COMPARING A PIECEWISE-LINEAR CLASSIFIER WITH GAUSSIAN MAXIMUM-LIKELIHOOD AND PARALLELEPIPED CLASSIFIERS IN TERMS OF ACCURACY AND SPEED

A piecewise linear classifier (PLC) was developed and tested to determ ine if it is superior to the Gaussian maximum likelihood classifier (G MLC) and parallelepiped classifier (PPC) for inventories of crop types in terms of classification accuracy and speed. The PLC was developed based upon the concepts of the single-sided decision surface, optimal weight vector, and seniority decision logic. These three classificatio n algorithms were evaluated using multitemporal digitized video data. The PLC was much faster than the GMLC, and yet provided similar classi fication accuracy. Although the PLC was somewhat slower than the PPC, it provided much higher classification accuracy than did the PPC. The PLC was determined to be an optimal alternative to the GMLC or PPC for inventories of crop types in terms of classification accuracy and pro cessing speed.