Multidimensional encoding of microspheres, as a novel signal output element, has demonstrated excellent detection performance; however, conventional microscopic counting methods for encoded microspheres are still limited by narrow fields of view, limited sampling coverage, and missed detection. To address these challenges, we developed an artificial intelligence-assisted microscopic video-stream digital immunoassay system. In this system, the size and color of microspheres are used for channel identification, while the counted number of microspheres serves as the digital signal for target quantification. By integrating continuous scanning imaging, object detection, size and color decoding, multi-object tracking, and trajectory-level duplicate-removal counting, the system successfully transforms fixed-field microscopic imaging into a continuous video-stream digital readout mode, significantly improving detection accuracy and sampling coverage. As a proof of concept, we applied this system to the simultaneous detection of enrofloxacin, kanamycin, and chloramphenicol, achieving a detection limit at the picogram-per-milliliter level. The results obtained from spiked milk samples were in high agreement with those of enzyme-linked immunosorbent assay. This study provides a low-cost, automated, and scalable digital readout system for multiplexed small-molecule detection.