Comparative Evaluation between Multispectral and Hyperspectral Data for Discrimination of Fruit Crops using Statistical Techniques

Abstract

Horticultural crops unlike field crops are perennial in nature, not having distinct phenology. It is difficult to discriminate horticultural crops using temporal multispectral data. Major limitation of multispectral data is lesser number of bands and mixed pixels which may not be able to discriminate fruit crops but the hyperspectral data has the advantage of having relatively large number of narrow, contiguous bands which lead to continuous spectral reflectance curve, making intricate details visible in the spectrum. For comparison of multispectral data with hyperspectral data, the hyperspectral data which have 2151 numbers of bands has been brought to multispectral level as because multispectral data has very less number of bands. Therefore, in the hyperspectral data, average at 50 nm, 100 nm and 250 nm interval was taken to reduces the data set into 42, 22 and 9 bands. The 4 tier statistical procedure which includes one way Analysis of variance (ANOVA), Classification and regression tree (CART), Jeffries-Matusita (J-M) distance and Linear discriminant analysis (LDA) technique was applied in the reduced band data set. The result of J-M distance and LDA were used to observe whether the reduced band data set can be able to discriminate the fruit crops. The study reveals the limitation of multispectral data in fruit crop discrimination. As the number of bands gets reduced the discriminative power of the data set also gets down.