A Computational Transmission Imaging Based on the SegNet Neural Network

To ensure image quality and reduce the bit error rate, conventional computational transmission imaging requires super-sampling, which leads to lengthy sampling times. By fully combining the advantages of ghost imaging and neural networks, this paper proposes a computational transmission imaging method based on the SegNet neural network. In this method, SegNet is used to construct an encoding network, where the original image is combined with a pre-set carrier image in the encoding network through cascaded convolution to achieve image encoding. A CNN is used to construct a decoding network to reconstruct the original image. During experiments, the ImageNet data set was used to train the neural network, and three evaluation metrics-BER (bit error rate), PSNR (peak signal-to-noise ratio), and CC (correlation coefficient)-were utilized for performance analysis. Compared with the state-of-the-art methods, it can be seen that the proposed method can effectively reduce the BER of information transmission to 0.1503, has strong anti-noise interference ability. Even when the transmission distance reaches 1.5*10^3, the CC value of the imaging effect of the proposed scheme can still be guaranteed to be above 0.9, and accurate information reconstruction can be achieved. A series of experimental results demonstrate that our method holds significant theoretical value and research potential.