Fig. 4: Intelligent recognition system for ADS based on SRMs.

a PCB board integrating SRMs-based 32 × 32 crossbar array and driver circuits. b Chip carrier with 32 × 32 crossbar array encapsulated on the PCB board. c SEM image of the 32 × 32 crossbar array. d Neuromorphic model architecture of YOLOv5 and YOLOv9 for the BDD dataset. The classification accuracy of e stoplight, f signpost, g pedestrian, h obstacle, i vehicle, and j total aspects based on the four models: YOLOv5, YOLOv9, M-YOLOv5, and M-YOLOv9, where the sample size is 100. The error bars represent the maximum and minimum intervals of the obtained batch results. k The precision, recall, and F1 score obtained by the four models YOLOv5, YOLOv9, M-YOLOv5, and M-YOLOv9 on the BDD dataset. l The total count of five labels—stoplight, signpost, pedestrian, obstacle, and vehicle in the BDD dataset.