Contrastive self-supervised representation learning framework for metal surface defect detection

Journal of Big Data

Table 5 Comparison of proposed model with the state-of-art deep learning models using the NEU-CLS dataset

Model	Architectures	Trainable parameters	Accuracy (%)	F1-score
Qayyum et al. [63]	InceptionV3	24 M	92.96	0.93
Konovalenko et al. [64]	ResNet152	60 M	86.29	0.86
Zeeshan et al. [65]	VGG19	138 M	61.85	0.55
Singh et al. [66]	ResNet101-SVM	44.5 M	70.23	0.70
Liu et. al. [28]	CNN-LSTM	1.2 M	67.5	0.68
Lee et. al [67]	AnoViT	13.97 M	93.0	0.93
Proposed	SimCLR-Skip-ConvNet	0.9 M	97.41	0.97