Classified-Cascaded Convolutional Neural Networks Enabled Imaging Recovery of Random Length GRIN Imaging Systems

Currently, the combination of the graded-index (GRIN) waveguides and convolutional neural networks (CNNs) has garnered remarkable attention, which already had unprecedented results in many scenes of biomedical endoscopic applications. Although CNNs can enhance image quality and fidelity, they are weak in eliminating diverse intrinsic distortions, acting as critical bottlenecks for in-vivo imaging systems. However, wide-spectrum GRIN endoscopes owns severe intristic distortions owning to its long focusing optical length, so the CNNs continue to pose formidable challenges. To address imaging distortion, we propose the classified-cascaded CNN (CC-CNN), which aims to identify virtual-real deteriorated images and facilitate image recovery. The CC-CNN demonstrates its prowess in imaging distortion correction under a limited set of data, owning to its extensive computational capabilities and innovative learning architectures. Our research introduces the CC-CNN as a promising candidate for endoscopic applications, particularly in response to the growing demand for real-time, in-vivo, invasive imaging transmission.