In the intricate and dynamic infrared imaging environment, the detection of infrared dim and small targets becomes notably challenging due to their feeble radiation intensity, intricate background noise, and high interference characteristics. To tackle this issue, this paper introduces a lightweight detection and recognition algorithm, named YOLOv5-IR, and further presents an even more lightweight version, YOLOv5-IRL. Firstly, a lightweight network structure incorporating spatial and channel attention mechanisms is proposed. Secondly, a detection head equipped with an attention mechanism is designed to intensify focus on small target information. Lastly, an adaptive weighted loss function is devised to improve detection performance for low-quality samples. Building upon these advancements, the network size can be further compressed to create the more lightweight YOLOv5-IRL version, which is better suited for deployment on resource-constrained mobile platforms. Experimental results on infrared dim and small target detection datasets with complex backgrounds indicate that, compared to the baseline model YOLOv5, the proposed YOLOv5-IR and YOLOv5-IRL detection algorithms reduce model parameter counts by 42.9% and 45.6%, shorten detection time by 13.6% and 16.9%, and enhance mAP0.5 by 2.4% and 1.8%, respectively. These findings demonstrate that the proposed algorithms effectively elevate detection efficiency, meeting future demands for infrared dim and small target detection.

With the development of infrared remote sensing technology and the deployment of satellite constellations, infrared video from orbital platforms is playing an increasingly important role in airborne target surveillance. However, due to the limitations of remote sensing imaging, the aerial targets in such videos are often small in scale, low in contrast, and slow in movement, making them difficult to detect in complex backgrounds. In this paper, we propose a novel detection network that integrates inter-frame residual guidance with spatio-temporal feature enhancement to address the challenge of small object detection in infrared satellite video. This method first extracts residual features to highlight motion-sensitive regions, then uses a dual-branch structure to encode spatial semantics and temporal evolution, and then fuses them deeply through a multi-scale feature enhancement module. Extensive experiments show that this method outperforms mainstream methods in terms on various infrared small target video datasets, and has good robustness under low-signal-to-noise-ratio conditions.