Edge Detection For Optical Synthetic Aperture Based On Deep Learning

As an important way to improve the resolution of optical imaging system,optical synthetic aperture has been widely applied to the field of ground and space telescopes.However,the boundaries of segmented aperture systems are much more complex than that of the whole aperture,and have more edge regions for imaging.On the other hand,for large-diameter optics,the steepness of the light at the edge of the mirror and the steep change of the edge phase affect the sharpness of the image reaching the detector,thereby affecting the peak judgment of the phase extraction.These problems have increased the difficulty of co-phase detection.Therefore,in order to reduce the loss of useful measurement data,we need to perform effective edge extraction on noise and high-frequency fringe images.Our main task is to identify the gaps between the sub-apertures and the edges of the projected fringes.In our research,we introduced Convolutional Neural Networks(CNN)and Generative Adversarial Nets(GAN),two models of Deep Learning,into the edge detection of optical synthetic aperture imaging,and analyzed the feasibility and advantage of the model in application.According to the requirement of synthetic aperture imaging edge detection,a large number of aperture fringe data sets are constructed through experiments and simulation.Mat Conv Net,a MATLAB-based toolbox,trained a convolutional neural network with multiple hidden layers and tested its effectiveness.As an input image is given,each intra-neighbor area around the pixel is taken into the network,and scanned pixel by pixel with the trained multi-hidden layers.The network output layer makes a judgment on whether the center of the input block is at the theoretical edge(ie,the sub-aperture boundary and the fringe phase critical).Both a GAN on pytorch was used to improve effect of low-frequency fringe detection,and the performance of our edge detection got better.Compared with the traditional edge detection algorithm or their improved algorithm,Deep Learning can better show the deep characteristics of data and reduce the workload of artificial design.Experiments on some images from testing set verify the effectiveness of the method and obtain a good edge detection result.At the same time,the processing efficiency is improved,which is several orders of magnitude faster than the traditional edge detection algorithm.The work in this paper provides a new approach to optical synthetic aperture imaging field and lays a solid foundation for the subsequent research on cophasing of subaperture.