融入注意力机制的交通标志检测算法OA
Traffic Sign Detection Algorithm with Attention Mechanism Incorporated
针对现有目标检测算法存在对交通标志识别准确率低、检测不完全等问题,提出一种将注意力机制融入 YOLO11n 的交通标志检测算法.首先,将卷积和注意力融合模块与 YOLO11n 主干部分结合,通过融合卷积操作和注意力机制,对图像的全局和局部特征进行有效建模,以提升检测精度;其次,利用全局注意力机制模块融入 YOLO11n 颈部,使模型对特征中的语义信息和位置信息提取更为充分,进而提高了模型的特征表达能力;最后,添加一个小目标检测层,保留更多浅层细节信息,以增强深层和浅层语义信息的融合,从而改善对小目标检测不完全的问题.实验结果证明,改进后的算法在 TT100K 数据集中,精确率、召回率、平均精度分别达 83.9%,70.7%,82.4%,与原模型 YOLO11n 相比较,分别提高了 5.7%,2.7%,6.3%,证明了改进算法的有效性.
In view of the flaws of low recognition accuracy and incomplete detection of traffic signs in existing target detection algorithms,a traffic sign detection algorithm has thus been proposed with an attention mechanism incorporated into YOLO11n.Firstly,the convolution and attention fusion module(CAFM)is integrated with the YOLO11n backbone,so as to effectively model both global and local features of images by combining convolution operations with attention mechanisms to enhance detection accuracy.Secondly,by incorporating the global attention mechanism module into the YOLO11n neck,the model is enabled to extract semantic and positional information from features more fully,thereby improving the feature expression ability of the model.Finally,a small target detection layer is added to retain more shallow detail information to enhance the fusion of deep and shallow semantic information,thus overcoming the incomplete detection of small targets.The experimental results show that the improved algorithm is characterized with a good accuracy,recall,and mean average precision(mAP)of 83.9%,70.7%,and 82.4%,respectively,in the TT100K dataset.Compared with the original model YOLO11n,it improves by 5.7,2.7,and 6.3 percentage points,verifying the effectiveness of the improvement.
彭杰;于惠钧
湖南工业大学 交通与电气工程学院,湖南 株洲 412007湖南工业大学 交通与电气工程学院,湖南 株洲 412007
信息技术与安全科学
交通标志检测YOLO11CAFM注意力机制GAM注意力机制
traffic sign detectionYOLO11CAFM attention mechanismGAM attention mechanism
《湖南工业大学学报》 2026 (3)
63-69,7
评论