Global-local hybrid cloud simulation with Gaussian-Perlin-FBM fusion and diffusion-attenuation for robust aircraft recognition in remote sensing imagery

With the advancement of the economy and society, aircraft recognition based on satellite remote sensing images has become increasingly important for applications such as airport traffic monitoring and regional economic analysis. However, cloud occlusion severely limits the usability of satellite imagery for accurate target recognition. According to global meteorological data, nearly 70% of the Earth’s surface is covered by clouds, posing a critical challenge for robust recognition. Most existing models are trained on clean datasets with little consideration of cloudy scenarios, while effective cloud simulation methods remain scarce. To address this gap, we propose a global-local hybrid cloud simulation framework for generating cloud-covered datasets. At the global scale, a Gaussian blur-based weighted fusion is applied to construct a new base image, on which Perlin noise and fractal Brownian motion (FBM) are used to produce large-scale cloud structures with enriched texture details, while replicating realistic atmospheric degradation. At the local scale, a diffusion-attenuation algorithm is developed to enable controllable cloud generation in specified regions, ensuring effective occlusion of aircraft targets. By mapping cloud thickness to atmospheric transmittance and incorporating stochastic attenuation, the method achieves fine-grained control of cloud position and extent while maintaining natural texture. The simulated dataset is then used to train aircraft recognition models, improving feature extraction under cloud interference. Experimental results demonstrate that the proposed approach significantly enhances recognition accuracy on cloud-contaminated test sets compared to the baseline model.