BiSpectral Mamba: Quantum-Enhanced CNN Achieves 84.8% Accuracy in Crop Analysis

Hyperspectral image (HSI) analysis is critical for precision agriculture but suffers from high dimensionality, spatial complexity, and limited labeled data. To address these challenges, researchers Mohammad Salman Khan, Ehsan Atoofian, and Saad B. Ahmed introduced a novel BiSpectral Mamba framework. The model first uses a multi-scale CNN backbone to extract hierarchical spatial-spectral features through multi-resolution fusion. A spectral attention mechanism then emphasizes informative spectral bands while suppressing noise. The refined features are processed by a BiSpectral Mamba module, which captures long-range dependencies in both forward and backward directions by treating hyperspectral features as sequential tokens. Additionally, class-weighted optimization and feature fusion strategies mitigate class imbalance and improve training stability.

Evaluated on the UAVHSI-Crop dataset, the framework achieved 84.83% overall accuracy, demonstrating robust spatial-spectral feature learning. The integration of convolutional, attention-based, and state-space components with quantum-inspired learning sets a new benchmark for crop classification. Beyond this, the model shows potential for broader agricultural and remote sensing applications, including crop disease detection, yield prediction, and soil moisture estimation. This work highlights the effectiveness of structured state-space and quantum-inspired architectures for hyperspectral image analysis, paving the way for more efficient and accurate precision agriculture tools.