高效轻量化的单光子三维成像方法OA

With the advancement of deep learning,single-photon imaging has gradually become an important and challenging research direction,and the introduction of deep learning is helpful for the three-dimensional(3D)reconstruction of single-photon images.Generally,single-photon images are sparse,noise-filled 3D images,with only a few valid signal echoes in their time channels.The existing single-photon imaging reconstruction architectures generally improve their performance by establishing larger backbone networks,which comes at the cost of higher GPU memory usage.Therefore,designing a lightweight yet effective model for deployment on edge devices has become one of the current focuses in related fields.This paper proposes a lightweight 3D image reconstruction architecture that achieves comparable results to other methods while significantly reducing computational requirements.Specifically,the architecture first utilizes a Swin Transformer network to extract temporal features of single-photon images,significantly reducing the dimensions of single-photon images through this time prediction network.Then,a densely cascaded multi-scale network(DCMNet)is employed to extract spatial domain features of single-photon images,ultimately completing the reconstruction of single photon images.This architecture improves the interconnection between decoding layers through a top-down cascade pathway and dense connections to generate high-quality multi-resolution depth outputs.Experimental results demonstrate that the proposed architecture can achieve commendable results while significantly reducing resource consumption.