An Electronic Component Recognition Algorithm Based on Deep Learning with a Faster SqueezeNet

Table 2

Parameters and runtime of the models.

CNN architecture	Data type	Compressed model size (MB)	TPR (FPR = 10e − 6) (%)	AVG inference time (ms)	AVG TensorRT inference time (ms)
PCA SVM	32	0.875	93.29	0.29	—
ResNetV2	32	9.10	96.995	4.65	—
DenseNet	32	2.95	96.995	3.39	—
EfficientNet	32	13.1	97.996	4.72	—
MobileNetV2	32	8.75	99.998	4.83	—
SqueezeNet	32	0.47	96.391	3.53	—
Ours (faster SqueezeNet)	32	0.255	99.999	2.67	0.65