Video super-resolution method for spacecraft approaching asteroids
-
摘要: 抵近探测的成像过程中,动态图像序列往往因为平台运动、抖动等,存在图像模糊、分辨率不够高等问题。本文针对抵近探测过程中图像序列的超分辨率开展研究,提出了一种基于BasicVSR++的视频超分辨方法。通过引入空间和通道注意力机制强化模型对细节特征的提取能力,同时结合共享投射权重、多组机制和采样点调制,提升了对齐模块的效果,在提升网络特征提取能力同时,弥补了正则卷积在长距离依赖性和自适应空间聚集方面的不足,同时将下采样与低通滤波器结合来减少图像高频成分,提高了模型对轻微图像抖动的鲁棒性。此外引入新的上采样模块来结合局部和全局特征,生成自适应上采样核扩展感受野,更好的对全局结构恢复和细节重建。仿真实验结果表明,本文提出的方法在峰值信噪比(PSNR)和结构相似性(SSIM)指标上,分别较原始方法提高了2.2%和2.1%。Abstract: In the imaging process of approach detection, dynamic image sequences often have problems such as image blur and insufficient resolution due to platform movement and jitter. This paper studies the super-resolution of image sequences in the process of approach detection and proposes a video super-resolution method based on BasicVSR++. By introducing spatial and channel attention mechanisms to enhance the model's ability to extract detail features, combined with shared projection weights, multi-group mechanisms and sampling point modulation, the effect of the alignment module is improved. While improving the network feature extraction capability, it makes up for the shortcomings of regular convolution in long-distance dependency and adaptive spatial aggregation. At the same time, downsampling is combined with a low-pass filter to reduce the high-frequency components of the image, which improves the robustness of the model to slight image jitter. In addition, a new upsampling module is introduced to combine local and global features, generate an adaptive upsampling kernel to expand the receptive field, and better restore the global structure and reconstruct details. The simulation experimental results show that the proposed method improves the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) indicators by 2.2% and 2.1% respectively compared with the original method.
-
-
计量
- 文章访问数: 44
- HTML全文浏览量: 6
- PDF下载量: 2
-
被引次数:
0(来源:Crossref)
0(来源:其他)