YOLOv8-GO: A Lightweight Model for Prompt Detection of Foliar Maize Diseases

Abstract

Disease is one of the primary threats to maize growth. Currently, maize disease detection is mainly conducted in laboratories, making it difficult to promptly respond to diseases. To enable detection in the field, a lightweight model is required. Therefore, this paper proposes a lightweight model, YOLOv8-GO, optimized from the YOLOv8 (You Only Look Once version 8) model. The Global Attention Mechanism was introduced before the SPPF (Spatial Pyramid Pooling Fast) layer to enhance the model’s feature extraction capabilities without significantly increasing computational complexity. Additionally, Omni-dimensional Dynamic Convolution was employed to optimize the model’s basic convolutional structure, bottleneck structure, and C2f (Faster Implementation of CSP (Cross Stage Partial) Bottleneck with two convolutions) module, improving feature fusion quality and reducing computational complexity. Compared to the base model, YOLOv8-GO achieved improvements across all metrics, with mAP@50 increasing to 88.4%, a 2% gain. The computational complexity was 9.1 GFLOPs, and the model could run up to 275.1 FPS. YOLOv8-GO maintains a lightweight design while accurately detecting maize disease targets, making it suitable for application in resource-constrained environments.