Study on automatic target identification method in substation 3D scene model

Figure 1.

Figure 2.

Figure 3.

Performance comparison of different algorithms

Model	mAP@0.5/%	mAP@0.5:0.95/%	Vol/MB	FPS/(f/s)
YOLOv5n	84.62	59.08	3.88	139
YOLOv5s	87.85	63.92	14.45	125
YOLOv5m	88.89	66.63	42.54	95
YOLOv5l	89.15	67.25	93.81	80
YOLOv5x	89.64	67.41	175.29	55
YOLOv3	81.82	46.34	246.64	43
YOLOv4	82.06	47.49	256.41	35
YOLOXs	81.38	50.88	36.64	32
SSD	40.54	20.25	98.58	43
Faster RCNN	50.29	23.21	115.37	15
Ours	90.37	68.94	8.72	92

Comparison of the results of the drop sampling experiment

Method	The number of original clouds	The number of clouds in the back point	Time(s)
Random drop sampling	281421	47251	0.1274
Uniform drop sampling	281421	47251	0.1532
Body drop sampling	281421	47251	0.1049

RMSE and time-consuming comparison results of different algorithms

No.	ICP		RANSAC+ICP		Ours
1	RMSE	Time/s	RMSE	Time/s	RMSE	Time/s
2	0.632	0.159	0.385	0.253	0.079	0.275
3	0.683	0.214	0.465	0.276	0.075	0.318
4	0.652	0.208	0.472	0.295	0.077	0.299
5	0.637	0.201	0.516	0.282	0.076	0.273
6	0.601	0.311	0.243	0.674	0.019	0.615
7	0.641	0.374	0.357	0.645	0.008	0.649
8	0.648	0.459	0.551	1.142	0.009	1.106
9	0.639	0.428	0.538	0.938	0.013	0.627
10	0.655	0.417	0.492	0.729	0.007	1.235

Improve the lateral contrast experiment of the module

Model	mAP@0.5/%	mAP@0.5:0.95/%	Model	mAP@0.5/%	mAP@0.5:0.95/%
YOLOv5	49.84	30.49	FPN	49.84	30.49
SE	50.69	32.95	PANet	51.06	32.95
BAM	50.93	33.41	GFPN	51.98	33.16
CBAM	51.42	33.87	PRFPN	52.31	33.83
CAM	52.08	34.62	BIFPN	52.87	35.46
Model	Loss value	mAP@0.5/%	Model	Loss value	mAP@0.5/%
IoU	0.0712	60.52	EIoU	0.0691	64.27
GIoU	0.0694	63.45	SIoU	0.0685	64.41
DIoU	0.0695	62.93	CIoU	0.0623	66.58

Removal experiment from the group point

Method	Parameter	Drop sampling	Removal from the group point	Time(s)
Radius filtering	k1 = 15	47251	47179	0.1042
	k1 = 25	47251	47103	0.1225
	k1 = 90	47251	47192	0.1563
Statistical filtering	r1 = 15, k2 = 10	47251	46935	0.0991
	r1 = 15, k2 = 25	47251	46326	0.1208
	r1 = 15, k2 = 45	47251	45834	0.1517

The experimental results of the model

Group	BiFPN	CAM	CIoU	Vol/MB	FPS/(f/s)	mAP@0.5/%	mAP@0.5:0.95/%
1	×	×	×	14.45	125	87.85	63.92
2	√	×	×	13.98	103	87.94	64.06
3	×	√	×	12.07	101	88.06	64.73
4	×	×	√	14.45	125	87.85	63.92
5	√	√	×	11.39	100	88.74	65.97
6	√	×	√	13.98	98	89.23	66.48
7	×	√	√	10.86	95	89.62	67.81
8	√	√	√	8.72	92	90.37	68.94