YDCNet: YCbCr and Depth-Aware Collaborative Approach for Remote Sensing Image Dehazing

Satellite imagery is indispensable for military reconnaissance and environmental monitoring, yet haze pollution severely degrades image quality and target recognition accuracy. Existing methods relying solely on the RGB color space often lead to suboptimal dehazing results with residual haze artifacts. To overcome this limitation, we propose a YCbCr and Depth-guided Collaborative Network (YDCNet) by integrating depth estimation and YCbCr features. First, a multi-stage architecture incorporating a frequency-domain attention-guided module (SWATT) is constructed to strengthen feature representation in both spatial and frequency domains. Second, a dilated shiftwise attention convolution module is designed to extract haze features with a large receptive field. Finally, a hybrid YCbCr feature constraint strategy is introduced, which suppresses color distortion through joint optimization and coordination of the depth estimation network and the pyramid network. Experiments show that YDCNet can achieve significant improvements in terms of PSNR and SSIM metrics, effectively improving visual quality.