Mirco Pattern Gaseous Detector (MPGD) plays a vital role in particle detection at The Heavy Ion Research Facility in Lanzhou and the High-Intensity Heavy Ion Accelerator Facility. The MPGD has amplification structures of a few micron meters. However, the pad size of the readout plane does not match the high granularity due to limitations on the integration level of readout electronics. To address this, using silicon pixel sensors with a pixel size of a few microns to read the MPGD becomes a good candidate, but this produces a vast amount of data from silicon pixels. Therefore, we have developed a Fast Online Multi-Track Locating (FOML) algorithm based on deep learning approaches. This FOML can extract the information of each track in real time and significantly reduces the data volumen. In our network, the lightweight and efficient effect is achieved by implicitly reusing features in the backbone network. In feature fusion, BiFPN and attention mechanisms are used for more comprehensive information transmission and fusion of different levels of features. Finally, learning the target location and category information through different network branches can reduce the computational complexity and improve the detection efficiency. The FOML is expected to achieve a detection speed of 452 frames per second while providing a position accuracy of ~2um.

deep learning; MPGD; Multi-Track; lightweight; efficient