Multidimensional Information Sensing Method Of Wireless Endoscopic Robot

Wireless endoscope robot has shown great potential in automatic/semi-automatic digestive tract lesion screening,in-vivo resident monitoring,targeted drug delivery,in-vivo biopsy and other future diagnostic and therapeutic applications.The performance of endoscope robot depends on its multi-dimensional information perception ability.This paper tries to build a unified multi-dimensional information perception framework for the new wireless endoscope robot and studies the common perception problems and methods under this framework.Due to the characteristics of endoscopic scene composite imaging,the work involves multi-spectral image,video image,depth image and other multi-dimensional information reconstruction,as well as the intelligent perception method of target tracking and recognition.Aiming at the characteristics of composite image acquisition of wireless endoscope robot,a framework of synchronous active illumination compression imaging system was built.Based on the unified framework,cooperative measurement schemes for multi-spectral compression imaging and time-domain compression imaging are presented.The multi-channel illumination,the compression coding device and the sensor cooperate to complete the compression measurement under the synchronous clock.Based on the idea of high signal-to-noise ratio signal assisted an estimation from the noised measurement,a multidimensional information perception framework based on edge information guidance is proposed.Based on the framework,a low-rank tensor reconstruction model is constructed for multi-dimensional information compression perception,and a robust iterative algorithm is proposed.Aiming at single-exposure imaging of multi-spectral coding aperture,a low-rank reconstruction algorithm based on edge information was proposed to effectively improve the reconstruction quality.In view of the characteristics of high dynamic intensity of endoscope scene,a compression sensing method based on adaptive aperture coding are proposed,which take an adaptive gray coding based on the predictions of the saturation region.Experimental results show that this method effectively reduces the probability of over-saturation distortion and improves the reconstruction accuracy of dynamic image blocks.Based on the proposed edge information guidance estimation framework,aiming at the problem of monocular multi-view depth estimation,a recursive variable parameter cost generation method is proposed on the end-to-end depth convolution network framework,which effectively improves the accuracy and stability of monocular dense depth image reconstruction.A wireless multipath fingerprint model is proposed to solve the problem that abnormal signals affect the positioning accuracy of wireless fingerprint location.The model projects the wireless location fingerprint feature descriptor into the regenerative kernel Hilbert space,and introduces the Euclidean distance between the nearest points of affine packet to measure the similarity of location.Based on its augmented Lagrangian function,an alternating iterative solution method for the model is given.This method improves the ability of the positioning system to suppress abnormal signals.Aiming at the characteristics of endoscopic online target tracking and relocation,a new rotation-invariant class-like Harbin statistical descriptor and a simplified statistical random forest discriminator based on confidence statistics are designed.Based on the framework of twin-deep neural network and manual descriptor compound,an online tracking and relocation method based on location fine fusion is proposed.Clinical endoscopic video testing has shown that its performance of tracking and relocation is superior to state-of-the-art twin-deep neural network.