ERL-YOLOv8n: A Potato Disease Detection Network for Complex Field Environments

Authors:
  • FGFeng Gong
    Feng Gong
  • YCYingsheng Chen
    Yingsheng Chen
Publisher website
Google Scholar
Abstract
To address the challenge of balancing accuracy and efficiency in potato disease detection models under complex field conditions, this study proposes an improved lightweight detection algorithm, ERL-YOLOv8n. The algorithm optimizes the YOLOv8n model in three key aspects: First, the Receptive Field Attention Convolutional Block Attention Module (RFCBAMConv) is introduced into the backbone network to enhance the model's adaptive perception of disease spot features at various scales. Second, an EPSA (ECA and Polarized Self-Attention) module is embedded in the neck network to improve the model's anti-interference capability in complex backgrounds by fusing multi-dimensional feature information. Finally, the LSGE (Large Selective Kernel Network and Spatial Group-wise Enhance) attention mechanism is incorporated to synergistically optimize the backbone and neck, effectively improving detection accuracy for large-scale targets and complex scenes. Ablation and comparative experiments demonstrate that the ERL-YOLOv8n model achieves significant improvements in key performance metrics while maintaining high detection speed. Compared to the original YOLOv8n model, its precision, recall, and mAP@50 increased by 1.4, 5.7, and 2.3 percentage points, respectively. The improved model exhibits enhanced robustness and superior detection performance, particularly in the early identification of early and late blight, providing reliable technical support for practical applications in precision agriculture.