Improved adaptive FPGA dark channel prior dehazing algorithm for edge applications in agricultural scenarios

In agricultural application scenarios, hazy environments often cause image blur, affecting the accuracy and efficiency of tasks such as crop monitoring, pest/disease identification, and drone inspection. This study presents an FPGA-accelerated edge computing system for an adaptive real-time dehazing platform in agricultural drones. The hardware acceleration unit implements an enhanced Dark Channel Prior algorithm, which features adaptive window filtering to refine dark channel images. Through field data acquisition and a self-developed adaptive mechanism, the system achieves adaptive processing across varying fog densities while mitigating the screen flickering inherent to adaptive systems. Based on the sky brightness distribution, a more effective sky-region segmentation strategy was designed to address overexposure in the sky region of dehazed images. Implemented on an Xilinx XC7S25 FPGA platform, the design supports 1920× 1080@60 Hz video input and integrates seamlessly with smart agriculture platforms through RTSP streaming. Using edge-computing strategies, the solution demonstrated efficient resource utilization (LUT: 8,241; FF: 3,168) and low power consumption (< 2 W). The experimental results show that the optimized algorithm maintains superior dehazing quality (outperforming other dehazing algorithms in the field on metrics including PSNR, SSIM, and information entropy). • FPGA-based real-time dehazing system for agricultural videos with 1920×1080@60Hz processing capability. • Improved dark channel prior (DCP) dehazing algorithm. • Introduces a concentration-triggered switching mechanism with adaptive parameter adjustment. • Proposes a novel luminance-based sky region segmentation strategy.