Non-Concentric Differential Model with Geographic Information-Driven Weights Allocation for Enhanced Infrared Small Target Detection

Infrared small target detection technology has received extensive attention due to its advantages in long-distance monitoring. However, there is much room for improvement in its performance due to complex backgrounds and the lack of distinct features in small targets. Many specific scenarios can lead to target loss, such as edge-adjacent targets, intersecting targets, low contrast caused by locally bright backgrounds, and false alarms induced by globally bright backgrounds. To address these issues, we have identified the positional correlation differences between the local background location and whether the target can be perceived by the human eye, thereby introducing geographic information weights to represent this correlation difference. We first constructed a non-concentric Gaussian difference structure to prevent the central target energy loss caused by traditional concentric filters. Based on this, we introduced Gabor filters, which have the capability of directional feature extraction and position correlation representation, into the non-concentric differential structure. By adjusting the relative position of the Gabor filter center and configuring frequency parameters based on geographic information, we optimized the filter weights to handle complex situations, such as targets being close to background clutter or other targets. Subsequently, an improved logarithmic function was applied to adjust the overall saliency of candidate targets, preventing the loss of low-contrast targets and the residual high-energy background clutter. Extensive experiments show that our method exhibits effective detection performance and robustness in four application scenes and three challenging image distribution scenes.