ABSTRACT
In agriculture, the prohibitively time-consuming manual fruit grading evaluation process can be improved with the use of computer vision and machine learning approaches. In this chapter, one of the novel non-destructive feature extraction technique called fuzzy weighted nuclear norm based two-dimensional linear discriminant analysis (FNN-2DLDA) is proposed. The proposed FNN-2DLDA technique is mainly divided into three phases: first, fuzzy set theory is used to incorporate the overlapped samples distribution in scatter matrices. Second, nuclear norm has been used to get the best discriminant features in the optimal projection direction. Finally, support vector machine (SVM) classifier combined with extracted fruit image features are used to grade the pomegranate and self-built mango fruits. The existing unilateral and bilateral dimensionality reduction techniques viz. F-norm based 2DLDA, Fuzzy 2DLDA, nuclear norm based 2DLDA and proposed FNN-2DLDA are evaluated on the pomegranate and self-built mango fruit image data set. The average classification accuracy, statistical performance measurements, and hypothesis test results show that the proposed FNN-2DLDA technique dominates the prevailing techniques since fuzzy membership and nuclear norm approach are incorporated to the traditional 2DLDA technique.
