Automobile Component Recognition Based on Deep Learning Network with Coarse-Fine-Grained Feature Fusion

Table 1

Deep learning-based image recognition in industry.

Ref.	Concept	Methodology	One image		Preprocessing	Small dataset	Performance
			Single target classification	Multiobject segmentation
[7]	Surface defect detection	CNN + Unet		√	—	No	—
[12]	Weld defect classification	VGG16 + SegNet-modified	√		Manually annotated	—	—
[13]	Subsurface defects detection	YOLO v5		√	Manually annotated	—	Precise = 92.6%
[14]	Quality monitor of pizza packages	ResNet18	√		Image flip	No	Precise = 99.74%
[15]	Recognition of mechanical material structure	Deep learning network	√		—	Yes	Accuracy = 90.8%
[16]	Vehicle types detection	G-YOLOX		√	Manually annotated	—	Precise = 95.0%
[17]	Damage detection of grotto murals	Ghost-C3SE YOLOv5		√	Manually annotated	Yes	Precise = 56.74%
This paper	Component recognition	PDLN	√		None	Yes	95.11%