Degree Assessment And Dangerousness Evaluation Of Karst Rocky Desertification Ased On SVM

Karst Rocky Desertification is the extreme form of land degradation in karst regions, known as "the cancer of the Earth". Karst rocky desertification is one of the most serious ecological and economic problems in the southwestern karst regions of China, which has become the third horrible ecological issues following desertification of the Northwest Territories and water and soil erosion of the loess areas.The main evaluation models used by previous researchers to assess the degree of rocky desertification and dangerousness are the linear weighted comprehensive model, the regression analysis model and fuzzy comprehensive evaluation model, etc, which mainly consider the linear relationship and can't reflect the reasonable and objective relationships between different indexes and between the indexes and the degree of desertification and the dangerousness level, what's more, the result of level classification is easily affected by human factors owing to the subjectivity when researchers determine the value of the indexes of these models. As rocky desertification land is a type of land, rocky desertification land evaluation belongs to the scope of the land resource evaluation, a few scholars have used BP neural network in the evaluation studies of the degree of dangerousness of rocky desertification, and many researchers of domestic and overseas have successfully applied the neural network in the classification and evaluation of land quality, which illustrated the feasibility of the application of neural network intelligent algorithm to the land quality classification and assessment. However, the methods based on neural network involve the disadvantages of huge sample demand, slow training and over-fitting. In consideration of the superiority of support vector machine of solving the small number of samples problems and the prevention of over-fitting and the high cost of collecting samples owing to the complex terrain, deep ravines, poor accessibility of karst area, the support vector machine is introduced into the assessment of the degree of rocky desertification and dangerousness in this paper, thus solving the problems of rocky desertification and risk evaluation using fewer samples. As study of methodology, if this method is feasible, important economic and social value will be made.Support Vector Machine is developed on the basis of statistiacal learning theory and structural risk minimization theory, which owns a better evaluating effectiveness than the neural network model based on empirical risk minimization, and it has advantages of small sample learning and better robustness. In this paper, a karst rocky desertification and risk evaluation method based on SVM was proposed after analyzing the feasibility of SVM model for that evaluation, and accuracy comparative experiments under different numbers of samples were done on the assessment of the degree of rocky desertification and dangerousness of the standard SVM model, PSO-LSSVM model and the TSPSO-LSSVM model to provide the foundation of selecting a model which can take all the accuracy of evaluation, the number of samples and computational efficiency into account. At the same time, the automatic extraction methods of two key evaluation indicators of vegetation coverage and exposed bedrock rate were studied. Based on the TSPSO-LSSVM model, the rocky desertification degree and risk of the study area were evaluated, and the evaluation results of the field were then verification surveyed. The verification results showed that SVM model for rocky desertification and risk assessment is very feasible.The main research results of the paper are as follows:(1) Built the index system for the degree of rocky desertification and dangerousness assessment, and quantitative classified the indexes.(2) Established an automatic extraction method for the exposed rate of bedrock and vegetation coverage for the studied area.(3) Analyzed the feasibility of the support vector machine model for rock desertification and risk evaluation, and introduced support vector machine model into that assessment.Built the standard SVM model, PSO-LSSVM model, and TSPSO-LSSVM model for the degree of rocky desertification and dangerousness assessment.(4) Carried out accuracy comparative experiments of the standard SVM model, PSO-LSSVM model, and TSPSO-LSSVM model of different test samples numbers and discussed the applicable premise and condition of each model, and then selected a TSPSO-LSSVM model for the assessment of the rocky desertification and dangerousness degree.(5) Established the evaluation steps of the TSPSO-LSSVM model for the assessment of the degree of rocky desertification, and applied the selected TSPSO-LSSVM model for quantitative classification of karst rocky desertification for the studied area, and made a comprehensive partition and results evaluation, and verified the feasibility of the method.