光伏组件清洁移动机器人地形分类方法研究  

STUDY ON TERRAIN CLASSIFICATION METHODS OF MOBILE ROBOTS FOR PHOTOVOLTAIC MODULES CLEANING

作　　者：李翠明[1] 刘爽 龚俊[1] Li Cuiming;Liu Shuang;Gong Jun(College of Mechanical and Electrical Engineering,Lanzhou University of Technology,Lanzhou 730050,China)

机构地区：[1]兰州理工大学机电工程学院,兰州730050

出　　处：《太阳能学报》2022年第3期210-215,共6页Acta Energiae Solaris Sinica

基　　金：甘肃省自然科学基金(18JR3RA139);甘肃省省级引导科技创新发展项目(2018ZX—13)。

摘　　要：针对移动机器人在光伏组件清洁过程中对光伏电站的非平坦地形分类问题,提出采用W-MC-MS准则优化视觉词袋模型的分类方法对地形图像进行分类。首先,对采集到的光伏电站实际地形图像进行SIFT特征提取,然后将这些特征利用K均值聚类算法进行聚类计算,生成地形图像的初始码本词典;引入W-MC-MS准则对其进行优化,以降低码本词典规模,提高视觉地形分类性能。实验以地形图像的平均分类精度和分类时间代价作为评价标准,对比码本词典优化前后的分类性能,验证了优化方法在地形分类中的有效性。Considering the classification of uneven terrain for photovoltaic power stations during photovoltaic panel cleaning by mobile robots,a classification method based on W-MC-MS was then proposed to optimize the Bag-of-visual-word Model to classify the terrain images.First of all,the SIFT features are extracted from the actual terrain images of the photovoltaic power stations;then,the K-means clustering algorithm is adopted to calculate these features and generate the initial code dictionary of terrain images;afterwards the WMC-MS criterion is adopted for optimization so as to reduce the size of codebook dictionary and improve the performance of visual terrain classification.The experiment considered the average classification accuracy and the cost of classification time regarding the terrain images as the evaluation standards,compared the classification performance of the codebook dictionary before and after optimization,and finally verified the validity of the optimization method in terrain classification.

关 键 词：移动机器人 光伏组件 K均值聚类 地形分类 视觉词袋模型 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]