Table 1: Performance of the benchmark algorithms on the testing set.

Benchmark algorithm	Recall	Precision F1 score	mAP
FRCNN R-FCN	0.7511	0.6186	0.6784	0.6596
FRCNN ResNet101	0.7396	0.5995	0.6623	0.6518
FRCNN Inception-v2-ResNet101	0.7554	0.6046	0.6716	0.6462
YOLOv5	0.7244	0.6081	0.6612	0.6304
EffDet	0.6939	0.6919	0.6929	0.6216

FRCNN Inception-v2-ResNet101 achieved the best recall, EffDet achieved the best precision and F¹ score, and FRCNN R-FCN achieved the highest mAP.

EffDet = EfficientDet; F¹ = harmonic mean of precision and recall; FRCNN = Faster region-based convolutional neural network; mAP = mean average precision;

R-FCN = region-based fully convolutional network; ResNet = residual neural network; YOLOv5 = You Only Look Once version 5.