Extraction Of Stellera Chamaejasme L. Based On HJ-HSI Hyperspectral Image Data

Stellera chamaejasme L. is one of the main poisonous species that distributing in alpine meadow in Qinghai Province. In recent years, the rapid spread of Stellera has serious harm to local animal husbandry and grassland ecosystem deterioration. Obtaining the distribution, proportion and variation information of Stellera, through the technology of Remote Sensing could provide an important basis for monitoring the Stellera’s distribution and assessing damage.It’s of great significance to maintain the sustainable development of animal husbandry in the region and grassland ecosystem balance.This paper used HJ-HSI hyper-spectral data and took Tibetan Autonomous Prefecture of Haibei in Qinghai province as the research area to launch the study of extracting the Stellera’s distribution through Remote Sensing. Hyper-spectral imaging can subtly detect Stellera’s spectral characteristics and efficiently overcomes the traditional remote sensing identification error caused by limited band. Starting from the perspective of spectrum waveform matching and based on the dimension reduction and denoising of HJ-HSI images, this study used two algorithms, Spectral Angle Mapper and Spectral Information Divergence, to identify the Stellera and acquires the following achievement:1. When the Stellera are in the full bloom stage, its unique phenological characteristics make its spectral reflectance and spectral reflectance of grass in the corresponding period a great difference. For specific performance, in the full bloom the spectral reflectance of Stellera in all bands are higher than the spectral reflectance of grass in the corresponding period and more than 70% at 900nm.For the community of Stellera, this apparent differences of spectral characteristics make the spectral reflectance of higher coverage of the community obviously different from grassland so that it provides the basis for the possibility of Stellera identification. Based on the analysis of spectral characteristic of the community, after Atmospheric Correction, the validity of the HJ-HSI spectrum of Stellera is inspected and it shows,after Atmospheric Correction, the HJ-HSI images could provides us with available spectral information and the community of Stellera and grass could be separated in the near-infrared band.2. Based on pros and cons analysis of different dimension reduction and denoising methods for the hyper-spectral remote sensing images, Principal components analysis and Minimum Noise Fraction Rotation are used to process the dimension reduction of HJ-HIS data ultimately. It turns out that the dimension reduction image data eliminate the correlation among the adjacent bands and separate the images and noise efficiently. Then, using the inverse Principal components analysis and inverse Minimum Noise Fraction Rotation not only makes the images retaining the spectrum waveform information but causes the data compression, the distinct decrease of data noise and high ratio of peak signal to noise. The purpose of the dimension reduction and denoising are achieved.3. Based on the theory of spectrum waveform matching, by building the Spectral Library of the Stellera and grass community and passing the dimension reduction and denoising images as the input data, SAM and SID are used respectively to extract and identify the Stellera. The results show that SAM of putting the image data after MNF transform as the data sources makes a highest precision of overall extraction, a 78.46% and Kappa coefficient is 0.5260. After counting the extraction proportion of the Stellera, it indicates that the Stellera damage proportion obtained by this method is same as the statistics of "The Qinghai province DuZaCao survey in 2012". It proves the SAM extraction technology based on the MNF transform can be applied in the identification of the Stellera.This research adopts the HJ-HIS data as data source to study the spectrum difference of the Stellera and grass community in full bloom and use two methods to extract and identify the community of the Stellera, and it turns out that the HJ-HIS data are potential and available in extracting and identifying the Stellera. Research results can provide remote sensing monitoring of the Stellera distribution and damage degree evaluation in western China with technical reference.