YOLO-ALW: An Enhanced High-Precision Model for Chili Maturity Detection

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Accurate chili pepper maturity detection is crucial for agriculture but challenged by occlusion, small target size, and color similarity.
• Existing object detection models struggle with these specific challenges in complex agricultural environments.

Purpose of the Study:

• To develop an enhanced object detection model, YOLO-ALW, for high-precision chili pepper detection and maturity recognition.
• To address limitations in detecting occluded, small, or background-similar chili peppers.

Main Methods:

• An enhanced YOLOv8n model (YOLO-ALW) was developed, incorporating an Alterable Kernel Convolution (AKConv) module in the head.
• The backbone features a Spatial Pyramid Pooling Fast-Large Separable Kernel Attention (SPPF_LSKA) module for multi-scale feature integration.
• Wise-Intersection over Union (Wise-IoU) loss function was employed to optimize bounding box regression.

Main Results:

• YOLO-ALW achieved a mean average precision (mAP_0.5) of 99.1%, with precision at 98.3% and recall at 97.8%.
• The model outperformed the original YOLOv8n by 3.4% in mAP, 5.1% in precision, and 9.0% in recall.
• Grad-CAM visualization confirmed improved focus on critical chili pepper features.

Conclusions:

• YOLO-ALW significantly enhances chili pepper detection accuracy, particularly under challenging conditions like occlusion and background similarity.
• The model's performance demonstrates its potential for reliable maturity recognition, supporting automated harvesting applications.
• The integration of AKConv, SPPF_LSKA, and Wise-IoU offers a robust solution for agricultural object detection tasks.