Automatic DNN architecture design using CPSOTJUTT for power system inspection

Journal of Big Data

Table 12 Test results for different object detection model

Model	PSIID		PSSID		PLOID		#Parameters
Model	mAP	mAR	mAP	mAR	mAP	mAR	#Parameters
YOLOv5	0.554	0.590	0.574	0.607	0.565	0.593	42.6 M
Faster R-CNN	0.656	0.702	0.633	0.658	0.621	0.672	40.23 M
Faster-FPN	0.710	0.803	0.717	0.765	0.735	0.779	36.25 M
Cascade R-CNN	0.793	0.868	0.802	0.832	0.837	0.871	256 M
CPGA-DNN	0.763	0.833	–	–	–	–	20.75 M
CPSOTJUTT-EM	0.847	0.897	–	–	–	–	132.6 M
CPGA-DNN	–	–	0.742	0.802	–	–	21.62 M
CPSOTJUTT-EM	–	–	0.843	0.879	–	–	139.7 M
CPGA-DNN	–	–	–	–	0.827	0.856	20.67 M
CPSOTJUTT-EM	–	–	–	–	0.892	0.917	129.5 M