Performance Analysis of YOLO Models for Strawberry Ripeness Detection in Challenging Environments

The application of deep learning in agricultural tasks, such as ripeness detection of fruits has undergone rapid advancements. However, the real time challenges faced by these models under adverse imaging conditions remain underexplored. This study evaluates the performance of You Only Look Once (YOLOv8) models for strawberry ripeness detection across four imaging scenarios: normal, blurred, darkened, and noisy conditions. A comprehensive dataset was prepared, incorporating diverse environmental challenges to simulate real-world conditions. The YOLOv8 model was trained and validated on each dataset, with performance evaluated using metrics such as precision, recall, mean Average Precision at Intersection over Union threshold 50% (mAP@50), and mAP across thresholds from 50% to 95% (mAP@50-95). Results reveal substantial variations in detection accuracy across conditions, highlighting the model’s sensitivity to image quality degradation, particularly in blurry and noisy scenarios. Despite struggling in blurry and noisy conditions, YOLOv8 demonstrated robust detection capabilities in normal and darkened conditions. This study provides valuable insights into the capabilities and limitations of current object detection models in strawberry cultivation and provides future directions for further scaling and enhancing the model into a cohesive setup for automated harvesting. These findings contribute to the development of more reliable machine learning systems in the field of agricultural automation