YOLO-AVCA-CBAMNet: Attention-driven framework for detection and classification of green pepper maturity stages

Area of Science:

• Computer Vision
• Machine Learning
• Agricultural Technology

Background:

• Accurate pepper berry maturity identification is crucial for optimal harvest timing and spice quality.
• Existing methods may struggle with diverse field conditions like varying illumination and complex backgrounds.

Purpose of the Study:

• To develop an integrated detection-and-classification framework, YOLO-AVCA-CBAMNet, for effective pepper berry maturity assessment in natural field settings.
• To enhance the reliability of maturity stage separation using complementary attention mechanisms.

Main Methods:

• Utilized a self-collected smartphone image dataset of pepper berries under varied conditions.
• Employed YOLOv8 for initial berry detection in cluttered scenes.
• Applied convolutional neural networks enhanced with Adaptive Visual Cortex Attention Module (AVCAM) and Convolutional Block Attention Module (CBAM) for classification.

Main Results:

• Achieved accuracy gains of 5-9% across different backbone architectures.
• The DenseNet121-based configuration reached a peak accuracy of 96.19%.
• Demonstrated improved discrimination of visually similar maturity stages through dual-attention mechanisms.

Conclusions:

• Attention-driven models show significant potential for interpretable, efficient, and scalable maturity assessment in precision agriculture.
• The YOLO-AVCA-CBAMNet framework offers a robust solution for real-world agricultural applications.
• The study validates the practical relevance of advanced deep learning techniques for crop monitoring.