CassavaViM: MambaVision with Efficient Multi-Scale Attention for Cassava Leaf Disease Recognition

Cassava leaf diseases represent a critical challenge to crop yield and quality. Despite the growing need for accurate disease identification, existing solutions often fail to adequately capture the subtle and diverse symptoms of these diseases. To address this gap, we propose Cassava ViM, a novel model for cassava leaf disease recognition that builds upon the Mamba Vision architecture. Cassava ViM leverages a multi-scale fusion strategy and integrates an Efficient Multi-Scale Attention (EMA) module with Cross-Spatial Learning, enabling it to effectively capture disease patterns across varying scales while refining feature integration. Through comprehensive experiments on the Cassava Leaf Diseases Dataset, Cassava ViM achieves a remarkable accuracy of 87.590/0, showing a substantial improvement over known state-of-the-art methods. This technology enhances cassava's early disease detection capabilities, facilitating timely intervention and improving agricultural productivity in southern regions such as the Greater Bay Area.