基于迁移学习的永磁除铁器输送带断裂检测

Fracture detection method of permanent magnet self-unloading iron remover conveyor belt based on transfer learning

为解决现有的永磁自卸除铁器输送带断裂检测方法不精确、检测装置易损坏、安装复杂的问题,提出一种基于迁移学习的视频检测方法。首先采用模板匹配技术选取弃铁输送带的特征图像,得到源图像中的ROI区域;然后通过对图片进行随机旋转、裁剪、亮度变换扩充数据量,并自制了数据集;随后基于Fine Tuning迁移学习方法,使用Pytorch架构,利用Python语言对ResNet18网络模型进行训练,解决了因样本数量不足、学习效果差的问题。实验结果表明:弃铁输送带断裂检测的准确率可达93.74%,图像处理速度为17帧/s;将训练好的ResNet18网络模型布署到Jetson TX2嵌入式开发平台,当检测到故障时可现场实时报警;通过TCP/IP协议与监控终端系统进行数据通信,进行数据与输送带图像的实时显示,最终构建弃铁输送带断裂检测系统。

In order to solve the problems of the existing methods for detecting the conveyor belt broken of permanent mag-netic self-unloading iron remover,such as inaccuracy,easy damage of the detection device and complex in-stallation,a detection method based on machine vision and transfer learning is proposed.Firstly,the template matching technology is used to select the feature image of the abandoned iron conveyor belt to obtain the ROI area in the source image;then the data volume is expanded by randomly rotating,cropping,and brightness transformation of the picture,and the data set is made by ourselves;subsequently,based on fine tuning transfer learning method,using Pytorch architecture and Python language,ResNet18 network model is trained,which solves the problems of insufficient samples and poor learning effect.The experimental results show that the method and system in this paper have an accuracy rate of 93.74%for the detection of broken iron conveyor belts,and the image processing speed is 17 frames/s.The trained ResNet18 network model is deployed to the Jetson TX2 embedded development platform,when the failure occurs,it can alarm in real time on the spot;communi-cate with the monitoring terminal system through the TCP/IP protocol,display the data and the conveyor belt im-age in real time,and finally a broken iron conveyor belt detection system is built.