Banana Ripeness Classification with Deep CNN on NVIDIA Jetson Xavier AGX
N Aishwarya, Vinesh Kumar R
Abstract
Efficiently managing supply chains, reducing food waste, and ensuring product quality in consumer goods relies significantly on the precise grading of banana maturity. This task is complex due to the subtle morphological and textural changes occurring during ripening. Addressing this challenge, this research introduces a deep YOLOv8 neural network approach for classifying bananas into six categories: fresh-ripe, fresh-unripe, overripe, ripe, rotten, and unripe. The study involves training and evaluating five YOLOv8 models—YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l, and YOLOv8x—using a dataset of 18,074 images. The models achieved detection accuracy ranging from 94.6% (YOLOv8n) to 96.3% (YOLOv8x) for mean average precision (mAP) with an Intersection of Union (IoU) of 0.5. Notably, YOLOv8s displayed strong potential for real-time fruit ripeness classification. For single image predictions, the estimated processing time on Nvidia Jetson Xavier AGX varied from 13.8ms (YOLOv8n) to 230.4ms (YOLOv8x), respectively.