A Machine Vision-Based Maturity Prediction System for Sorting of Harvested Mangoes

Seasonal fruits, like mango (Mangifera Indica L.), are harvested from gardens or farms in batches; the mangoes present in each batch are not uniformly matured, therefore, sorting of mangoes into different groups is necessary for transporting them into different locations. With this background, this paper proposes a machine vision-based system for classification of mangoes by predicting maturity level, and aimed to replace manual sorting system. The prediction of maturity level has been performed from the video signal collected by the Charge Coupled Device (CCD) camera placed on the top of the conveyer belt carrying mangoes. Extracted image frames from the video signal have been corrected and processed to extract various features, which were found to be more relevant for the prediction of maturity level. Recursive feature elimination technique in combination with support vector machine (SVM)-based classifier has been employed to identify the most relevant features among the initially chosen 27 features. Finally, the optimum set of reduced number of features have been obtained and used for classification of the mangoes into four different classes according to the maturity level. For classification, an ensemble of seven binary SVM classifiers has been combined in error correcting output code, and the minimum hamming distance-based rule has been applied in decision making phase. For the experimental study, the mangoes of five different varieties were collected from three different locations and in three different batches. The obtained experimental result found to provide an average classification accuracy up to 96%.