Date Fruit Classification System using Deep Transfer Learning

Abstract

Nowadays, developing an accurate computer vision system for classifying the type and maturity stage of date fruit is essential for designing an efficient and smart robotic harvesting solution. In this study, we proposed an effective deep transfer learning-based system for automatically classifying five distinct date types on bunches (Naboot saif, Khalas, Barhi, Meneifi, and Sullaj) with four different maturity stages (Immature, Khalal, Rutab, and Tamar) using transfer learning approach on pre-trained deep convolutions neural networks. Resenet-50 and AlexNet networks were trained on the available dataset for the two classification tasks (date types and maturity stage). The findings of performed various experiments suggested that the proposed Resenet-50 outperforms the AlexNet with maximum achieved accuracies of 95.1% and 94.7% for date type and maturity stage classification, respectively. The study’s findings are to be utilized in designing an intelligent robotic harvesting mechanism for efficient monitoring and harvesting of date fruits in a natural environment. Keywords Machine vision, Transfer learning, Neural networks, Real-time systems, Task analysis, Monitoring, Residual neural networks