Smart agriculture: optimizing mango grading with machine learning and image enhancement

Mango grading plays a vital role in ensuring fruit quality and maintaining market value. Traditional manual grading methods are often subjective, inconsistent, and inefficient, especially in large-scale production settings. In contrast, automated systems based on computer vision and deep learning provide a more reliable alternative. In this study, we propose an advanced mango grading system that integrates U2-Net for background removal and Adaptive Contrast Enhancement (ACE) to enhance image details, thereby improving classification accuracy. We evaluate the effectiveness of various preprocessing techniques in highlighting surface texture features of mangoes and compare the performance of different model architectures. Experimental results indicate that the proposed method significantly outperforms the top-performing AI CUP model and the CGAN + YOLO approach in appearance-based grading, without relying on complex processing workflows. Furthermore, to address the challenge of limited sample sizes, we explore the feasibility of using a Siamese Neural Network under data-scarce conditions. These findings underscore the potential of our approach for developing efficient and reliable automated fruit grading systems, contributing to the advancement of smart agriculture and enhancing quality control standards in commercial markets.