基于贝叶斯优化的卷积神经网络结合机器学习分类器的滑坡易发性预测研究
Landslide Susceptibility Prediction Based on Bayesian Optimization Convolutional Neural Network Combined with Machine Learning Classifier

现有使用机器学习进行滑坡监测方法存在滑坡易发性特征选择超参数多为连续变量，数值约束松散的问题。为解决此问题，提出一种结合CNN-LR-SVM的滑坡易发性监测方法。使用CNN提取滑坡易发性特征值，捕获输入数据中的上下文信息，消除繁琐的传统特征选择过程，提升特征选择准确率。将提取到的特征值与LR、SVM分类器结合。之后，使用贝叶斯优化方法寻找CNN、LR和SVM超参数，考虑历史信息，优化下一步超参数设置方法，解决超参数优化数值约束问题。经实验验证，本方法在选择不同优化算法与激活函数时相比传统单一机器学习模型(以CNN与SVM为例)均有良好性能表现。
The existing landslide monitoring methods using machine learning have the problem that most of` the hyperparameters selected for landslide susceptibility characteristics are continuous variables and the numerical constraints are loose. To solve this problem, a landslide susceptibility monitoring method combined with CNN-LR-SVM is proposed. Using CNN to extract landslide prone feature val-ues, capture context information in input data, eliminate complicated traditional feature selection processes, and improve feature selection accuracy. The extracted eigenvalues are combined with LR and SVM classifiers. Then, using the Bayesian optimization method to find CNN, LR and SVM hyperparameters, considering the historical information, the next step of the hyperparameter setting method is optimized to solve the numerical constraint problem of hyperparameter optimization. The experimental results show that the proposed method has good performance compared with the traditional single machine learning model (take CNN and SVM as examples) when selecting different optimization algorithms and activation functions.