An Intelligent Guava Grading System Based on Machine Vision

Abstract

Ensuring efficient grading of guavas is crucial for timely postharvest storage and maximizing profits. Currently, the subjective nature of manual grading underscores the need for more sophisticated methodologies. However, employing machine vision for intelligent grading faces hurdles due to the diverse characteristics of guavas and the high development costs. This research targets the limitations in the guava grading process and introduces an intelligent system to overcome them. The system's structure and operational procedures were outlined, establishing diverse standards encompassing guava color, shape, size, and integrity. Image capture and preprocessing of guavas are completed. Employing the RGB model, the study performed color feature extraction and guava recognition, alongside diameter and integrity assessment through edge detection. Following a thorough analysis of various models, ResNet50 emerged as the preferred choice for guava image evaluation and depth recognition. Subsequently, an intelligent guava grading system was developed using Microsoft Visual Studio 2017. Experimental results demonstrated outstanding grading accuracy of 98.05%, with grading speed averaging 5.47 times faster than manual methods. Compared to traditional manual grading techniques, the system excelled in work efficiency, speed, reliability, and robustness.