基于改进SSD的安全帽佩戴检测方法

Safety helmet wearing detection method based on improved SSD

安全帽是保障工业现场施工人员生命安全的关键性防护装备之一。针对现有安全帽佩戴检测速度慢及小目标识别准确率低的问题,提出了一种基于改进SSD算法的安全帽佩戴检测方法。采用VGG16作为基础模型,综合Faster R-CNN与YOLO算法的优点,在保证检测准确率的同时提高检测速度。利用不同卷积层的特征图,在多个特征图上产生多个候选框,提高了小目标检测的准确率;采用Adam优化器实现训练过程中神经网络的快速收敛;同时利用VGG16预训练模型的泛化能力进行迁移学习,加速了训练过程并减少了对数据量的需求。测试实验表明,相比于Faster R-CNN目标检测算法,该模型检测速度显著提升;相同的迭代次数下,相比于SSD模型,该方法对佩戴安全帽检测的平均准确率提升8.4%,达到了91.7%。

Safety helmet is one of the key protective equipment to ensure the safety of construction workers in industrial sites.Aiming at the existing problems of slow detection speed of helmet wearing and low accuracy of small target recognition,a method of detecting helmet wearing based on improved SSD algorithm is proposed.Using VGG16 as the basic model,combining the advantages of Faster R-CNN and YOLO algorithms,it improves the detection speed while ensuring the detection accuracy.Using the feature maps of different convolutional layers,multiple candidate frames are generated on multiple feature maps,which improves the accuracy of small target detection,the Adam optimizer is used to achieve rapid convergence of the neural network during the training process,at the same time,the VGG16 pre-training model is used.The generalization ability of migration learning accelerates the training process and reduces the demand for data.Test experiments show that compared to the Faster R-CNN target detection algorithm,the detection speed of this model is significantly improved;under the same number of iterations,compared to the SSD model,the average accuracy of the method for wearing helmet detection is increased by 8.4%,reaching this is 91.7%.