부산대학교 도서관

In monitoring scenarios for material processing and accident status evaluation, high-speed X-ray imaging at the microsecond or even nanosecond level is utilized. Real-time, long-term high-speed X-ray image processing is difficult to realize with existing technology; main limitations are detector system structure and the need for high-speed transfer and storage of a large number of digital signals. Therefore, this article proposes an irradiated photoelectric neural network for high-speed real-time X-ray image recognition based on a pixelated radiation detector. We use a silicon photomultiplier (SiPM)-coupled scintillator to realize this pixelated radiation detector. The weights of a neural network are mapped to the bias voltage of SiPM pixels, so that the SiPM array itself constitutes a neural network. By analyzing the output signals of an array of SiPMs, an X-ray image can be recognized in real-time. The whole process does not involve complicated digital circuits and digital signal processing. The feasibility of the method was verified by experiments. As the initial stage of the study, we achieved an X-ray image recognition speed of 500000 frames/s. The simplicity and innovation of this detector fully demonstrate its future application prospects. It is especially suitable for applications that require long-term continuous monitoring of radiographic images and immediate feedback of results, such as online monitoring of high-speed material processing, observation of the cosmic environment, and other continuously changing scenes of X-ray images.