Decision template fusion for classifying Indian edible oils using singular value decomposition on NIR spectrometry data

Abstract

Edible oil is dominant part of human diet and available in various forms in the market. On the other hand, quality assurance in food industry is nowadays an important parameter for public concern and awareness. In this context, this paper presents a discrimination methodology of edible oils such as mustard, olive, rice bran, sunflower and soybean oils using a portable non-destructive near infrared spectrometer (NIR). Two tier approaches are taken for oil discrimination purpose. First is the significant features extraction from each oil's spectrum using singular value decomposition technique and the second one is the classification of oils using multiple classifier combination approach using decision template fusion technique. Support vector machine and multilayer perceptron classifier are also applied here. Experimental results clearly indicate the efficacy of decision template fusion technique with combination of support vector machine and multilayer perceptron classifier as compared to combination of same type of classifier.