ABSTRACT
In view of the slow convergence of the typical denoising network, which is prone to excessive smoothing of images and insufficient recovery of detailed features, this paper proposed a deep denoising network architecture based on the auto-encoder network and the generative adversarial network. That took advantage of the adversarial learning of generators and discriminators, making denoising results sharper and clearer, improving the over-smoothing issue of images, and enhancing the restoration of specific information. Meanwhile, the residual structure was introduced into encoders of the auto-encoder network, which further augmented their denoising capacity and accelerated the convergence of the loss function of the generative adversarial network. In the end, the YOLOv3 network was respectively employed to detect the sonar pulse power spectrum images before and after denoising and analyze the performance of the proposed deep denoising network in the sonar pulse detection. The experimental results manifest that both the power signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) of the denoised image are improved, compared with those of the low signal-noise sonar pulse power spectrum image and that the target detection accuracy is remarkably lifted, which proves the proposed the effectiveness of the deep denoising network in improving the quality of the power spectrum image about the sonar pulse signal, and the significance of denoising process in the sonar pulse detection.
- Lee S, Park B, Kim A. Deep Learning from Shallow Dives: Sonar Image Generation and Training for Underwater Object Detection[J]. 2018.Google Scholar
- Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks[C]//Advances in neural information processing systems. 2012: 1097-1105.Google Scholar
- Ganggang, Dong, Guisheng, A Review of the Autoencoder and Its Variants: A Comparative Perspective from Target Recognition in Synthetic-Aperture Radar Images[J]. IEEE Geoscience and Remote Sensing Magazine, 2018.Google Scholar
- Gong M, Niu X, Zhang P, Generative Adversarial Networks for Change Detection in Multispectral Imagery[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(12):2310-2314.Google ScholarCross Ref
- He K, Zhang X, Ren S, Deep Residual Learning for Image Recognition[J]. IEEE, 2016.Google ScholarCross Ref
- Zhang H, Gong M, Zhang P, Feature-Level Change Detection Using Deep Representation and Feature Change Analysis for Multispectral Imagery[J]. IEEE Geoscience and Remote Sensing Letters, 2016, PP(99):1-5.Google ScholarCross Ref
- Redmon J, Farhadi A. YOLOv3: An Incremental Improvement[J]. arXiv e-prints, 2018.Google Scholar
- Redmon J, Divvala S, Girshick R, You Only Look Once: Unified, Real-Time Object Detection[C]// Computer Vision & Pattern Recognition. IEEE, 2016.Google Scholar
- Redmon J, Farhadi A. YOLO9000: Better, Faster, Stronger[C]// IEEE Conference on Computer Vision & Pattern Recognition. IEEE, 2017: 6517-sss6525.Google Scholar
- Lin T Y, Dollar P, Girshick R, Feature Pyramid Networks for Object Detection[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE Computer Society, 2017.Google Scholar
Recommendations
Lossless-constraint Denoising based Auto-encoders
In this paper, we address the poor generalization ability problem of traditional auto-encoder on noise data, and propose a Lossless-constraint Denoising (LD) method, which can enhance the anti-noise ability and robustness of auto-encoders. We ...
Research on Wavelet Denoising for Pulse Signal Based on Improved Wavelet Thresholding
PCSPA '10: Proceedings of the 2010 First International Conference on Pervasive Computing, Signal Processing and ApplicationsPulse signal is the non-stationary random signal, the signal denoising is an important task before analyzing it. Based on wavelet thresholding denoising method presented by Donoho, a new compromising threshold function is proposed. Compared with ...
Image Denoising Based on the Dyadic Wavelet Transform
ICCIMA '03: Proceedings of the 5th International Conference on Computational Intelligence and Multimedia ApplicationsSince subsampling does not take place in image dyadic wavelet transform at each level, image representation in dyadic wavelet domain compared with wavelet series reconstruction is very redundant and part of disturbance of image dyadic wavelet ...
Comments