Mahalanobis distance and maximum likelihood based classification for identifying tobacco in Pakistan

Abstract

Classifying cash crops through satellite based remote sensing has proved to be effective for reliable ground based agricultural statistics. In this study, frequently used simple and fast classification algorithms i.e., Mahalanobis Distance and Maximum Likelihood Classification (MLC) are compared for classifying tobacco crops by the end of June in north-western Pakistan. High Geometric Resolution imagery of SPOT-5 (2.5m) is used as the base image for comparison over a large pilot region. Our results indicate that MLC is more accurate than its simple form Mahalanobis distance with overall accuracy of 93.91% and kappa coefficient of 0.9181. Though it is visually seen that MLC has over-estimated tobacco crops in the unclassified region but this effect is mitigated with the help of two additional classes namely `interfering separation' and `interfering settlements'. It is recommended to use and compare MLC for future detection of tobacco crops in north-western Pakistan.