1 January 2026 in IEEE Geoscience and Remote Sensing Letters
2 June 2026 in Expert Systems with Applications
20 July 2026 in Measurement Science and Technology
To address the challenges of structural ambiguity in small-scale targets and uneven thermal-feature distribution in unmanned aerial vehicle (UAV) infrared object detection, this paper proposes a detection network, DSR-SCNet, that integrates spatial enhancement and channel calibration. The network enhances the representation of infrared object features by incorporating two key modules. First, a depthwise spatial residual (DSR) Block is proposed to improve local structural representation by progressively aggregating spatial context via cascaded depthwise convolutions, enabling more robust modeling of weak and blurred infrared targets under low computational cost. Second, a spatial-channel feature calibration module is introduced, consisting of multi-scale axial spatial attention and self-attentive channel aggregation, to jointly perform adaptive feature calibration from both spatial and channel perspectives. Specifically, in the spatial dimension, a multi-scale axial modeling strategy is employed to enhance salient responses along the height and width directions. In the channel dimension, a self-attention-based channel aggregation mechanism is adopted to emphasize semantic features while suppressing background interference. Built upon these modules, DSR-SCNet achieves mAP50 scores of 88.3%, 66.2%, and 93.4% on the HIT-UAV, VEDAI, and NUAA-SIRST datasets, respectively, demonstrating the effectiveness of the proposed method for small-scale UAV infrared object detection.
16 July 2026 in Infrared Physics & Technology