Efficient and Accurate Identification of Maize Rust Disease Using Deep Learning Model

Pei Wang ^1* Jiajia Tan ¹ Yuheng Yang ¹ Tong Zhang ¹ Pengxin Wu ¹ Xinglong Tang ² Hui Li ^1* Xiongkui He ³ Xinping Chen ¹

• ¹ Southwest University, Chongqing, China
• ² Chongqing Academy of Agricultural Sciences, Chongqing, Chongqing Municipality, China
• ³ China Agricultural University, Beijing, Beijing Municipality, China

Common corn rust and southern corn rust are two typical diseases that affect maize during growth stages. Accurate differentiation between these rust species is crucial for understanding their occurrence patterns and associated pathogenic risks. To address this, a specialized Maize-Rust model is developed in this study for precise distinguish of these similar phenotypic symptoms. Initially, a SimAM module is integrated into the YOLOv8s backbone network to enhance feature extraction for both rust types. Additionally, a BiFPN is introduced to improve fusion across scales, particularly for detecting small disease spots. To expedite detection, a DWConv is used to streamline the model structure. Through the training and testing of the data set, the accuracy, average accuracy, recall rate and F1 value of the improved model are 94.6%, 91.6%, 85.4% and 0.823, respectively. The classification accuracy of the new model is 16.35% and 12.49% higher than that of Faster-RCNN and SSD models, respectively. The speed of detecting a single rust image is 16.18 frames/s. The proposed model is also deployed on mobile phones to achieve real-time data collection and analysis. It provides an effective support for the accurate detection of large-scale outbreaks of rust in the field.