Cluster-Detection-Based Local Super-Resolution Network for Remote Sensing Image Enhancement

Remote sensing images are essential for various applications such as environmental monitoring and urban planning. However, the spatial resolution of these images often hinders the accurate detection of small-scale or densely packed targets. In this work, we advocate a novel method that introduces clustering into local super-resolution networks to enhance image resolution and improve the detectability of small-scale and dense regions. First, K-means clustering is applied to identify dense areas based on both the color and spatial properties of image pixels, efficiently pinpointing regions where small objects are concentrated. Next, a lightweight cross-attention-based super-resolution network is employed to enhance the resolution of these key regions, which improves object detection accuracy. The proposed method is computationally efficient, incorporating depthwise separable convolutions and focusing the cross-attention mechanism on regions of interest, thereby minimizing overhead. Extensive experiments on the WV-3 and DOTA datasets demonstrate that our method significantly improves image quality, outperforming existing methods in terms of both PSNR and SSIM.