BTE-ShapeNet: Background and Target Enhancement with Shape Perception Network for infrared small target detection

Infrared small target detection (IRSTD) presents significant challenges due to low contrast, blurred edges, and shape variability of targets in infrared images, which complicate their separation from the background and severely degrade detection performance. To address these challenges, we present a Background and Target Enhancement with Shape Perception Network(BTE-ShapeNet). Specifically, to tackle the issue of insufficient multi-scale feature perception, we design an enhanced scale sensitivity block(SSB) that strengthens the model’s ability to recognize small targets at different scales through multi-scale convolutional features and an adaptive weighting mechanism. Secondly, to address the issues of background complexity and the emergence of false alarms, we propose a background-target attention blocks (BTABs), BTABs refine the separation between background and target features by employing a dual enhancement mechanism for both target and background, and further integrate background and target features through multiple spatial-channel cross-attention transformer blocks, thereby enhancing background suppression capabilities. Additionally, considering the problems of low contrast and blurred edges, we design a shape perception and detail restoration blocks(SPDR), which combines large convolutions and central difference convolutions to effectively enhance the target edge information while preserving its shape characteristics. Experimental results on the IRSTD-1K, NUAA-SIRST, and NUDT-SIRST datasets demonstrate that BTE-ShapeNet outperforms state-of-the-art methods in detection accuracy, particularly under low signal-to-noise ratios and complex backgrounds, significantly improving detection precision while effectively reducing false alarms and miss detection.