AVIG: A Real-Time Visual Inspection for Guava Grading System Using Computer Vision and XGBoost

The conventional approach to manual inspection can be described by its considerable time and labor demands and the unavoidable occurrence of human error when faced with the inspection of a large number of products. Therefore, the application of an automated visual inspection system is frequently employed. The camera(s) can easily capture visual features for automatic visual inspection. However, it is essential to note that there are limitations when using this method to estimate the size-based grading system. The volume measurement is an essential feature in the field of automatic image analysis for fruit and vegetable grading, specifically guava. In this study, we propose a novel image analysis approach that utilizes a dual-view camera system to enhance the accuracy of volume approximation, particularly in scenarios involving non-perfect spherical objects. Hence, this research presents the training process of a comprehensive model for estimating the volume of objects with irregular shapes, named AVIG (Automatic Visual Inspection for Grading). The model utilizes the ellipsoid approximation technique, a dual-view camera, and XGBoost algorithm. The results from the accuracy comparison demonstrate that XGBoost shows the highest degree of accuracy, specifically 86%, thereby outperforming the performance of other algorithms. The comprehensive experiment conducted in this study has shown that the proposed architecture, AVIG, exhibits superior performance compared to previous research in terms of automatic evaluation of moving objects.