To address the challenges of severe noise and difficult real-time processing in raindrop images under complex outdoor lighting conditions, this paper proposes an optimized Two-Dimensional Variational Mode Decomposition (2D-VMD) algorithm for image denoising and implements a hardware-accelerated system based on the ZYNQ heterogeneous platform. At the algorithmic level, the optimal number of modes (K=6) is determined through multi-metric evaluation. A selective reconstruction strategy is constructed by integrating criteria such as frequency-domain energy and physical correlation, which effectively suppresses noise while significantly enhancing the preservation of raindrop edge and morphological details. At the hardware level, a parallel pipeline acceleration architecture is designed, covering stages from image capture and RGB-to-YCbCr conversion to the core 2D-VMD computation. Key operations, including complex multiplication and two-dimensional FFT, are deeply mapped to the Programmable Logic (PL) side. Experimental results show that the proposed algorithm achieves comprehensive performance on raindrop image sequences with a PSNR of approximately 24.97 dB, an SSIM of 0.8336, and a maximum edge preservation ratio of 0.98. Furthermore, the hardware-accelerated system meets the real-time processing requirement of 4K@60 fps. The conclusion demonstrates that the proposed algorithm-hardware co-design scheme excels in both denoising effectiveness and processing real-time performance, providing an efficient and feasible technical approach for reliable visual monitoring in high-altitude unattended environments.
Xiong et al. (Thu,) studied this question.