Non-invasive Grading System for Banana Tiers using RGB Imaging and Deep Learning

Food losses transpire at postharvest and processing operations in developing countries, commonly caused by inaccurate manual classification of horticultural crops. The modernization of agricultural facilities and emerging technologies in agriculture has provided solutions for such losses and even increased productivity in a short duration and with higher precision. A non-invasive classification of bananas is presented in this paper, which grades banana tiers into different categories using digital images of bananas applied with deep learning techniques. The main objective of this paper is to develop a tier-based grading system for clustered fruits such as bananas and classify them in terms of quality (export class, middle class, and reject class), maturity (green, turning yellow, yellow, and overripe), and size (small, medium, and large). The classification models for the different grading parameters are developed using transfer learning and a fine-tuned VGG16 Deep CNN architecture. The system was able to automate the assembly line-like process of classifying bananas with a satisfactory overall accuracy using only a minimal number of image samples. The non-invasive technique will also serve as a paradigm for classifying other clustered fruits or horticultural crops.