摘要
为了快速准确地识别出红外图像中温度值实现缺陷检测,提出了面向电力设备红外图像的温度值识别算法。针对温度值区域背景复杂的问题,根据红外图像直方图自适应确定阈值进行预处理;结合轮廓与相对位置信息,准确定位温度值区域,并实现字符分割;建立温度值图像数据集,采用卷积神经网络进行训练和测试;基于MATLAB的App Designer模块,设计温度值识别与记录系统。结果证明,该算法对温度值识别准确率达到98.6%,高于传统的字符识别算法,能够实现快速识别与准确记录温度值,有效降低了电力巡检人员的劳动强度。
To quickly and accurately recognize the temperature value of an infrared image and realize defect detection,we propose a temperature value recognition algorithm for the infrared image of power equipment.Owing to the complex background of the temperature value region,the adaptive threshold is determined for preprocessing based on the infrared image histogram.Combined with the contour and relative position information,the precise location of the temperature value is segmented.The infrared image temperature values dataset is established,trained,and tested by the convolutional neural network.The temperature value recognition and recording system are designed based on the App Designer module of MATLAB.The experiment demonstrates that the accuracy of the proposed method reaches 98.6%,which is higher than the traditional character recognition algorithm.The proposed method can quickly and accurately recognize and record the temperature value,effectively reducing the labor intensity of power inspectors.
作者
王凯旋
任福继
倪红军
吕帅帅
汪兴兴
WANG Kaixuan;REN Fuji;NI Hongjun;LYU Shuaishuai;WANG Xingxing(School of Mechanical Engineering,Nantong University,Nantong 226019,China;Department of Intelligent Information Engineering,Tokushima University,Tokushima 7708501,Japan)
出处
《智能系统学报》
CSCD
北大核心
2022年第3期617-624,共8页
CAAI Transactions on Intelligent Systems
基金
江苏高校优势学科建设工程项目(PAPD)
德岛大学研究集群项目(2003002).
关键词
电力设备
红外图像
自适应阈值
图像分割
字符识别
卷积神经网络
缺陷检测
仿真系统
power equipment
infrared image
adaptive threshold
image segmentation
character recognition
convolutional neural network
defect detection
simulation system