Conditional TransGAN-Based Data Augmentation for PCB Electronic Component Inspection

Table 2

Experimental results (mAP in percentage) on the self-developed dataset. The highest scores for each column are marked in bold.

Models	Settings	Capacitor	Diode	Optocoupler	mAP
Faster R-CNN ResNet101	w/o augmentation	90.1	91.8	92.6	91.5
	W/IP aug	90.4	92.5	93.3	92.1
	w/TransGAN	90.7	93.2	93.0	92.3
	w/cGAN	91.2	92.9	93.1	92.4
	w/cTransGAN	93.0	93.9	93.5	93.4
YOLOV3 DarkNet	w/o augmentation	92.6	93.1	94.2	93.3
	w/IP aug	93.3	93.1	94.8	93.7
	w/TransGAN	93.1	93.2	95.6	94.0
	w/CGAN	94.0	94.5	95.2	94.6
	w/cTransGAN	94.9	95.1	95.8	95.3
SCNet ResNet101	w/o augmentation	93.8	94.1	94.5	94.1
	w/IP aug	94.5	94.7	96.3	95.2
	w/TransGAN	94.7	95.3	95.9	95.3
	w/cGAN	94.1	95.4	96.7	95.4
	w/cTransGAN	95.2	95.5	97.9	96.2

Table 2 gives the metrics for the best models trained by the Faster R-CNN ResNet101, YOLOV3 DarkNet, and SCNet ResNet101 models using different enhancements on the two dataset tasks, respectively, where the metrics include the AP value and the mean AP value (mAP) for each subtarget. The range of AP values is from 0% to 100%, with higher values demonstrating better detection of the target. As can be seen in Tables 2, the detection results using cTransGAN are almost always optimal.