基于深度学习的变电站多目标行人检测算法研究

Research on multi-objective pedestrian detection algorithms for substations based on deep learning

传统变电站行人检测算法通常是以数字波形式进行数据采集和处理的,无法对行人进行精准检测,针对该问题,提出了基于深度学习的变电站多目标行人检测算法研究。在深度学习网络基础上进行网络训练,将原始图像从RGB模型转换为HSV模型,分别在3个通道内进行图像数据预处理。采用动态自适应池化方法提取图像特征,根据每个池化域内部不同情况,自适应调整对应池化权值,借助池化因子实现对图像精准层抽象特征的提取。利用结构化边缘检测器生成边缘图像,选取行人候选框,将提取的结果作为深度学习网络输入,获取红马甲行人在变电站的全部信息,通过具体检测过程,完成多目标行人检测。在INRIA行人数据集支持下进行算法验证,并由结果可知,该算法最高检测精准度可达到98%,为变电站行人安全提供保障。

Traditional pedestrian detection algorithms in substations usually collect and process data in the form of digital waveforms,which can not accurately detect pedestrians. Aiming at this problem,a multi-objective pedestrian detection algorithm for substations based on in-depth learning is proposed. On the basis of deep learning network,network training is carried out. The original image is transformed from RGB model to HSV model,and image data is preprocessed in three channels. The dynamic adaptive pooling method is used to extract image features. According to the different conditions within each pooling domain,the corresponding pooling weights are adaptively adjusted,and the abstract features of the image precision layer are extracted by using the pooling factor. Structured edge detectors are used to generate edge images,select pedestrian candidate boxes,and input the extracted results as a deep learning network to obtain all the information of pedestrians in the substation. Through the specific detection process,multi-objective pedestrian detection is completed. The algorithm is validated with the support of INRIA pedestrian data set,and the results show that the maximum detection accuracy of the algorithm can reach 98%,which provides security for pedestrian safety in substations.