【摘要】 鉴于由于类型、规模、自然环境等的差异,不同港口在高分遥感影像上表现出的形状、方向、纹理等特征往往具有较大差异,因而利用低层特征在高分遥感影像上进行港口检测面临较大挑战,提出一种利用非监督提取算法提取高分遥感影像中层特征用于港口检测的方法。该方法通过对大量样本的自主学习,提取能够识别港口的可分图斑作为中层特征表达,再用词袋模型构建基于可分图斑的特征向量,并利用SVM分类器检测港口。最后,探讨了可分图斑提取算法中相关参数设置对检测结果的影响,并利用测试影像集对该方法进行了验证,结果表明其检测正确率达97.5%。更多还原

【Abstract】 Due to the differences in types,sizes and environments,different harbors in high-resolution remote sensing(HRRS)images show different features of shape,orientation and texture,which brings a great challenge to detect harbors with low-level features in HRRS images.In this paper,we propose a new harbor detection method based on an unsupervised mid-level feature extraction algorithm.Firstly,discriminative patches were extracted as mid-level features by learning from a large amount of samples.Then,feature vectors of discriminative patches detected from HRRS images were extracted to build a bag-of-words model.After that,a SVM classifier was built based on bag-of-words model for harbor detection.Finally,we discussed the parameter setting of discriminative patches extraction algorithm and used a test image set to testify the effect of the proposed method in this paper.The result showed that the accuracy of this method is 97.5%.更多还原