YOLO-Punica

YOLO-Punica improves the YOLOv8n object detection algorithm to enable real-time monitoring of pomegranate fruit development with reduced computational requirements.

Key Features:

• Dual-Path Downsampling Module (DPDM): Integrated into the backbone network to enhance detection precision and computational efficiency.
• Cross-Scale Feature Fusion Module (CCFM): Combined with DPDM in the neck structure to reduce parameters, memory consumption, and model size while improving detection accuracy.
• Lightweight Architecture: Achieves 45.8% reduction in parameters, 28% reduction in GFLOP, and 43.7% reduction in model size compared to YOLOv8n.
• Real-Time Embedded Processing: Processes images at 14.3 frames per second on embedded devices for low-computational-power environments.

Scientific Applications:

• Pomegranate Fruit Development Detection: Provides real-time machine vision monitoring of pomegranate fruit development to support accurate intelligent agricultural management.
• Agricultural Machine Vision Enhancement: Offers a methodological framework applicable to improving detection models in other agricultural contexts.

Methodology:

YOLO-Punica modifies the YOLOv8n architecture by integrating DPDM into the backbone network and both CCFM and DPDM into the neck structure, evaluated on public datasets using mean average precision (mAP).