基于结构重参数化的遥感影像超分轻量化重建方法

Lightweight Super Resolution Reconstruction Method of Remote Sensing Image Based on Structural Reparameterization

遥感影像成像过程中各种因素导致获取的影像分辨率较低进而难以达到预期观测效果,需要借助超分辨率重建技术实现质量增强。然而,大多数遥感影像超分辨率重建算法集中于提升超分网络模型的性能,忽略推理速度对超分辨率重建算法同样重要。文章设计了一种基于结构重参数化的遥感影像超分轻量化重建方法,在推理时通过参数等价转换减少模型参数和浮点运算数,从而实现更快的推理速度。采用AID与NWPU-RESISC45遥感数据集进行实验,根据典型评估指标峰值信噪比和结构相似性,将本文提出的ECBASR方法与经典的超分重建方法进行对比。实验结果表明,ECBASR取得了良好的重建性能,大幅减少了运行占用内存,加快了推理速度。

Due to various factors in the imaging process,the resolution of the remote sensing image obtained is low and it is difficult to achieve the desired observation effect.Quality enhancement needs to be achieved with the help of super resolution reconstruction technology.In view of the fact that most of the remote sensing image super resolution reconstruction algorithms focus on improving the performance of the super separation network model,ignoring the reasoning speed is also important to the super resolution reconstruction algorithm,a remote sensing image super resolution lightweight reconstruction method based on structural re-parameterization is designed.In inferencing,the model parameters and floating-point operands are reduced by parameter equivalent transformation,so as to achieve faster inferencing speed.Experiments are carried out on AID and NWPU-RESISC45 remote sensing data sets,and the ECBASR method proposed in this paper is compared with the classical super resolution reconstruction method according to the peak signal-to-noise ratio and structural similarity of typical evaluation indicators.The experimental results show that ECBASR achieves good reconstruction performance,greatly reduces the memory occupied by operation,and speeds up the speed of inferencing.