section image classification model according to claim 1, wherein in step 7, the weights are corresponding to:
the length of the landslide, the thickness of the front edge of the landslide, the thickness of the rear edge of the landslide, the thickness of the middle part of the landslide, the existence of artificial excavation marks on the landslide, the type of strong and weak gliding force and the steepness and slowness of the landslide are equal to [0.10,0.16,0.20,0.18,0.21,0.07 and 0.08 ].
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011483680.9A CN112529084B (en) | 2020-12-16 | 2020-12-16 | Similar landslide recommendation method based on landslide section image classification model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011483680.9A CN112529084B (en) | 2020-12-16 | 2020-12-16 | Similar landslide recommendation method based on landslide section image classification model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112529084A CN112529084A (en) | 2021-03-19 |
CN112529084B true CN112529084B (en) | 2022-05-03 |
Family
ID=75000554
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011483680.9A Active CN112529084B (en) | 2020-12-16 | 2020-12-16 | Similar landslide recommendation method based on landslide section image classification model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112529084B (en) |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101363730A (en) * | 2008-08-07 | 2009-02-11 | 北京林业大学 | Investigation and sortation method for disaster danger area in mountanious region |
KR20120084560A (en) * | 2011-01-20 | 2012-07-30 | 삼성전자주식회사 | Method and apparatus for providing and receiving disaster information |
CN102819023B (en) * | 2012-07-27 | 2014-09-17 | 中国地质大学(武汉) | Method and system of landslide recognition of complicated geological background area based on LiDAR |
US20140245210A1 (en) * | 2013-02-28 | 2014-08-28 | Donan Engineering Co., Inc. | Systems and Methods for Collecting and Representing Attributes Related to Damage in a Geographic Area |
US20150019262A1 (en) * | 2013-07-11 | 2015-01-15 | Corelogic Solutions, Llc | Method and system for generating a flash flood risk score |
CN104200100B (en) * | 2014-09-01 | 2017-02-15 | 重庆大学 | Three-dimensional slope stability prediction method based on sliding displacement analysis |
CN109241902B (en) * | 2018-08-30 | 2022-05-10 | 北京航空航天大学 | Mountain landslide detection method based on multi-scale feature fusion |
CN110532872A (en) * | 2019-07-24 | 2019-12-03 | 宁德市公路局 | A kind of landslide hierarchy system and method based on convolution supporting vector neural network |
CN110929939B (en) * | 2019-11-26 | 2023-04-07 | 电子科技大学 | Landslide hazard susceptibility spatial prediction method based on clustering-information coupling model |
-
2020
- 2020-12-16 CN CN202011483680.9A patent/CN112529084B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN112529084A (en) | 2021-03-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103632168B (en) | Classifier integration method for machine learning | |
CN107291688A (en) | Judgement document's similarity analysis method based on topic model | |
CN103996057A (en) | Real-time handwritten digital recognition method based on multi-feature fusion | |
CN113344050A (en) | Lithology intelligent identification method and system based on deep learning | |
CN108830312A (en) | A kind of integrated learning approach adaptively expanded based on sample | |
CN110310012B (en) | Data analysis method, device, equipment and computer readable storage medium | |
CN110993102A (en) | Campus big data-based student behavior and psychological detection result accurate analysis method and system | |
CN110929746A (en) | Electronic file title positioning, extracting and classifying method based on deep neural network | |
CN112215696A (en) | Personal credit evaluation and interpretation method, device, equipment and storage medium based on time sequence attribution analysis | |
CN106934055B (en) | Semi-supervised webpage automatic classification method based on insufficient modal information | |
CN105786898B (en) | A kind of construction method and device of domain body | |
CN109523514A (en) | To the batch imaging quality assessment method of Inverse Synthetic Aperture Radar ISAR | |
De Stefano et al. | Layout measures for writer identification in mediaeval documents | |
CN116629258B (en) | Structured analysis method and system for judicial document based on complex information item data | |
CN112529084B (en) | Similar landslide recommendation method based on landslide section image classification model | |
CN117112648A (en) | College student learning portrait generation system and method based on intelligent education data | |
Zhang et al. | Computational method for calligraphic style representation and classification | |
CN116434273A (en) | Multi-label prediction method and system based on single positive label | |
Depositario et al. | Automated Categorization of Research Papers with MONO Supervised Term Weighting in RECApp | |
CN115205877A (en) | Irregular typesetting invoice document layout prediction method and device and storage medium | |
Faigenbaum-Golovin et al. | Writer characterization and identification of short modern and historical documents: reconsidering paleographic tables | |
CN101710392B (en) | Important information acquiring method based on variable boundary support vector machine | |
CN104778479A (en) | Image classification method and system based on sparse coding extractor | |
Sridevi et al. | Knowledge representation and answer evaluation system using language processing algorithm | |
JP7291347B2 (en) | Drawing retrieval device, model generation device, drawing retrieval method, and model generation method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |