Research On Key Technologies Of Hyperspectral Image Preprocessing And Its Application

Hyperspectral image is widely used in precision agriculture,object survey,environmental detection,urban remote sensing,etc,for its "space spectrum integration" quality.While the precondition for those applications is the development of image processing technology.Data preprocessing and data analysis are the two parts of hyperspectral image processing which including image segmentation,normalization,filtering,feature extraction,classification and recognition.In this paper,some optimization algorithm of image segmentation,image normalization and image filtering algorithms is proposed,a wheat scab classification model of more than 98% accuracy is also developed.The main work and contribution of this paper are as follows:1)To solve the problem about separating wheat and background,Continuous Projection Algorithm is proposed to extract 20 bands which is the feature band of hyperspectral image,the spectral angle of the feature bands are calculated,then the hyperspectral image is segmented accurately by the spectral angle of the feature bands.To further improve accuracy,morphological operator,etched and opened,is used.By this way,the segmentation accuracy of wheat and background is almost 100%.2)To solve the problem that the large mean square error between calculated reflectance spectrum and calibration spectrum,A Dual Reflectance Plate Reflectance Inversion method is proposed.The image acquisition method is reasonably designed and the noise evaluation system of hyperspectral image is established according to the noise distribution characteristic of the swing sweep hyperspectral image acquisition equipment.The reflectance spectrum which mean square error is less than 0.0001 is computed by the noise value of the evaluation system and the DN value of the dual reflectance plate.3)To solve the problem that savitzky Golay(SG)filtering algorithm does not make any different on the accuracy of the hyperspectral image classification,a TSG filtering algorithm is propose,and a multi-dimensional evaluation system for hyperspectral image space and spectral quality evaluation is also constructed.The hyperspectral image features is keeped,the signal-to-noise ratio of image is higher,and the spectral features are enhanced by using TSG filtering.The hyperspectral image is segmented and filtered by the methods that have proposed above,extracted top six principal components by principal component analysis algorithm.Finally the wheat seed pixels of hyperspectral image are classified by SVM algorithm based on these six principal components.By this way,the accuracy of the wheat scab classification system is 98.69%,a new idea about the detection of wheat scab by hyperspectral is also provided.