CSCWD：面向边缘设备轻量微小目标检测的跨尺度通道知识蒸馏
原标题：CSCWD: Cross-Scale Channel-wise Knowledge Distillation for Lightweight Tiny Object Detection on Edge Devices
AI 摘要
该研究提出跨尺度通道知识蒸馏（CSCWD）训练框架，将高分辨率 YOLO11m-P2 教师的空间表征迁移到紧凑的 YOLO11n 学生模型，且不改变学生推理结构。在 Drone-vs-Bird 验证协议下，YOLO11n-CSCWD 达到 50.17% mAP@0.5 和 59.73% 召回率，较基线分别提升 2.92 和 3.55 个百分点；跨尺度对齐比同尺度通道蒸馏再提升 2.09 个百分点。在 DUT-Anti-UAV 零样本评估中 mAP@0.5 从 48.29% 升至 50.06%，并在 Raspberry Pi 5 上以 258 万参数实现 50.32% mAP@0.5、82.32 毫秒延迟（12.15 FPS），运行与内存开销与基线基本一致。
正文节选
CSCWD: Cross-Scale Channel-wise Knowledge Distillation for Lightweight Tiny Object Detection on Edge Devices Abstract Real-time tiny object detection in aerial imagery is constrained by the weak spatial evidence of very small objects and the loss of high-resolution detail in lightweight detectors. This study presents Cross-Scale Channel-wise Knowledge Distillation (CSCWD), a training-time framework that transfers high-resolution spatial representations from a YOLO11m-P2 teacher to a compact YOLO