Photoelectric Detection System and Machine Learning Recognition Method of Its Detection Images

Abstract

In the photoelectric detection system, the photoelectric detector can convert the optical signal to be measured into a current signal, and the current amplifier transforms the current signal output by the detector into a voltage signal for amplification. In this study, the photo-multiplier tube (PMT) is selected as the photoelectric detector. Compared with other photoelectric detectors, it can obtain higher internal gain, higher sensitivity, and better response performance. The current amplifier is prepared by pre-amplifier and voltage amplifier. In order to capture photoelectric signals well, a large-format scanning system is set up to design each component module, control module, and host computer module of the system. Besides, a machine learning-based algorithm is proposed, namely semi-supervised manifold image recognition algorithm, which is used for identify photoelectric detection images. In the test process, the printed circuit board (PCB) and sapphire material are firstly used as the substrate of the current amplifier, and their influences in the circuit are compared. The peak value of the output noise of each substrate circuit is around 2.8 mV when the input of the current amplifier is short-circuited. Then, the signal gain and signal bandwidth of the photoelectric detection system remain stable when there is no optical signal input. During the process of changing the system signal gain ratio, the noise output of the system is the lowest when the voltage of PMT is 0.50 V and the current amplifier gain is set to 2.2 × 105 V/A. The proposed recognition algorithm can identify different types of targets well. After the image is projected into a two-dimensional space by the algorithm, the distance between classes increases, and the targets in the class promote aggregation, thereby enhancing the identify-ability between samples.