OIRD: a multi-scenario dataset for intelligent laser interference image restoration

With the widespread proliferation of Unmanned Aerial Vehicles (UAVs) in military and security domains, electro-optical detection systems face growing threats from laser interference. The resulting laser spots and halos significantly degrade image quality, thus reducing object detection accuracy. Current research on deep learning-based restoration methods faces two critical bottlenecks: the lack of large-scale dedicated laser interference datasets and poor model generalization Trained on general-purpose datasets. This study constructs the Optical Interference Restoration Dataset (OIRD), the first large-scale dataset specifically for laser interference image restoration, utilizing a methodology of "multisource real data fusion integrated with physical simulation". Comprising airborne/ground-based laser interference and glare subsets, OIRD integrates a real spot database, physical laws, and multi-instance superposition to ensure the authenticity of the simulated interference. Following automated screening, manual validation, and augmentation procedures, the dataset yields 483,649 high-quality annotated pairs with bidirectional scene coverage. Experimental validation using the and augmentation model demonstrates the dataset's effectiveness, achieving a test-set PSNR of 41.580 dB and SSIM of 0.990. Typical restoration results successfully remove laser spots and recover obscured details. This work provides critical data support for laser interference image restoration research. Future work plans include expanding into dynamic scenes and exploring multimodal fusion techniques.