| 改进多尺度特征融合的工业现场目标检测算法 |
| |
| 引用本文： | 刘瑞昊 1,于振中 2,孙 强 2. 改进多尺度特征融合的工业现场目标检测算法[J]. 机械与电子, 2022, 0(11): 40-45 |
| |
| 作者姓名： | 刘瑞昊 1 于振中 2 孙 强 2 |
| |
| 作者单位： | 1. 江南大学物联网工程学院,江苏 无锡 214122 ;2. 哈工大机器人国际创新研究院人工智能研究所,安徽 合肥 230601 |
| |
| 摘 要： | 为了提高工业现场等复杂场景下的小目标检测的准确率,降低工业现场的安全事故发生率,基于 YOLOv3 提出了一种改进多尺度特征融合方法。该方法增加了Inception _ shortcut 模块,优化网络的输出宽度,使用工业现场的监控视频作为数据集以及利用 k-means 算法对检测目标重新聚类,引入了 PANet 多尺度特征融合结构,精简了 YOLOv3 的网络检测输出层。在创建工业现场安全帽、安全绳数据集 FHPD 、FSRPD 以及 PASCAL VOC2007 数据集上的实验结果表明,改进算法的 mAP 比原始 YOLOv3 提高了许多。改进的多尺度特征网络融合增加了参数,但检测速度仍满足算法的实时性要求。
|
| 关 键 词： | 特征融合 目标检测 YOLOv3 算法 安全帽检测 安全绳检测 |
Improved Multi-scale Feature Fusion for Industrial Field Object Detection Algorithm |
| |
| LIU Ruihao1,YU Zhenzhong2,SUN Qiang2. Improved Multi-scale Feature Fusion for Industrial Field Object Detection Algorithm[J]. Machinery & Electronics, 2022, 0(11): 40-45 |
| |
| Authors: | LIU Ruihao1 YU Zhenzhong2 SUN Qiang2 |
| |
| Affiliation: | ( 1.School of Internet of Things Engineering , Jiangnan University , Wuxi 214122 , China ; 2.Institute of Artificial Intelligence , HRG International Institute for Research and Innovation , Hefei 230601 , China ) |
| |
| Abstract: | To improve the accuracy of small target detection in complex scenes such as industrial sites and reduce the incidence of safety accidents in industrial sites , an improved multi-scale feature fusion method is proposed based on YOLOv3 algorithm.The method added the Inception _ shortcut module to optimize the output width of the network , the surveillance videos of the industrial site were used as the data set and?k-means algorithm was used to re-cluster the detection targets , the PANet multi-scale feature fusion structure was introduced , and the network detection output layer of YOLOv3 was simplified.The experimental results on the industrial field helmets , safety rope datasets FHPD , FSRPD and PASCAL VOC2007 datasets showed that the mAP of the proposed algorithm is higher than the original YOLOv3. The improved multi-scale feature network fusion increases the parameters , but the detection speed still meets the real-time requirements of the algorithm. |
| |
| Keywords: | feature fusion object detection YOLOv3 algorithm helmet detection safety rope detection |
|
| 点击此处可从《机械与电子》浏览原始摘要信息 |
|
点击此处可从《机械与电子》下载免费的PDF全文 |
|
