Iot-Based Multiclass Fruit Classification System using Transfer Learning

Abstract

Agricultural products, fruits, and vegetables, can be of varying quality. Proper sorting and grading of these products are essential in national and international trade, hence the country's socioeconomic development. Currently, most agricultural sectors depend on the manual sorting of products, which is tedious and inaccurate. This paper addresses this issue by designing an IoT-based fruit sorting system using transfer learning. It uses a transfer learning-based classifier to identify the input fruit class. The fruit to be classified passes through a conveyor belt. The digital camera interfaced with the classifier takes the fruit image and gives it to the classifier network. The classifier network is a predesigned model trained using the fruits to be classified. The result of the classifier couples to a microcontroller-based system. Based on the type of the classified fruit, the microcontroller activates the piston system connected to the conveyor belt. The collecting boxes placed near the conveyor belt-piston arrangement contain the sorted fruits. A microcontroller-based system monitors the collecting box status. The sensor attached to the box sends signals regarding its status, whether full or not. If the box is 80% full, the microcontroller alerts that box overflow can be avoided. The box status is updated on the concerned website using a cloud system for further procedure.