To address the adverse impacts of fog interference on observation systems, including target recognition difficulties and accuracy reduction, a novel image defogging algorithm based on a global atmospheric optimization model was proposed. A comprehensive defogging framework was constructed, integrating an adaptive dark channel refinement mechanism, a transmittance joint optimization strategy, and an atmospheric light iterative correction method. Firstly, a dynamic dark channel calculation method based on multi-scale windows was designed, with nonlinear enhancement coefficients introduced to improve fog concentration sensitivity. Secondly, a transmittance optimization model incorporating guided filtering and contrast constraints was established to resolve image edge distortion issues. Finally, an atmospheric light correction algorithm combining regional clustering screening and transmittance weighting was developed to overcome the limitations of fixed-region sampling in conventional methods. Experimental results demonstrated that compared with Wiener filtering and multi-scale Retinex models, the proposed method achieved minimum improvements of 16.6% in PSNR(peak signal-to-noise ratio), 19.2% in SSIM(structural similarity), and 16.9% in Entropy metrics. When applied to YOLOv5 object detection models, precision and recall rates were enhanced by over 10.3% and 12.7% respectively, significantly strengthening image analysis and target identification capabilities of observation systems in complex environments.