Data Mining For Forest Space Information Features From Remote Sensing Images

Global environment is becoming worse and worse. Natural disasters are frequent, which seriously impact on people's life, production, economic development and social stability, such as drought,sandstorm,flood,etc. In order to alleviate these natural disasters, intensive protection and research for forest have become extremely important. The forest coverage is fargoing, traditional method of point observation is difficult to comprehensively research the forest. Remote sensing technology offered an effective analytical way for the forest's spatial information. In recent years, data mining won great attention by the informational industry, we urgently need to convert abundant data into useful information and knowledge. Remote sensing image contains lots of information so that it becomes the most potential field in data mining.Date mining based on study area's remote sensing multi-band images. Remote sensing gray image contains lots of information, but its level is not rich that the observation and analysis are difficult satellite multi-channel information can not be completely efficient used, based on this poor availability and the phenomenon of different spectra for the same object and the same spectrum for different objects that greatly affect the classification accuracy, bring up a method to distinguish and classify surface features. This method is based on wavelet spectra characteristics,wavelet textural features and BP neural network. To analyze the forest's change by ratio vegetation index and normalized vegetation index's change. To analyze vegetation's coverage and different type of surface's influence on Surface temperature; Combining with wavelet analysis theory and artificial neural network to distinguish between Spectral features similar trees and corps.Research results show that wavelet bp neural network can effectively identify spectrum similar trees and crops. With wavelet and textural features's BP neural network to distinguish and classify surface features can obviously improve the accuracy to identify farmlands, shady slopes and sunny slopes, as well as soil, rock and water, which show serious confusable spectrum phenomenon. The identification accuracy reaches 89.1% for sunny slopes,88.6% for shady slopes,87.8% for farmlands,98.0% for soil and water. Total error is 6.4%. The algorithm is superior to maximum likelihood and neural network algorithms that just rely on spectral information. Classification rules will be more targeted and reasonable, as well as identificatic accuracy will be improved, with distinguishing shady slopes and sunny slopes on the comph hilly topography in remote sensing images. Remote sensing vegetation index and vegetatic coverage provides an effective way to quantitative analyze the forest's coverage change aboi upriver of Li-Jiang.The research innovations include:bring up a new method to distinguish and classify surfa( features, this method combined with wavelet analysis theory and neural network theor selecting the best band combination from remote sensing multi-band image,selecting optimu wavelet basis,extracting wavelet spectral signature and textural features,Using BP neur network to distinguish and classify surface features.This method can improve remote sensir gray image's use ratio and reduce the classification cost and computational complexity, with higher accuracy in recognition and classification.