Automated Quality Evaluation Of Strawberries: A Computer Vision Approach

Abstract

This study addresses the challenge of real-time strawberry defect detection in agricultural settings by enhancing the You Only Look Once v8 (YOLOv8) object detection algorithm. Strawberry quality is crucial for consumer satisfaction and market value, but manual inspection for defects is labor-intensive and prone to error. Computer vision offers a promising solution, and YOLOv8’s speed and accuracy make it a suitable candidate for real-time deployment. The study proposes an improved YOLOv8 model through tuning incorporating specialized data augmentation techniques, as well as enhancing inference speed, to address the unique characteristics of strawberry defects. The model was trained on a diverse dataset of strawberry images capturing various defects. The effectiveness of the proposed approach was evaluated through rigorous testing in real-world scenarios. Results demonstrate a significant improvement in defect detection accuracy and speed compared to the baseline YOLOv8 model. The enhanced YOLOv8 achieved a higher mean Average Precision (mAP) and a faster inference time, enabling real-time defect detection on a standard computing device. To showcase the practicality of the proposed approach, a real-time demonstration system was developed, allowing for seamless integration into existing strawberry sorting processes. The demonstration system successfully identified defects in real-time, proving valuable feedback to farmers for quality control. This study’s contribution lies in its successful adaptation of the modified YOLOv8 model for the specific task of strawberry defect detection, its rigorous real-world evaluation, and the development of a practical demonstration system. The findings pave the way for the widespread adoption of computer vision in the agricultural industry, leading to improve product quality, reduced waste, and enhanced efficiency in strawberry production