Object identification in computational ghost imaging based on deep learning
- PDF / 3,115,026 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 64 Downloads / 231 Views
Object identification in computational ghost imaging based on deep learning Jianbo Li1 · Mingnan Le1 · Jun Wang1 · Wei Zhang1 · Bin Li1 · Jinye Peng1 Received: 10 March 2020 / Accepted: 8 September 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Processing method plays an important role in accelerating imaging process in ghost imaging. In this study, we propose a processing method with the Hadamard matrix and a deep neural network called ghost imaging hadamard neural network (GIHNN). We focus on how to break through the bottleneck of image reconstruction time, and GIHNN can identify an object before the imaging process. Our research reveals that the light intensity value contains the feature information of the object and expands the possibility of further applications of artificial intellectual techniques in computational ghost imaging.
1 Introduction As a relatively new imaging method, ghost imaging(GI) has been applied to many related fields since Pittman and Y.H. Shi’s first GI experiment in 1995 [1]. Due to its indirect imaging characteristics, it has applications in military and medicine [2–4] etc. In the above application scenarios, researchers need to use ghost imaging method to achieve the image first and then identify the object. Although ghost imaging has advantages over traditional imaging methods, its imaging speed is still not satisfactory. At present, there are various methods for improving imaging speed, most of which focus on improving data acquisition or processing methods, such as the compressed sensing (CS) algorithm [5–8]. Yang et al. [9] achieved object identification with a technique based on CS. However, CS has high computational complexity, and this reconstruction process leads to the inefficiency of identification. In recent years, deep learning method has been used in ghost imaging [10–14], and in this study, we propose a method which benefits from deep learning [15] to classify the features directly before the imaging process. We first use a method based on computational ghost imaging (CGI) [16–18] to extract features of the object by a hadamard matrix [19]. Then we use a deep neural network called ghost imaging Hadamard neural networks (GIHNN) to learn the * Mingnan Le [email protected]; [email protected] 1
School of Information Science and Technology, Northwest University, Xi’an 710127, China
classification features to realize the task of object identification [20] before the imaging process, and the calculation time is reduced significantly. Consequently, our method can be employed in the scenario which requires rapid automatic object identification based on ghost imaging system.
2 Method 2.1 Computational ghost imaging Based on the CGI method, we get the raw data, and the setup required for CGI is shown in Fig. 1. The light generated by the light source (632.8 nm He-ne laser) irradiates the digital micromirror device (DMD, DLP LightCrafter, 0.3 WVGA chipset), and then the light reflected by the DMD illuminates the object. Finally, the
Data Loading...
