Lightweight Adaptive Feature Sampling-based High Dynamic Range Infrared Image Detail Extraction Network
Mingle Ma, Xinke Yang, Huanan Li, Rui Fan, Chaoxiong Dai, Peng Wang · 2025
Accurate detail extraction from infrared images is fundamental for subsequent analysis tasks such as target detection, fault diagnosis, and thermal anomaly identification, as these details often contain critical information about object boundaries, thermal variations, and potential defects. However, extracting fine details from such images remains challenging due to their inherent characteristics of low contrast, high noise, and complex parameter tuning requirements in traditional processing methods. This paper proposes a novel high dynamic range infrared image detail extraction method addressing the difficulties of traditional infrared image processing in detail extraction and parameter adjustment. Based on grouped deformable convolution and depth-wise separable convolution, we propose a lightweight adaptive feature sampling module (LAFSM) that can effectively achieve precise detail and basic component separation of high-bit infrared images. Compared with traditional algorithms, our method requires no manual parameter adjustment and significantly reduces parameter count. Experimental results demonstrate that this method performs excellently in high-bit infrared image detail extraction, with significant advantages in edge preservation and noise suppression, providing a new technical approach for high dynamic range infrared image preprocessing.