模型

Loss

AP

RECALL

识别正确率

误识别率

漏识别率

FasterRCNN

0.003

0.761

0.694

71.3%

19.1%

28.7%

YOLOV5

0.036

0.350

0.400

65.2%

17.1%

34.8%

UNET

0.009

74.2%

19.7%

25.8%

MaskRCNN

0.071

0.771

0.656

72.3%

19.1%

27.7%

SE + MaskRCNN

0.074

0.746

0.712

73.9%

19.4%

26.1%

Cbam + MaskRCNN

0.007

0.778

0.696

80.3%

17.8%

19.7%