Research On The Information Extraction Of Alteration Minerals By Using CASI/SASI Hyperspectral Data Based On Measured Spectra

Hyperspectral Remote Sensing refers to the remote sensing image of high spectral resolution, which has become an important emerging technology in the field of geological and mineral explorations. To use hyperspectral remote sensing for extracting the mineralization information is one of the hot topics on the researches of remote sensing geology. A variety of theories for information extraction of alteration minerals using hyperspectral data have been proposed and databases of mineral spectra have been established by domestic and foreign scholars. However, because of the effects of environmental factors and measurement restrictions in the field, various variations of the mineral spectral characteristics occurred, while some drawbacks and uncertainty in the existing mineral extraction methods. Leading to the incorrect results in mineral identification and mineral mapping. It is particularly important to select proper mineral spectra database and improve the existing method for the mineral information extraction.The research take the uranium mineralization area of Xuemisitan in Xinjiang as the study area. According to full analysis of the measured spectral absorption characteristics of different altered minerals and the conventional mineral information extraction, proper method for the hyperspectral mineral alteration information extraction was put forward for the study area by using the CASI/SASI data and field measured spectral data. The main contents and conclusions are as follows:1. A systematic evaluation on the quality of the CASI/SASI data was conducted respectively from the aspect of information quantity, Signal to Noise Ratio(SNR) of image, data dimension reduction and noise reduction. Besides, the evaluation results were compared with commonly used Hyperion hyperspectral data. The evaluation results indicates that CASI and SASI are both sensors with excellent imaging qualities. The stability of CASI data is the best. The SASI data contains the maximum information. Besides, CASI/SASI data possesses the higher SNR, especially in the near infrared range, SASI possesses the maximum SNR value. The data acquired by the two sensors contains a lot of random noise, but the effective signal and noise data can be well separated. The data after data dimension reduction and noise reduction can be applied to the identification and extraction of mineral information.2. The pre-treatment process of CASI/SASI data and filed measured spectral data was studied. In order to eliminate the influence of outlier spectra in mineral mapping, we put forward a method of combining the overall measurements and local measurements to eliminate outlier spectra. It laid the foundation for mineral identification and mapping.3. The mineral spectral characteristics and impact factors of the mining area were analyzed. Results showed that without changes in the internal structure and composition of minerals, the position of the absorption was stable. The absorption position of the diagnostic spectral characteristics in the mining area was determined. Compared with the USGS spectra, it indicates that the spectra in the mining area has undergone some alteration. Moreover, the reliability of mineral extraction by using measured spectra was proved in the spectral characteristics.4. Spectral dimension reduction is a necessary step in the hyperspectral mineral identification and mapping. However, the band selection in the conventional dimension methods is based on some basic calculation of statistical information, which does not consider the image spatial structure and the unique features of the objects. In this research, a dimension reduction method of hyperspectral minerals extraction based on the measured spectral characteristics and image spatial information is proposed, and the diagnostic spectral characteristics and high resolution image spatial information which are important for the identification were taken into account. The experimental results show that the clustering degree for each alteration mineral of this method is better, which verifies the reliability and superiority of this dimension reduction method.5. The mineral mapping result by using measured spectra and USGS spectra respectively for comparative analysis, shows that the mineral identification and mapping accuracy is higher by using measured spectra. In view of the uncertainty of the mapping results by the conventional spectral angle mapping method in different threshold, this paper presents an extraction method of hyperspectral mineral alteration information based on multi-source information fusion. In this method, the decision tree is established for the mineral information extraction by combing the image information and the positions of diagnostic spectral features, which has been proved to have a good reliability.