ASPC-YOLOv8 detection algorithm for miniature traffic signs
To address the misdetection and omission of miniature traffic signs under partial occlusion and complex background, this paper proposes a traffic sign detection framework based on YOLOv8s.First, the adaptive spatial pyramid convolution module ( ASPC) is built to replace the Neck part of the Conv module.A new ASPC2f module is designed to replace the part of the C2f module, reducing the number of model parameters and improving the detection performance.To mitigate the semantic information loss due to the inconsistency of scale in detecting the miniature targets, a miniature target detection layer is introduced to enhance the effective fusion of deep and shallow semantic information.Finally, the EIOU is employed to replace original bounding box loss function to improve the network bounding box regression performance.Our experimental results show the method achieves 89.7% mean average precision (mAP) on the TT100K traffic sign dataset, 6.2 percentage points higher compared to that of the original model.Meanwhile, it is 9 .4 percentage points higher in the mean average precision of the microtargets and reduces the number of parameters by 2.6 MB.