An intelligent mangosteen grading system based on an improved convolutional neural network

Efficient grading of mangosteens is vital in ensuring timely post-harvest storage and preservation for maximizing profits. Currently, manual grading is susceptible to subjective biases, thereby warranting a more intelligent grading approach. Innovative solutions for automated fruit grading have been developed based on computer vision. However, intelligent grading of mangosteens based on computer vision is challenging due to the different appearance and complex characteristics of mangosteens, coupled with the high development costs and challenges in widespread adoption of the grading technology. This study aims to address the limitations in mangosteen grading system. A specialized hardware setup is designed to efficiently transfer the fruits to the conveyor belt using a toggling material device. In addition, this work proposed a novel fruit grading model based on computer vision approach namely New MobileNetV3 InceptionV3 Network (NewMoInNet) model. Furthermore, a data visualization platform tailored to the mangosteen grading system's requirements is developed. Experimental results demonstrated an impressive grading accuracy of 97.15%, with an average grading speed 5.06 times faster than the manual method. In conclusion, the proposed system demonstrated significant speed, reliability, efficiency in work, and robustness compared to the conventional grading approach.