摘要
首先采用基于同态滤波技术的局部直方图均衡化算法和自适应中值滤波算法消除复合绝缘子憎水性图像的高频噪声。其次,鉴于憎水性图像中水珠引起的反光和透明等干扰,采用最大类间方差作为目标函数和遗传算法作为阈值的优化算法,获取了良好的分割效果。最后,将最大水珠区域的图像的面积比、形状因子、伸长度、7个不变矩共10个特征参数输入BP神经网络,对7个憎水性等级进行判定,结果表明训练准确率和测试准确率分别高达94%和90%。
The high-frequency noise of composite insulator hydrophobic image is firstly eliminated based on the local histogram equalization algorithm for homomorphic filtering and adaptive median filtering algorithm. Then,in view of the reflection and transparency caused by water droplets in the hydrophobic image,good segmentation effect is obtained through adopting maximum interclass variance as objective function and genetic algorithm as a threshold optimization algorithm. Finally,ten feature parameters,including area ratio,shape factor,extension degree,seven invariant moments,are placed into the BPNN,and the results indicate that the training accuracy and testing accuracyare as high as 94% and 90%,respectively.
作者
张广东
张玉刚
温定筠
姚境
王晓飞
高立超
郭陆
ZHANG Guang-dong;ZHANG Yu-gang;WEN Ding-jun;YAO Jing;WANG Xiao-fei;GAO Li-chao;GUO Lu(Cansu Electric Power Research Institute of State Grid,Lan Zhou,Gansu 730070,China;Gan Su Electric Power Company of State Grid,Lan Zhou,Gansu 730010,China;School of Electrical and Information Engineering,Hunan University,Changsha,Hunan 410082,China)
出处
《计算技术与自动化》
2019年第4期19-24,共6页
Computing Technology and Automation
基金
国网甘肃电力公司科技项目资助(522722160021)
关键词
憎水性自动检测
同态滤波
遗传阈值
BP神经网络
automatically hydrophobic detection
homomorphic filtering
genetic threshold
BP neural networks
