DTF-YOLO: A Thermal-Texture Decoupling and Recalibration Detector for Infrared-Visible Small Object Detection

All-weather, high-precision perception of airport airside targets such as distant personnel, ground vehicles, and small hazards is essential for safety control. Visible-light monitoring degrades under low illumination, dust, and fog, whereas infrared sensing captures heat-emitting targets but lacks scene texture; infrared–visible fusion has therefore become a mainstream route to robust detection. In fused images, however, thermal saliency and visible texture remain deeply entangled, modality interference varies across scenes and feature levels, and strided downsampling causes irreversible loss of small-target information. This paper proposes DTF-YOLO, a thermal-texture decoupling and recalibration detector built on a restructured YOLOv11 backbone. A Thermal-Texture Decoupling Recalibration (TTDR) module uses parallel thermal-radiation and texture-detail perception paths to recalibrate fused features at the channel and spatial level; a fully information-preserving Bi-directional Focus module replaces conventional downsampling with spatial-channel slice rearrangement; and a four-scale P2–P5 fusion neck improves sensitivity to tiny targets. Evaluated on the public M3FD and LLVIP datasets as proxy benchmarks, DTF-YOLO achieves 87.32% mAP@0.5 and 80.33% recall with 2.91 M parameters on M3FD, and 95.26% mAP@0.5 on LLVIP, demonstrating a favorable accuracy–efficiency trade-off and potential for airport airside safety monitoring.