Vegetation detection of close-range images for landslide monitoring

Considering of the problems caused by vegetation which has serious effects on automatic landslide monitoring in close-range photogrammetry, this paper presents a vegetation detection algorithm for close-range images based on texture features and naive Bayes classifier. Some meaningful discussions and analysis have been done by carrying out a series of experiments especially focusing on problems such as the effectiveness of the algorithm, the effects of image contrast stretching on vegetation detection, the generality problem of samples training and so on. By comparing with another detection method which is based on visual cognition features, it proves the availability and the validity of this method. Last, by applying the result of vegetation detection to landslide monitoring, we can effectively eliminate the vegetation's interference with landslide deformation monitoring. The experimental results show that the vegetation detection algorithm presented in this paper can almost extract the vegetation regions from close-range images and the result is satisfying.