【目的】为改进车道线检测算法的精度,优化自动驾驶系统中的车道保持,提出了一种新的车道线检测算法——基于空洞金字塔池化的通道和空间注意力机制的车道线检测算法(channel and position attention mechanism with atrous spatial pyr...【目的】为改进车道线检测算法的精度,优化自动驾驶系统中的车道保持,提出了一种新的车道线检测算法——基于空洞金字塔池化的通道和空间注意力机制的车道线检测算法(channel and position attention mechanism with atrous spatial pyramid pooling for ultra fast structure-aware deep lane detection,CPAM-UFLD)。【方法】首先,在超快速结构感知深度车道检测(ultra fast structure-aware deep lane detection,UFLD)算法的基础上融入了空洞金字塔池化模块(atrous spatial pyramid pooling,ASPP),以有效地捕捉车道线图像中不同尺度的特征;使用了通道和空间注意力机制(channel and position attention mechanism,CPAM)以关注图像中的关键区域;同时,为平衡类别权重和提高定位精度,使用了包括加权交叉熵损失函数等的四种损失函数;其次,提出了一种亮度改善模块,该模块旨在提升输入图像的质量,从而增强车道线的识别度。【结果】本算法在TuSimple数据集上的检测精度由原来的95.86%提升至96.56%;同时,在CULane数据集上,检测精度由原来的72.2%提升至73.7%。【结论】通过算法的改进,可以有效提高车道线检测的精度,这为智能网联汽车自动驾驶系统的环境感知提供了理论参考。展开更多
安全帽在保护施工人员免受事故伤害方面发挥着至关重要的作用。然而,由于种种原因,工人们并没有严格执行戴安全帽的规定。为了检测工人是否佩戴安全帽,本文提出了基于YOLOv8改进的目标检测算法(CPAM-P2-YOLOv8)。在YOLOv8的颈部添加CPA...安全帽在保护施工人员免受事故伤害方面发挥着至关重要的作用。然而,由于种种原因,工人们并没有严格执行戴安全帽的规定。为了检测工人是否佩戴安全帽,本文提出了基于YOLOv8改进的目标检测算法(CPAM-P2-YOLOv8)。在YOLOv8的颈部添加CPAM结构,增强网络对图片的特征提取,在YOLOv8的头部引入处理后的小目标检测层P2。CPAM-P2-YOLOv8提高了目标检测的精确度,实验结果表明,改进模型的精确度达到了91%。与YOLOv8对比,CPAM-P2-YOLOv8的mAP50提高了1.0%,参数量减少了17%,同时通过对比发现,CPAM-P2-YOLOv8比YOLOv8在检测小目标方面更有优势。与YOLOv10对比,CPAM-P2-YOLOv8的mAP50提高1.9%。使用知识蒸馏方法,使CPAM-P2-YOLOv8的精确度进一步提升,达到91.4%。Safety helmets play a crucial role in protecting construction workers from accidents and injuries. However, due to various reasons, workers did not strictly adhere to the rule of wearing safety helmets. To detect whether workers are wearing helmets, this article proposes an improved object detection algorithm based on YOLOv8 (CPAM-P2-YOLOv8). Add CPAM structure to the neck network of YOLOv8 to enhance the feature extraction of safety helmets, and introduce a processed small object detection layer P2 at the head of YOLOv8. CPAM-P2-YOLOv8 improved the accuracy of object detection, and experimental results showed that the improved model achieved an accuracy of 91%. Compared with the YOLOv8 model, CPAM-P2-YOLOv8 improved mAP50 by 1.0% and reduced parameter count by 17%. Through comparison, it was found that CPAM-P2-YOLOv8 has more advantages in detecting small targets than YOLOv8. Compared with YOLOv10, the mAP50 of CPAM-P2-YOLOv8 increased by 1.9%. By using the knowledge distillation, the precision of CPAM-P2-YOLOv8 was further improved to 91.4%.展开更多
文摘【目的】为改进车道线检测算法的精度,优化自动驾驶系统中的车道保持,提出了一种新的车道线检测算法——基于空洞金字塔池化的通道和空间注意力机制的车道线检测算法(channel and position attention mechanism with atrous spatial pyramid pooling for ultra fast structure-aware deep lane detection,CPAM-UFLD)。【方法】首先,在超快速结构感知深度车道检测(ultra fast structure-aware deep lane detection,UFLD)算法的基础上融入了空洞金字塔池化模块(atrous spatial pyramid pooling,ASPP),以有效地捕捉车道线图像中不同尺度的特征;使用了通道和空间注意力机制(channel and position attention mechanism,CPAM)以关注图像中的关键区域;同时,为平衡类别权重和提高定位精度,使用了包括加权交叉熵损失函数等的四种损失函数;其次,提出了一种亮度改善模块,该模块旨在提升输入图像的质量,从而增强车道线的识别度。【结果】本算法在TuSimple数据集上的检测精度由原来的95.86%提升至96.56%;同时,在CULane数据集上,检测精度由原来的72.2%提升至73.7%。【结论】通过算法的改进,可以有效提高车道线检测的精度,这为智能网联汽车自动驾驶系统的环境感知提供了理论参考。
文摘安全帽在保护施工人员免受事故伤害方面发挥着至关重要的作用。然而,由于种种原因,工人们并没有严格执行戴安全帽的规定。为了检测工人是否佩戴安全帽,本文提出了基于YOLOv8改进的目标检测算法(CPAM-P2-YOLOv8)。在YOLOv8的颈部添加CPAM结构,增强网络对图片的特征提取,在YOLOv8的头部引入处理后的小目标检测层P2。CPAM-P2-YOLOv8提高了目标检测的精确度,实验结果表明,改进模型的精确度达到了91%。与YOLOv8对比,CPAM-P2-YOLOv8的mAP50提高了1.0%,参数量减少了17%,同时通过对比发现,CPAM-P2-YOLOv8比YOLOv8在检测小目标方面更有优势。与YOLOv10对比,CPAM-P2-YOLOv8的mAP50提高1.9%。使用知识蒸馏方法,使CPAM-P2-YOLOv8的精确度进一步提升,达到91.4%。Safety helmets play a crucial role in protecting construction workers from accidents and injuries. However, due to various reasons, workers did not strictly adhere to the rule of wearing safety helmets. To detect whether workers are wearing helmets, this article proposes an improved object detection algorithm based on YOLOv8 (CPAM-P2-YOLOv8). Add CPAM structure to the neck network of YOLOv8 to enhance the feature extraction of safety helmets, and introduce a processed small object detection layer P2 at the head of YOLOv8. CPAM-P2-YOLOv8 improved the accuracy of object detection, and experimental results showed that the improved model achieved an accuracy of 91%. Compared with the YOLOv8 model, CPAM-P2-YOLOv8 improved mAP50 by 1.0% and reduced parameter count by 17%. Through comparison, it was found that CPAM-P2-YOLOv8 has more advantages in detecting small targets than YOLOv8. Compared with YOLOv10, the mAP50 of CPAM-P2-YOLOv8 increased by 1.9%. By using the knowledge distillation, the precision of CPAM-P2-YOLOv8 was further improved to 91.4%.
