Multi-Domain Divide-and-Conquer Method for UAV Infrared Image Super Resolution Based on State Space Module

To address the challenges of low resolution and sparse information in drone infrared images, this paper proposes a multi-domain divide-and-conquer drone infrared image super-resolution algorithm based on a state space model. The algorithm incorporates a state space model structure utilizing a selective scanning mechanism. By modeling long-range dependencies within features through the state space model, the approach effectively suppresses infrared image noise while enhancing the reconstruction capability for texture details and small target structures. During the feature mapping stage, a method based on Haar wavelet transform is designed to decouple high-frequency and low-frequency image features. Furthermore, spatial pyramid pooling and selective state space equations are employed to enhance local texture continuity and global semantic consistency. Finally, an adaptive multi-domain attention fusion method is introduced. This method aligns structural features across spatial and frequency domains using multi-scale convolution and incorporates an adaptive weighted cross-attention mechanism. It dynamically adjusts feature fusion weights via learnable parameters to improve model robustness. Extensive experimental results demonstrate that, compared to existing algorithms, the proposed method achieves superior reconstruction of infrared image details and textures, enhanced performance across multiple quantitative evaluation metrics.