基于图像识别的弓网接触点检测方法

李凡,杨杰,冯志成,陈智超,付云骁

Pantograph-catenary contact point detection method based on image recognition

Fan LI,Jie YANG,Zhicheng FENG,Zhichao CHEN,Yunxiao FU

表 3 PMSE-BiSeNet和主流模型在弓网数据集中的结果对比

Tab.3 Comparison of results between PMSE-BiSeNet and mainstream models in proposed pantograph-catenary data set

模型 基础网络结构 mIOU/% P/M FLOPs/G FPS/帧 I9-12900(CPU) JETSON TX2 DeepLab v3[19] ResNet50 67.47 41.81 171.09 2.68 — PSPNet[20] ResNet50 80.31 46.71 184.74 2.69 — DenseAspp[21] Densenet121 80.93 9.17 43.09 8.39 — EncNet[29] ResNet50 80.40 33.60 147.08 2.30 — Fcn8s[14] Vgg16 80.16 30.02 320.87 5.85 — BiSeNet v1[22] ResNet18 85.45 12.80 13.03 23.05 6.68 BiSeNet v2[23] — 87.52 3.34 12.35 33.28 10.50 PMSE-BiSeNet — 87.50 4.57 6.73 49.80 12.60