Research Of Target Reconstruction Algorithm For Computational Ghost Imaging

Computational ghost imaging is a new computational imaging technique based on optical system modulation and signal processing,which uses optical modulation and computational inversion to solve the nonlinear relationship between scene and image.This new imaging technique can not only take advantage of the traditional optical imaging,but also simplify the imaging system,and utilize information such as spectrum,space and time resolution,so it can be widely used in military detection,medical imaging and other fields.However,this new imaging method is limited by its imaging quality and imaging time,and it is difficult to complete fast and high-quality reconstruction,which restricts the its further promotion and application.In this paper,the reconstruction algorithm and object classification method in computational ghost imaging system are systematically and deeply researched.By analyzing the related techniques and their shortcomings,we propose a novel computational ghost imaging methods,which are sought to achieve the requirements of obtaining high quality images while reducing the time cost.The main work of the research in this paper is as follows:(1)A compressed computational ghost imaging method based on region segmentation is proposed to solve the imaging quality problem of local tiny regions in reconstructed images.Compared with other traditional methods,the proposed method not only significantly reduces the number of samples and the spatial intensity calculation of the target region,but also improves the image quality of local tiny regions.The experimental results show that when sampling 3000 times,the peak signal-to-noise ratio(PSNR)of the proposed method is improved by more than 9 d B compared with the traditional computational ghost imaging method,and also increased by about 49.57%compared with the sampling 500 times.(2)A computational ghost imaging method based on convolutional neural network(CNN)is proposed to solve the imaging quality and the speed of reconstructed images under low sampling condition.The proposed method is further improved and optimized as an image recovery model based on denoising convolutional neural networks,which can fully exploit the internal features of images.The experimental results show that the proposed method can reconstruct the measured object with high quality when the sampling rate is 0.08 and the imaging quality is higher than other methods.Meanwhile,our method takes about 0.06 s without sacrificing the image quality when the single image reconstruction,which greatly improves the speed of image reconstruction.(3)A computational ghost imaging object classification based on mini VGG is proposed to solve the ghost imaging object classification under low sampling condition.The proposed method is capable of fast automatic recognition and classification of object information that cannot be distinguished by the human eye when high-quality reconstructed images are not available.The experimental results show that the method can obtain 98.41% classification accuracy when the sampling rate is 0.1,and the required time to execute the program for single image classification is about 0.064 ms,which is better than other methods in terms of time and accuracy.Moreover it can achieve the object classification task quickly.In summary,the imaging system combined region segmentation and convolutional neural network methods with computational ghost imaging has the characteristics of high imaging quality and fast imaging speed,which provides a new solution to the bottleneck problem of computational ghost imaging technology in imaging quality and imaging speed at this stage.