| Title |
Research of Low Resolution Infrared Image Deep Convolution Neural Network with Meta Data Fusion for Embedded System |
| Authors |
홍용희(Yong-hee Hong) ; 진상훈(Sang-hun Jin) ; 구영모(Young-mo Koo) ; 지호진(Ho-jin Jhee) |
| DOI |
https://doi.org/10.5573/ieie.2023.60.7.13 |
| Keywords |
Meta data; VGG; Low resolution; Infrared; Synthesize image |
| Abstract |
In this paper, the low resolution infrared image convolution classifier accuracy was improved by meta data which made up of relative distance and angle. The meta data is fused with reinforced VGG style network and Global Average Pooling layer structure. Then accuracy is improved with maintaining computational complexity. The meta data is fused before 1x1 convolution layer which located in the reinforced VGG style network. And it makes output after 1x1 convolution and Global Average Pooling layer. The fused network which use indirect meta data input with full connection network shows 0.51% improvement and 96.36% classifier accuracy. The fused network which use direct meta data input with changing active function into Leaky ReLU shows 0.49% improvement and 96.34% classifier accuracy. The classifier accuracy can be improved by adding meta data for low resolution infrared image convolution network. |