Small traffic sign detection in real time faces significant challenges, such as complex backgrounds that may confuse and hide signs. The weather conditions can further reduce visibility and affect the precision of detection. Furthermore, false detection may occur due to certain traffic sign’s similar shapes and colors, particularly when signs are small. To overcome the listed issues, advanced methods with reliable feature extraction and effective processing are needed to locate and classify signs with high precision in real time. Deep learning enables the extraction of high level features using mechanisms such as multiscale attentions and pyramids based multiscale features. In this paper, we suggest a new method with limited computational power. We update You Only Look Once YOLOv5 architecture by merging a spatial attention SA. A spatial attention is merged into the backbone of the model to focus on critical areas within images. The study outcome on the GTSDB dataset shows that the mAP50 increased from 90.2% to 93% which indicates that spatial attention has a positive impact model performance.