A lightweight co-optimization model for field sunflower disease identification

Xiao Wu¹Liqian Zhang^1,2*Yaogeng Wang¹Yunli Bai^1,2

• ¹College of Computer and Information Engineering, Inner Mongolia Agricultural University, Hohhot, China
• ²Inner Mongolia Autonomous Region Key Laboratory of Big Data Research and Application of Agriculture and Animal Husbandry, Hohhot, China

Accurate identification of crop diseases is crucial for ensuring crop quality and yield. However, existing deep learning models lack robustness in complex field environments and suffer from large model parameter sizes, making them difficult to deploy on resource-constrained devices. To address these challenges, this paper proposed a lightweight YOLO-CGA model for sunflower disease identification, and then the model was deployed on a Raspberry Pi to bridge the gap between laboratory models and field applications. The proposed model incorporates three key improvements: (1) Designing a CBAM_ADown module that integrates attention mechanisms with the asymmetric down sampling to enhance feature extraction and noise suppression in complex image backgrounds. (2) Replacing C2f module of YOLOv8n-cls with the C3ghost module, which utilizes ghost convolution to reduce the number of parameters while preserving fine-grained features. (3) Constructing the AFC_SPPF module, which aggregates multi-scale disease features through a multi-branch adaptive fusion structure, and improves recognition performance for diverse lesions. Through applying these strategies, our proposed model YOLO-CGA achieves high accuracy on three major datasets which are 98.48% (BARI-Sunflower dataset), 98.32% (Cotton disease dataset), and 91.11% (FGVC8). And the number of parameters of proposed model is only 0.92M, which is much less than other models. Finally, the model was deployed on a Raspberry Pi end device for fulfill the demand of field disease identification.