The Impact of Data Compression on Supervised Classification of Airborne Multispectral Imagery

摘要

An assessment was made of the impact of on-board data compression on the information which a user could extract from multispectral imagery by supervised classification. To simulate the effect of data compression, available airborne MSS data on CCT were compressed and reconstructed and stored on another CCT in the original format. Both CCTs, containing the original and reconstructed miltispectral images of an agricultural scene in the Netherlands, were processed by a program for supervised maximum likelihood classification, while the ground locations of the training areas were exactly identical. The effects of data compression and the influence of training area orientation on inter- and intra-class statistics and on the final classification results were investigated. Data compression enhances overall class separability, but reduces distinction between similar classes. Along scan lines, assignment to classes becomes more uniform. Training samples from reconstructed imagery are less representative for along track signal variation, causing many image lines to remain unclassified.