An End-to-End Deep Learning Approach for Plate Recognition in Intelligent Transportation Systems

Table 1

Network configuration summary: “”, “”, and “” stand for kernel size, stride, and padding size, respectively.

Type	Configuration
Input
Conv	#kernels: 64, : , : 1, : 1
Max pooling	Windows: , : 2
Conv	#kernels: 128, : , : 1, : 1
Max pooling	Windows: , : 2
Conv	#kernels: 256, : , : 1, : 1
Conv	#kernels: 256, : 33, : 1, : 1
Max pooling	Windows: , : 2
Conv	#kernels: 512, : , : 1, : 1
Batch normalization
Conv	#kernels: 512, : , : 1, : 1
Batch normalization
Max pooling	Windows: , : 2
Conv	#kernels: 512, , : 1, : 0
Map to sequence
Bidirectional LSTM	#hidden unit: 256
Bidirectional LSTM	#hidden unit: 256
Bidirectional LSTM	#hidden unit: 256
Transcription