Journal of Sensors

Research Article

A Robust Residual Shrinkage Balanced Network for Image Recognition from Japanese Historical Documents

Table 1

The comparison results with other deep learning models in the Kuzushiji-MNIST dataset.
Deep learning modelsBalanced accuracyNumber of parameters
Proposed (8901/10,000)12,573,130
Proposed-basic (8800/10,000)12,573,130
ResNet-18 (8733/10,000)11,192,817
Wide ResNet-18 (8834/10,000)55,913,732
Stochasticdepth-18 (8281/10,000)11,223,780
MobileNet (8130/10,000)3,314,852
MobileNet-v2 (8391/10,000)2,369,316
ShuffleNet (8714/10,000)1011676
ShuffleNet-v2 (8370/10,000)1,360,464
ACNN (8573/10,000)71,470,602
DCNN (8463/10,000)22,070,634
CNN (8426/10,000)22,070,634