Simultaneous Identification on Tomato Variety and Maturity Based on Local and Global Feature Fusion.

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Tomato variety and maturity classification is vital for crop evaluation but challenging due to environmental factors like leaf occlusion.
• Existing methods struggle with accuracy and efficiency in complex field conditions.

Purpose of the Study:

• To develop an innovative, simultaneous detection model for tomato variety and maturity.
• To overcome limitations of current identification methods in complex agricultural environments.

Main Methods:

• An improved YOLOv8n model incorporating frequency-adaptive dilated convolution (FADC) and high-level screening-feature path aggregation network (HSPAN) for feature fusion.
• Integration of channel attention and feature selection mechanisms for enhanced local and global feature integration.
• Utilization of Powerful-IoU (PIoU) loss function and a dynamic detection head for improved bounding box accuracy and adaptive feature extraction.

Main Results:

• The proposed model demonstrated superior global perception capabilities.
• Achieved the highest detection accuracy compared to other models evaluated.
• Exhibited lower computational complexity, indicating efficiency.

Conclusions:

• The developed model effectively addresses challenges in tomato variety and maturity detection.
• The combination of FADC, HSPAN, PIoU loss, and dynamic detection head significantly improves recognition accuracy and efficiency.
• This approach offers a promising solution for automated agricultural monitoring and yield assessment.