LAMOST Fiber Positioning Unit Detection Based on Deep Learning

Abstract

Abstract The double revolving fiber positioning unit (FPU) is one of the key technologies of The Large Sky Area Multi-Object Fiber Spectroscope Telescope (LAMOST). The positioning accuracy of the computer controlled FPU depends on robot accuracy as well as the initial parameters of FPU. These initial parameters may deteriorate with time when FPU is running in non-supervision mode, which would lead to bad fiber position accuracy and further efficiency degradation in the subsequent surveys. In this paper, we present an algorithm based on deep learning to detect the FPU’s initial angle using the front illuminated image of LAMOST focal plane. Preliminary test results show that the detection accuracy of the FPU initial angle is better than 2.°5, which is good enough to distinguish those obvious bad FPUs. Our results are further well verified by direct measurement of fiber position from the back illuminated image and the correlation analysis of the spectral flux in LAMOST survey data.