High-throughput end-to-end aphid honeydew excretion behavior recognition method based on rapid adaptive...

Area of Science:

• Agricultural Entomology
• Computer Vision
• Machine Learning

Background:

• Aphids are major agricultural pests and plant virus vectors.
• Current methods for monitoring aphid behavior are inefficient and lack real-time capabilities.
• Understanding aphid feeding activity and plant resistance requires accurate behavior analysis.

Purpose of the Study:

• To develop an automated, end-to-end framework for detecting multiple aphid behaviors.
• To enhance the accuracy and real-time capability of aphid behavior recognition.
• To provide a reliable tool for evaluating plant resistance and managing aphid pests.

Main Methods:

• Created the first fine-grained dataset for aphid Crawling Locomotion (CL), Leg Flicking (LF), and Honeydew Excretion (HE) behaviors.
• Developed a rapid adaptive motion feature fusion algorithm for high-granularity spatiotemporal feature extraction.
• Optimized the RT-DETR detection model by integrating a spline-based activation function and the Kolmogorov-Arnold network (RK50 module).

Main Results:

• The proposed framework achieved an average precision of 85.9%.
• The RK50 module improved mean average precision (mAP50) by 2.9% compared to the baseline model.
• Demonstrated superior performance in detecting small-target honeydew compared to mainstream algorithms.

Conclusions:

• The research presents an innovative solution for automated monitoring of fine-grained aphid behaviors.
• The developed framework offers improved accuracy and efficiency over conventional methods.
• The study provides a valuable reference for insect behavior recognition research, with resources shared on GitHub.