| 引用本文: | 黄新波,高玉菡,张烨,赵隆,伍逸群,孙苏珍.基于联合分量灰度化算法和深度学习的玻璃绝缘子目标识别算法[J].电力自动化设备,2022,42(4): |
| HUANG Xinbo,GAO Yuhan,ZHANG Ye,ZHAO Long,WU Yiqun,SUN Suzhen.Glass insulator target recognition algorithm based on joint component grayscale algorithm and deep learning[J].Electric Power Automation Equipment,2022,42(4): |
|
| 摘要: |
| 针对相近色干扰、不同光照条件下玻璃绝缘子颜色特征不明显而无法准确识别的问题，提出一种基于联合分量灰度化算法和深度学习的玻璃绝缘子目标识别算法。首先，提出一种联合分量灰度化算法，通过补偿玻璃绝缘子目标区域的颜色特征实现目标增强；然后，在均匀分块的基础上，采用动态分块阈值进行玻璃绝缘子图像粗分割，并结合玻璃绝缘子的颜色和空间信息等多尺度高维特征，提出一种双尺度分类卷积神经网络算法实现玻璃绝缘子图像细分割；最后，将细分割得到的所有子图像进行合并，实现复杂背景下玻璃绝缘子目标的准确识别。实验结果表明，所提算法能对图像中存在相近色干扰、光照变化影响的玻璃绝缘子目标进行精准识别，且其在Dice参数、杰卡德系数2项识别指标上均达到90%以上，平均识别准确率高达92%。 |
| 关键词: 玻璃绝缘子 联合分量灰度化算法 动态分块阈值分割 双尺度分类卷积神经网络 深度学习 |
| DOI：10.16081/j.epae.202201032 |
| 分类号:TM216+.4 |
| 基金项目:陕西省自然科学基础研究计划-一般项目(青年)(2019JQ-843)；西安市科技计划项目(GXYD7.12)；陕西省教育厅科研计划项目(21JK0661) |
|
| Glass insulator target recognition algorithm based on joint component grayscale algorithm and deep learning |
|
HUANG Xinbo, GAO Yuhan, ZHANG Ye, ZHAO Long, WU Yiqun, SUN Suzhen
|
|
School of Electronics and Information, Xi’an Polytechnic University, Xi’an 710048, China
|
| Abstract: |
| Aiming at the problem that the color characteristics of glass insulators are not obvious and can not be recognized accurately under similar color interference and different lighting conditions, the glass insulator target recognition algorithm based on joint component grayscale algorithm and deep learning is proposed. Firstly, a joint component grayscale algorithm is proposed, which realizes the target enhancement by compensating the color features of target region of the glass insulator. Then, based on the uniform block segmentation, the dynamic block threshold is used for rough segmentation of glass insulator images. Meanwhile, combining with the multi-scale and high-dimensional characteristics of glass insulators, such as color and spatial information, a dual-scale classification convolutional neural network algorithm is proposed to achieve fine segmentation of glass insulator images. Finally, all sub-images obtained by fine segmentation are combined to achieve accurate recognition of glass insulator targets in complex background. The experimental results show that the proposed algorithm can accurately recognize the glass insulator target from the images with similar color interference and different lighting influences, and both of its two recognition indicators, i. e. Dice parameter and Jaccard coefficient are more than 90%,and the average recognition accuracy rate is as high as 92%. |
| Key words: glass insulator joint component grayscale algorithm dynamic block threshold segmentation dual-scale classification convolutional neural network deep learning |
