牟亮等：基于梯度压缩的YOLO V4算法车型识别 9 h) (c) (d) e (f) 图13同步流样本检测结果对比.(a)YOLO V4在场景1的检测结果：(b)YOLO V4在场景2的检测结果：(c)YOLO v4在场景3的检测结果：(d) YOLO V4GCCD在场景1检测结果：(e)YOLO V4GCCD在场景2检测结果：(f)YOLO V4GCCD在场景3检测结果 Fig.13 Comparison of synchronous flow sample detection:(a)detection results of YOLO v4 in scenario 1:(b)detection results of YOLO v4 in scenario 2;(c)detection results of YOLO v4 in scenario 3;(d)detection results of YOLO v4 GC CD in scenario 1;(e)detection results of YOLO v4 GC CD in scenario 2;(f)detection results of YOLO v4 GC CD in scenario 3 (b) d el (f0 图14阻塞流样本检测结果对比.(a)YOLO v4在场景1的检测结果：(b)YOL0v4在场景2的检测结果：(c)YOLO V4在场景3的检测结果：(d) YOLO V4GCCD在场景1检测结果：(e)YOLO v4GCCD在场景2检测结果：(f)YOLO V4GCCD在场景3检测结果 Fig.14 Comparison of blocked flow sample detection:(a)detection results of YOLO v4 in scenario 1;(b)detection results of YOLO v4 in scenario 2;(c) detection results of YOLO v4 in scenario 3;(d)detection results of YOLO v4 GC CD in scenario 1:(e)detection results of YOLO v4 GC CD in scenario 2;(f)detection results of YOLO V4 GC CD in scenario 3 1.00 1.00 1.00 (b) 0.98 (a) 0.95 (c) 0.95 0.96 0.90 0.85 0.90 080 0.80 0.88 0.65 0.75 0.60 0.86 0.55 0.70 0.8 0.00.10.20.30.40.50.60.70.80.91.0 0.50.0010.20.30.40.50.60.70.80.91.0 0.66.00.1020.30.40.50.60.70.80.91.0 Kecall Recall Recall 图15针对不同车型的P-R曲线.(a)小汽车：(b)公交车：(c)货车 Fig.15 P-R curves for different models:(a)car,(b)bus;(c)truck(a) (b) (c) (d) (e) (f) 图 13 同步流样本检测结果对比. （a）YOLO v4 在场景 1 的检测结果；（b）YOLO v4 在场景 2 的检测结果；（c）YOLO v4 在场景 3 的检测结果；（d） YOLO v4 GC CD 在场景 1 检测结果；（e）YOLO v4 GC CD 在场景 2 检测结果；（f）YOLO v4 GC CD 在场景 3 检测结果 Fig.13 Comparison of synchronous flow sample detection: (a) detection results of YOLO v4 in scenario 1; (b) detection results of YOLO v4 in scenario 2; (c) detection results of YOLO v4 in scenario 3; (d) detection results of YOLO v4 GC CD in scenario 1; (e) detection results of YOLO v4 GC CD in scenario 2; (f) detection results of YOLO v4 GC CD in scenario 3 (a) (b) (c) (d) (e) (f) 图 14 阻塞流样本检测结果对比. （a）YOLO v4 在场景 1 的检测结果；（b）YOLO v4 在场景 2 的检测结果；（c）YOLO v4 在场景 3 的检测结果；（d） YOLO v4 GC CD 在场景 1 检测结果；（e）YOLO v4 GC CD 在场景 2 检测结果；（f）YOLO v4 GC CD 在场景 3 检测结果 Fig.14 Comparison of blocked flow sample detection: (a) detection results of YOLO v4 in scenario 1; (b) detection results of YOLO v4 in scenario 2; (c) detection results of YOLO v4 in scenario 3; (d) detection results of YOLO v4 GC CD in scenario 1; (e) detection results of YOLO v4 GC CD in scenario 2; (f) detection results of YOLO V4 GC CD in scenario 3 1.00 (a) 0.98 0.96 0.94 0.92 0.90 0.88 0.86 0.84 0.0 0.1 0.2 Precision Recall 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Precision Recall (b) 1.00 0.95 0.90 0.85 0.80 0.75 0.70 0.65 0.60 0.55 0.500.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Precision Recall (c) 1.00 0.90 0.85 0.80 0.75 0.70 0.65 0.95 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 YOLO v4 GC CD YOLO v4 YOLO v4 GC CD YOLO v4 YOLO v4 GC CD YOLO v4 图 15 针对不同车型的 P−R 曲线. （a）小汽车；（b）公交车；（c）货车 Fig.15 P−R curves for different models: (a) car; (b) bus; (c) truck 牟 亮等： 基于梯度压缩的 YOLO v4 算法车型识别 · 9 ·