CSC-Net: A Lightweight Model for Remote Real-Time Monitoring of Infrared Small Targets

Infrared small target detection (IRSTD) is an important subject in many fields such as real-time monitoring and drone applications. Drones typically require real-time processing and transmission of infrared image data during flight. Lightweight models can reduce computational and communication burdens. While existing methods perform target detection effectively, resource-intensive demands hinder remote real-time detection. To provide high-performance target detection and analysis in resource-constrained environments, we propose a cross-sensing compression network (CSC-Net) for IRSTD in this paper. Specifically, the compression mixer module (CMM) enhances the hybrid scheme’s adaptability to different locations by exploring both semantic and spatial dimensions, thus improving image association flexibility. The filter perception module (FPM) dynamically adjusts input contributions for enhanced cross-sensing, emphasizing the benefits of multilayered semantics. Experiments on SIRST and IRSTD-1K show that the proposed CSC-Net can achieve accurate target location and segmentation. In particular, its visual effects and resource consumption can meet the remote real-time monitoring scenario.