一种像素与对象相结合的林区建筑物识别方法

针对林区建筑物遥感监测技术需求，为构建GF-2数据在林区建筑物识别中的应用方法，选取蜀南竹海风景名胜区为研究区，根据所选区域建筑物的GF-2影像特征，研究形成了像素级和对象级相结合的林区建筑物识别方法。首先利用基于递归特征消除法的随机森林算法对预处理后的GF-2影像进行特征筛选；然后通过对比支持向量机和随机森林分类器识别的建筑物结果，选用支持向量机分类器所得研究区建筑物作为像素级识别结果；融合像素级建筑物识别结果和多尺度分割得到的影像对象，识别出该研究区建筑物目标。结果表明：利用支持向量机分类器进行像素级建筑物识别，其结果的正确率、完整率和质量均高于随机森林分类器；提出的像素级和对象级相结合的建筑物识别方法既保留了简单易行的优势，也避免了椒盐现象，在正确率、完整率和质量上均比像素级方法和对象级方法有所提高，在质量上分别比像素级方法和对象级方法提高了0.20和0.13，该方法可为主管单位有效监管林区内违规建筑物提供技术支撑。

Building Recognition Method in Forest Districts Combining the Pixel-level and Object-level

To meet the technical requirements of building monitoring in forest districts by using remote sensing images， The Southern Sichuan Bamboo Sea is selected as the study area to form the application method of building recognition from GF-2 data. According to image characteristics of the building in the selected area， a building recognition method that combines pixel-based and object-based methods in the forest district has been proposed. First， Random Forest-Recursive Feature Elimination is used to perform feature selection on the pre-processed GF-2 images. By comparing the results of the buildings identified by using SVM classifier and RF classifier， the building in the study area obtained by SVM classifier has been selected as the pixel-level building recognition result. Then the image objects are obtained using multiresolution segmentation method， and the building targets in the study area are identified by fusing both the pixel-level building result and the image objects. The results show that the correctness， completeness and quality of the building recognition result using SVM classifier are higher than RF classifier in the pixel-level. The proposed building recognition method combining pixel-level and object-level that not only retains the advantages of simplicity and ease of use， but also avoids the phenomenon of salt and pepper. The correctness， completeness and quality of the method are better than the pixel-level or the object-level method and the quality has been improved by 0.20 and 0.13， respectively. This method can provide technical support for the superior authorities to effectively supervise illegal buildings in forest districts.