Automatic Detection and Grading of Multiple Fruits by Machine Learning

Abstract

Classification of various types of fruits and identification of the grading of fruit is a burdensome challenge due to the mass production of fruit products. In order to distinguish and evaluate the quality of fruits more precisely, this paper presents a system that discriminates among four types of fruits and analyzes the rank of the fruit-based on its quality. Firstly, the algorithm extracts the red, green, and blue values of the images and then the background of images was detached by the split-and-merge algorithm. Next, the multiple features (30 features) namely color, statistical, textural, and geometrical features are extracted. To differentiate between the fruit type, only geometrical features (12 features), other features are used in the quality evaluation of fruit. Furthermore, four different classifiers k-nearest neighbor (k-NN), support vector machine (SVM), sparse representative classifier (SRC), and artificial neural network (ANN) are used to classify the quality. The classifier has been contemplated with four different databases of fruits: one having 4359 color images of apples; out of which 2342, are with various defects, second having 918 color images of avocado out of which 491 are of with various defects, third having 3805 color images of banana out of which 2224 are with various defects, and fourth having 3050 color images of oranges out of which 1590 are with various defects. The system performance has been validated using the k-fold cross-validation technique by considering different values of k. The maximum accuracy achieved for fruit detection is 80.00% (k-NN), 85.51% (SRC), 91.03% (ANN), and 98.48% (SVM) for k = 10.The classification among Rank1, Rank2, and defected maximum accuracy is 77.24% (k-NN), 82.75% (SRC), 88.27% (ANN), and 95.72% (SVM) achieved by the system. SVM has seen to be more effective in quality evaluation and results obtained are encouraging and comparable with the state of art techniques.