Mausoleum Scenic Spot5 Fusion And Classification Techniques

In this paper, the author took ZhongShan Cemetery as the study region .On the basis of geometric correction, Topographic Normalize and Spectral Character Analysis for original SPOT5 remote sensing images data, Then a series of general image transformations such as interactive contrast stretching, ratio enhancement, principal components transformation, tasseled-cap transformation, MNF transformation (minimum noise fraction rotation) and image color combine, were implemented. After discussing and analysising the Statistical Characters of multi-spectral band images, two indexes were calculated to estimate the best bands union for color combination, one is the determinant of the co-variance matrix, the other is Optimum Index Factor. It can be seen from the result that for color combination the original optimal bands were Band 4,Band 2 and Band 1, the best mixed images were PC1,BR and MNF1. On the basis of studying images fusion technology, the SPOT5 panchromatic band were merged with Band 4,Band 2 and Band 1 through six different data fusion algorithms, which were HIS transform, K-L transform, K-T transform, Brovey fusion and wavelet transform fusion. We estimated the fusion images in both subjective and objective factors. The objective parameters involved entropy, correlation coefficient and so on. And then each merged images were used to images classified automatically. Among classified automatically, special vector drawings, seed pixel expanded and majority filter were used to select training swatches. At one time, the author groped for the classified model of decision trees united with mahalanobis-Distance. Lastly, the classified images were dealed with such as clump analysis, sieve analysis and eliminate analysis. It can be showed from this research, the fused images had higher spatial resolution while maintaining the basic spectral contents of the original band4,2,1 images, and the visual effect and the accuracy of the classification had been greatly enhanced. Among the six fusion images, the Brovey image ,wavelet transform image and K-L transform image are good fusion methods, which can be applied in forest area resource survey .But while the study region was classified into three forest styles, the highest classification accuracy is only 74.60%, the whole kappa coefficient is only 0.6972.There is a further way in practicality. Through attempting to introduce into decision tree idea, firstly picked-up forest styles which are easily mixed classification each other, then separating the mixed styles and using mahalanobis-Distance to classify the residual image, lastly merging the each forest styles. It can be seen that the whole classification accuracy and kappa coefficient are improved.