基于matlab深度学习工具箱来设计卷积神经网络用来对图像上的水体部分进行识别,并生成水体陆地二值化图像。采用的是9层卷积神经网络用来对图像进行特征提取和分类,水体识别的准确率可以达到96%以上。
Landslide susceptibility mapping by deep learning and slope unit
r.landslide is a free and open source add-on to the open source Geographic Information System (GIS) GRASS software for the mapping of landslide susceptiblity. The tool was developed in Python language and works on the basis of Artificial Neural Networks (ANNs) fed with environmental parameters and landslide databases. The module can be used by natural disaster management bodies and land use planning organs as a support tool for the elaboration of landslide susceptibility maps in an agile and efficient manner.
This is an one class classification problem, which is based on support vecter domain description (SVDD)and Sequential minimal optimization(SMO)algorithm. This work has been tested on pubilc data set, like Iris data set from http://archive.ics.uci.edu/ml/datasets/Iris.
One-Class SVM implementations for temporal segmentation of human activities
An Unmanned Aerial Vehicle based on DJI Phantom 4, incorporating Clarifai Computer Vision API which can detect natural crisis such as forest fires and debris flows in rural areas.
An Unmanned Aerial Vehicle based on DJI Phantom 4, incorporating Clarifai Computer Vision API which can detect natural crisis such as forest fires and debris flows in rural areas.
A web application that displays estimates for the probability and volume of debris flows that may be produced by a storm in a recently burned area
Thresh – Software for Tracking Rainfall Thresholds for Landslide and Debris Flow Occurrence (A collection of Fortran programs, Python scripts, shell scripts and Windows batch files along with sample data)
Implementing post-fire debris flow models with simulated fire data to enable pre-fire mitigation of debris flow hazards.
D-Claw extension of GeoClaw for solving two-phase model for landslides and debris flows.