基于深度学习的激光无线充电保护系统
Laser wireless charging protection system based on deep learning
钱绣洁 1陈瀚林 1马海霞 1杨雁南 1蓝建宇2
作者信息
- 1. 南京航空航天大学 物理学院 物理系,南京 211100,中国
- 2. 中国航天科技集团有限公司上海空间电源研究所,上海 200245,中国
- 折叠
摘要
为了提高智能家居远程激光无线充电系统使用的安全性,采用了一种基于深度学习的激光无线充电保护系统.针对贴附于智能家居表面的光伏电池属于小目标,具有不易识别的难点,改进得到了YOLOv7-NH网络模型,设立保护监测区,并融入帧间差分法用于实时监控充电区域;通过创设原理分析-算法框架搭建-环境调试等环节,编写了对充电目标所在区域进行图像监控的保护算法,并搭建了测试系统.结果表明,当激光发射端与充电目标距离在10m以内,基于该算法搭建的保护系统的响应启动时间均低于1 ms,即当移动异物以低于常规速率 1.5 m/s进入大小为 40 mm×40 mm的保护监测区时,该保护系统能够在其运动到激光束所在光路前停止激光发射.这一结果对室内激光远程无线充电保护技术的发展是有帮助的.
Abstract
In order to improve the safety of the remote laser wireless charging system used in smart homes,a laser wireless charging protection system based on deep learning was adopted.Meanwhile,in response to the small target of photovoltaic cells attached to the surface of smart homes,which were difficult to identify,a YOLOv7-NH network model was improved to establish a protection monitoring area and incorporate inter frame difference method for real-time monitoring of charging areas.A protection algorithm for image monitoring of the area where the charging target was located was written through the steps of creating a principle analysis algorithm framework building environment debugging,and a testing system was built.The test results show that when the distance between the laser emitting end and the charging target is within 10 m,the response start time of the protection system built based on this algorithm is less than 1ms.That is,when a moving foreign object enters the protection monitoring area with a size of 40 mm×40 mm at a speed of 1.5 m/s below the normal speed,the protection system can stop laser emission before it moves to the optical path where the laser beam is located.This result is helpful for the development of indoor laser remote wireless charging protection technology.
关键词
激光技术/激光远程无线充电/深度学习/YOLOv7-NH/帧间差分法Key words
laser technique/laser remote wireless charging/deep learning/YOLOv7-NH/inter frame difference method引用本文复制引用
基金项目
国家自然科学基金资助项目(51577091)
出版年
2024
国家科技期刊平台
NSTL
万方数据