A study on an efficient citrus Huanglong disease detection algorithm based on three-channel aggregated attention.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Pathology

Background:

• Citrus Huanglong disease (HLB) presents complex field symptoms challenging traditional detection methods.
• Existing methods suffer from low efficiency and insufficient recognition accuracy.
• There is a need for advanced algorithms for accurate and rapid HLB identification.

Purpose of the Study:

• To develop an efficient and accurate detection algorithm for citrus Huanglong disease (HLB).
• To improve upon the You Only Look Once (YOLO)v8 architecture for enhanced disease feature recognition.
• To provide a tool for early and accurate identification of HLB in field conditions.

Main Methods:

• An improved YOLOv8 algorithm incorporating a novel character to float (C2f) Attention inverse residual moving block (IRMB) module.
• Integration of a three-channel aggregated attention module (Powerneck) for efficient cross-scale feature interaction.
• Optimization of the detection head using structural reparameterization for accelerated inference.

Main Results:

• The proposed model achieved a mAP50 of 97% and an accuracy of 91.5% on a citrus dataset.
• Inference speed was improved by 14.6% to 370 frames per second (FPS).
• The enhanced model outperformed the original YOLOv8 and other mainstream models in accuracy and speed balance.

Conclusions:

• The improved YOLOv8 algorithm effectively addresses the limitations of traditional citrus disease detection.
• The C2f Attention IRMB and Powerneck modules synergistically enhance model performance for HLB detection.
• This algorithm offers practical significance for reducing pesticide misuse and improving orchard management efficiency.