Abstract: 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.

Key words: building recognition, histogram of oriented gradients, feature extraction, steerable filters, support vector machine

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F1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。

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