Application of Data-Driven Iterative Learning Algorithm in Transmission Line Defect Detection

Table 2

The precision of algorithm test.

Algorithm	Classes	Recall	Precision	AP	mAP	Training time (h)
Initial training data + Faster RCNN	lsqxz	0.751	0.684	0.65	0.596	6.11
	lmxs	0.643	0.685	0.541
Initial training data + feature pyramid + deformable convolution + Faster RCNN	lsqxz	0.78	0.689	0.68	0.625	8.05
	lmxs	0.668	0.702	0.569
The first iterative learning + sample mining + feature pyramid + deformable convolution + Faster RCNN	lsqxz	0.892	0.794	0.803	0.704	39.05
	lmxs	0.696	0.787	0.604
The second iterative learning + sample mining + feature pyramid + deformable convolution + Faster RCNN	lsqxz	0.927	0.8	0.854	0.785	118.80
	lmxs	0.752	0.814	0.715