DELP-YOLOv12: a lightweight deployable model for maize pest and disease detection

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Maize yield is threatened by pests and diseases, necessitating accurate detection.
• Existing models struggle with small objects, feature perception, and edge device deployment.
• Field conditions like variable illumination complicate pest and disease identification.

Purpose of the Study:

• To develop a deployable and lightweight maize pest and disease detection framework.
• To improve recognition of small-scale pests and irregular lesions.
• To enhance model efficiency for edge device deployment in agriculture.

Main Methods:

• Proposed DELP-YOLOv12 framework based on YOLOv12 architecture.
• Integrated Dynamic RepConvBlock with NAM (DRN) module for enhanced feature representation.
• Developed Lightweight Feature Enhancement Detection Head (LFEDH) for small-object recognition.
• Incorporated Efficient Channel Attention (ECA) for improved feature discrimination.
• Applied Layer-Adaptive Sparsity for Magnitude-based Pruning (LAMP) for model compression.

Main Results:

• DELP-YOLOv12 achieved 94.1% precision, 89.6% recall, and 94.2% mAP50.
• Outperformed baseline models across all performance metrics.
• Reduced parameter count by ~68% and computation cost by ~60%.
• Maintained real-time inference on embedded hardware (e.g., NVIDIA Jetson Orin NX).

Conclusions:

• DELP-YOLOv12 effectively balances accuracy, efficiency, and deployability for maize pest and disease detection.
• Enhances recognition of small/irregular targets with low computational demand.
• Offers a practical solution for real-time agricultural monitoring and intelligent pest management.