Study On The Hyperspectral Image Preprocessing

People pay more attention to hyperspectral remote sensing, because it has more band numbers,more higher spectral resolution and more information. Compare with traditional multi-spectral remote sensing, hyperspectral remote sensing can acquire images in hundreds of registered, and its spectral resolution can reach up to 10nm, so that it can distinguish objects what former multi-spectral remote sensing can't, build the foundation to the development of calibration remote sensing.In case of so much spectral bands and such huge quantities of data, preprocessing becomes more and more important in the data processing and analysis, it affect quantitative analysis and data exploration directly. This paper discussed the hyperspectral preprocessing technique, and concentrated on hyperspectral image de-noising,spectral reflectance retrieval and geometric correction method. In general, the major works of this paper are as follows:1. Hyperspecyral image noise's origin and characteristic are systematically analyzed, and several common filter methods are introduced. Comparison and analysis are made by filter experiments with strip noise and spectral field random noise. First, aimed at strip noise's high frequency characteristic in PHI image, puts forward an improved de-striping method, it can eliminate image's strip noise perfectly. Then, apply wavelet threshold method to spectral filter experiment. Results proved that wavelet threshold filter method can preserve spectral characteristic and weak the random noise clearly at the same time.2. Hyperspectral image radiation preprocessing techniques are studied deeply. The origins of radiation errors and their effect to hyperspectral image are systematically analyzed. According to the characteristic of radiation errors, several radiation correction methods are discussed. Paid more attention to the study of hyperspectral image reflectance, and three hyperspectral image reflectance retrieval methods are expatiated, and compared and analyzed through the experiment.3. Hyperspectral image geometric preprocessing techniques are studied deeply. First, the imaging equations of frame perspective sensor and linear pushbroom sensor are introduced; then, several geometric correction algorithms about linear pushbroom imaging spectrometer are introduced and analyzed. Here this paper mainly studies on geometric correction method based on POS row orientation data, which is realized across programmed.4. A set of near-full experimental system based on the technologies and methods mentioned in above is established.