基于改进HOG特征的建筑物识别方法  

Building recognition method based on improved HOG feature

作　　者：杨松[1,2,3] 李盛阳 邵雨阳[1,2] 郑贺[1,2,3] YANG Song;LI Shengyang;SHAO Yuyang;ZHENG He(Technology and Engineering Center for Space Utilization,Chinese Academy of Sciences,Beijing 100094,China;Key Laboratory of Space Utilization,Chinese Academy of Sciences,Beijing 100094,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院空间应用工程与技术中心,北京100094 [2]中国科学院太空应用重点实验室,北京100094 [3]中国科学院大学,北京100049

出　　处：《计算机工程与应用》2018年第7期196-200,共5页Computer Engineering and Applications

基　　金：中国科学院空间科学研究院培育项目(No.Y5031211WY)

摘　　要：随着机器学习方法的广泛应用,建筑物识别技术得到了快速的发展,识别的准确性一直是人们关注的重点。梯度方向直方图(HOG)特征提取方法中的梯度求解方式不能有效提取建筑物的边界特征,直接影响了识别的准确性,提出基于方向可控滤波器的HOG算法,利用支持向量机学习方法实现建筑物的识别。实验结果表明,该方法在平均准确率、TP、FP、召回率、精确率和F_1值等指标上优于基于方向可控滤波器的建筑物识别方法,证明了该方法可以有效识别建筑物。With the development and the extensive application of machine learning methods,the building recognition technology is developed rapidly.For the building recognition,the recognition accuracy is still the focus of attention.The traditional gradient method of Histogram of Oriented Gradients(HOG)can not effectively describe the boundary characteristics of the building and it directly affects recognition result.The HOG method based on the steerable filters is proposed,and Support Vector Machine(SVM)is used as the training method.The experimental result is analyzed according to the average accuracy rate,TP,FP,Recall,Precision and F1.The results show that the proposed method has better performance than the steerable filtered-based building recognition method,and it is proven that the proposed method can effectively identify buildings.

关 键 词：建筑物识别 梯度方向直方图 特征提取 方向可控滤波器 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]