MNCM: Hybrid attention MobileNet for wheat (Triticum aestivum L.) disease identification

Abstract

Wheat is a staple crop of global significance, essential for food security and agricultural sustainability. Its production is frequently threatened by foliar diseases such as leaf rust, septoria, and powdery mildew, which can severely reduce both yield and grain quality. Accurate and timely identification of these diseases is therefore critical for effective crop management and loss prevention. However, existing deep learning models often struggle to balance recognition accuracy with computational efficiency, limiting their practicality for real-world field deployment. To address this challenge, we propose MNCM, a novel lightweight deep learning model for wheat disease identification. The model enhances MobileNetV3-large by integrating a channel prior convolutional attention module to refine local feature extraction and a multi-head self-attention mechanism to capture long-range dependencies across feature maps. This hybrid design improves feature discrimination while maintaining a compact architecture suitable for edge devices. Evaluated on a dataset of 7493 images encompassing five wheat leaf conditions, the proposed model achieves an overall classification accuracy rate of 97.49%, outperforming comparative models including OverLoCK-B, MobileMamba, and ConvNeXt. These results demonstrate that MNCM offers a robust and efficient solution for automated wheat disease diagnosis, with strong potential for integration into smart agriculture systems and field-deployable early warning platforms.