A microcontroller based machine vision approach for tomato grading and sorting using SVM classifier

Advances in computer vision have led to the development of promising solutions for challenging problems in agriculture. Fruit grading and sorting are complex problems which require a great deal of human expertise. In this paper, we propose a non-destructive system for sorting and grading tomatoes, which is confounding even for expert human sorters. The proposed system performs classifications of tomatoes in three stages with digital images of samples captured in an experimental setup deployed using microcontroller. In the first stage, a binary classification is performed to discriminate tomatoes from other species using a species vector constructed from these images. In the second stage, the tomatoes are classified into ripe and unripe categories based on the color attribute. Then, the defects in the fruits are identified using Gabor wavelet transform to segment the infected regions of these images. The third stage identifies three types of defects namely the black spots, cankers and Melanose, based on a defect vector constructed from additional color and geometric features. Due to the complexity involved in solving such a non-linear problem, the proposed system is implemented as a cascade of two support vector machine classifiers. The performance of this system is assessed with the accuracy, specificity, sensitivity and precision metrics. The experimental results and comparative analyses with similar methods testify the efficacy of the proposed system over existing systems on the sorting and grading of tomatoes. The results obtained from each of the three classification stages i.e. Tomato/Non-Tomato, Good/Defective and the type of defect in the case of defective are communicated to the microcontroller to enable the respective motor, so that the given fruit is classified and collected in the respective bin. (C) 2020 Elsevier B.V. All rights reserved.