基于SSD算法的智能定员检测研究  

Research on Intelligent Assignment Detection Based on SSD Algorithms

作　　者：陈响洲 杨余旺[1] 沈兴鑫 CHEN Xiangzhou;YANG Yuwang;SHEN Xingxin(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094

出　　处：《计算机与数字工程》2021年第8期1555-1559,共5页Computer & Digital Engineering

基　　金：国防基础科研计划项目;江苏省科技重点及面上项目(编号:BE2018393);苏州市重点产业技术创新项目(编号:SYG201826)资助。

摘　　要：为了保障企业生产线的安全生产,针对生产线定员检测问题,论文提出了一种基于SSD目标检测算法的生产线定员检测算法,在传统SSD算法基础上优化了小尺寸目标的检测效果。该算法通过在ResNet101网络和添加的卷积层上选取适合的特征层进行提取多尺寸图像特征,然后产生一连串大小固定的边框和相应概率,最后使用非极大值抑制法得到最佳的预测值,实现对生产线人员数量实时监控。论文实验通过对生产线人员检测数据集进行测试,SSD_ResNet模型在该数据集上达到了87%的mAP,从人员准确率的角度验证了该算法的有效性。In order to ensure the safe production of the production line of the enterprise,this paper proposes a production line qualification detection algorithm based on the SSD target detection algorithm for the production line personnel inspection problem.Based on the traditional SSD algorithm,the detection effect of the small size target is optimized.The algorithm extracts multi-size image features by selecting appropriate feature layers on the ResNet101 network and the added convolutional layer,and then generates a series of fixed-size borders and corresponding probabilities.Finally,the non-maximum suppression method is used to obtain the best prediction value.Real-time monitoring of the number of production line personnel is achieved.In this paper,the SSD_ResNet model achieves 87%of mAP on the dataset by testing the dataset of the production line personnel.The effectiveness of the algorithm is verified from the perspective of personnel accuracy.

关 键 词：目标检测 SSD算法 生产线定员 神经网络 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]