AMS-YOLO: multi-scale feature integration for intelligent plant protection against maize pests

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Maize is a vital global food crop threatened by pests, impacting yield and quality.
• Current pest detection methods struggle with diverse pest appearances, similarities, and complex field conditions.
• Accurate and efficient pest identification is essential for effective agricultural management and sustainable crop protection.

Purpose of the Study:

• To develop an enhanced detection model for maize pests that overcomes limitations of existing methods.
• To improve the accuracy and efficiency of maize pest identification in challenging agricultural environments.
• To create a lightweight and deployable model for precision agriculture applications.

Main Methods:

• Developed AMS-YOLO, an enhanced detection model based on YOLOv8n.
• Integrated three synergistic modules: SMCA attention mechanism, MSBlock multi-scale feature fusion, and AMConv optimized downsampling.
• Trained and evaluated the model on a dataset of 13 common maize pests across developmental stages.

Main Results:

• AMS-YOLO achieved 90.0% precision, 89.8% recall, and 94.2% mAP50, outperforming YOLOv8n.
• Demonstrated superior performance compared to other state-of-the-art methods like SSD and RT-DETR.
• The model is lightweight (5.3MB) with reduced parameters and computational requirements, suitable for edge devices like Jetson Nano.

Conclusions:

• AMS-YOLO effectively addresses challenges in maize pest detection through targeted architectural enhancements.
• The model's lightweight design and high accuracy enable field deployment in resource-constrained environments.
• This advancement supports precision pesticide application, resource optimization, and intelligent plant protection for sustainable agriculture.