The invention discloses an improved YOLOv8n
traffic sign target detection
algorithm, and relates to the technical field of
small target detection. According to the method, the
small target detection precision is remarkably improved, the mAP50 value on a CCTSDB
data set is improved by 5% compared with the original YOLOv8n, and the mAP50: 95 value is improved by 4.4%; on a RoadSign
data set, the mAP50 value is increased by 4.3%, the mAP50: 95 value is increased by 3%, and the detection precision of various traffic signs such as indication, prohibition and warning is improved; according to the invention, the adaptability of complex scenes is enhanced, the DCNv3 module adapts to irregular-shaped targets, the CBAM module focuses key features, and the P2 layer strengthens
small target capture, so that the model can still perform stable detection in complex scenes such as shielding, moving blur, light variation and the like; according to the method provided by the invention, the problem of
class imbalance is relieved, a WIOUv3
loss function effectively solves the problem of non-uniform
data set class distribution, for example, the mAP50 value of a small crowd class such as a parking sign (Stop) in a RoadSign data set is increased from 76.2% to 93.1%.