AS-YOLO: Enhanced YOLO Using Ghost Bottleneck and Global Attention Mechanism for Apple Stem Segmentation

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Stem removal is a labor-intensive bottleneck in fruit harvesting, impacting fruit quality and marketability.
• Automated fruit harvesting requires accurate and rapid fruit and stem segmentation.

Purpose of the Study:

• To develop an enhanced You Only Look Once (YOLO) model, named AppleStem (AS)-YOLO, for precise apple stem segmentation.
• To improve computational efficiency and feature extraction for real-time applications.

Main Methods:

• The AS-YOLO model integrates a ghost bottleneck for parameter reduction and computational efficiency.
• A global attention mechanism is employed to enhance feature extraction capabilities.
• The model was evaluated on custom-built and open datasets.

Main Results:

• AS-YOLO achieved high accuracy with mean average precision (mAP)@50 of 0.956 and mAP@50-95 of 0.782.
• The model demonstrated a real-time inference speed of 129.8 frames per second (FPS).
• AS-YOLO outperformed existing state-of-the-art segmentation models.

Conclusions:

• The AS-YOLO model shows significant potential for real-time application in automated fruit harvesting.
• This advancement contributes to the broader field of agricultural automation by addressing a key harvesting challenge.