February 24, 2025 | Journal of Food Composition and Analysis |
Researchers from Universiti Malaya and Chuzhou University, China, conducted a study aimed at improving the efficiency and accuracy of mangosteen grading by developing an automated grading system. Traditional manual grading methods are often labor-intensive, inconsistent, and prone to error, which can impact post-harvest quality management and economic returns.
The study introduced a specialized hardware system integrated with a machine learning-based grading model, Faster-FRNet. This model is an enhanced version of the Faster R-CNN framework, incorporating ResNet50 and a Feature Pyramid Network (FPN) to enable better multi-scale feature extraction. The system was designed to provide a scalable and high-accuracy grading solution suitable for large-scale agricultural applications.
Experimental results demonstrated that the proposed system achieved a grading accuracy of 98.75% and a mean Average Precision (mAP) of 0.68. This performance surpassed that of the standard Faster R-CNN with ResNet50 (0.51 mAP) and VGG16 (0.45 mAP). In addition to improved accuracy and speed, the system reduced computational complexity, making it more practical for real-world use.
The findings highlight the potential of the Faster-FRNet model to enhance post-harvest management in mangosteen production. The system provides a promising solution for broader application in the agricultural sector, supporting quality control and productivity.
