| 基于改进SSD模型的输电线路巡检图像金具检测方法 |
| |
| 引用本文： | 戚银城,江爱雪,赵振兵,郎静宜,聂礼强.基于改进SSD模型的输电线路巡检图像金具检测方法[J].电测与仪表,2019,56(22):7-12. |
| |
| 作者姓名： | 戚银城 江爱雪 赵振兵 郎静宜 聂礼强 |
| |
| 作者单位： | 华北电力大学电气与电子工程学院,河北保定,071003;山东大学计算机科学与技术学院,山东青岛,266237 |
| |
| 基金项目： | 国家自然科学基金项目（61871182、61401154、61773160、61302163）,北京市自然科学基金项目（4192055）,河北省自然科学基金项目（F2016502101、F2017502016、F2015502062）,中央高校基本科研业务费专项资金项目（2018MS095、2018MS094）,模式识别国家重点实验室开放课题基金（201900051） |
| |
| 摘 要： | 为了解决航拍图像金具智能检测问题,提出一种基于改进SSD模型的输电线路航拍巡检图像金具目标检测方法。对巡检图像进行多角度旋转并自适应裁剪扩充样本,以解决金具在图像中占比过小导致大量漏检问题,使用改进的IoU得到对目标尺度更敏感的默认框；进一步针对图像中金具目标密集问题,在模型中加入对密集目标检测有效的斥力损失,提高模型对密集遮挡金具的检测效果。在包含6934个训练目标框和1879个测试目标框的11类金具数据集中进行实验,使用文中方法的金具检测准确率达到了75.64%,相对于使用原始SSD模型检测准确率提升了4.73%,且检测框更贴合目标。
|
| 关 键 词： | 金具 SSD 目标检测 IoU 斥力损失 遮挡 |
| 收稿时间： | 2019/7/26 0:00:00 |
| 修稿时间： | 2019/7/26 0:00:00 |
Fittings Detection Method in Patrol Images of Transmission Line Based on Improved SSD |
| |
| Qi Yincheng,Jiang Aixue,Zhao Zhenbing,Lang Jingyi and Nie Liqiang.Fittings Detection Method in Patrol Images of Transmission Line Based on Improved SSD[J].Electrical Measurement & Instrumentation,2019,56(22):7-12. |
| |
| Authors: | Qi Yincheng Jiang Aixue Zhao Zhenbing Lang Jingyi and Nie Liqiang |
| |
| Affiliation: | School of Electrical and Electronic Engineering,North China Electric Power University,School of Electrical and Electronic Engineering,North China Electric Power University,School of Electrical and Electronic Engineering,North China Electric Power University,School of Electrical and Electronic Engineering,North China Electric Power University,School of Computer Science and Technology,Shandong University |
| |
| Abstract: | In order to solve the problem of intelligent detection of fittings in aerial photography images, this paper proposed a fittings detection method based on modified SSD in power transmission inspection images.The patrol image is rotated from multiple angles and adaptively tailored to expand the sample to solve the problem of a large number of missed detection caused by the small proportion of fittings in the image.Use the improved IoU to get the default box that is more sensitive to the ground truth scale.Further aiming at fittings is dense in the images, adding effective repulsion loss to the model for dense object detection, improve the detection effect of model for dense occlusion fittings.The experiment is based on the dataset of 11 kinds of fittings including 6934 training object boxes and 1879 test object boxes, the detection accuracy of the fittings using the method in the paper reaches 75.64%, which is 4.73% higher than that using the original SSD model, and the detection box is more suitable for the object. |
| |
| Keywords: | fittings SSD object detection IoU repulsion loss occlusion |
| 本文献已被 万方数据 等数据库收录！ |
| 点击此处可从《电测与仪表》浏览原始摘要信息 |
|
点击此处可从《电测与仪表》下载全文 |
|
