摘要
针对k近邻(k-nearest neighbor,KNN)算法在土地覆盖分类中存在将山体阴影覆盖下植被误分成水体的问题,提出改进的KNN算法。改进算法充分利用神经网络能有效区分山体阴影覆盖下植被和水体的特性,实现BP神经网络与KNN算法的融合,整体提高了北京市密云区土地覆盖分类精度。实验结果表明:相对于支持向量机(support vector machine,SVM)、随机森林、BP神经网络和KNN算法,改进算法分类精度最高,达到了95.20%,分类精度比未改进KNN算法提高了6.43%。改进算法的Kappa系数在对比算法中也是最高的,达到0.93。此外,实验结果也表明改进算法可应用于中分辨率遥感图像分类中。
Aiming at the problem that k-nearest neighbor(KNN) algorithm misclassified vegetation under shadow of mountain into water area in land cover classification, an improved KNN algorithm was proposed. It combined the BP neural network algorithm with KNN algorithm taking advantage of the neural network’s characteristics of the high accuracy of distinguish vegetation under shadow of mountain from water area. The experimental results show that classification accuracy of the improved KNN algorithm is the highest when comparing with support vector machine(SVM), random forest, BP neural network and KNN algorithm, and accuracy of classification reaches 95.20%. The classification accuracy of the improved KNN algorithm is 6.43% higher than that of the unimproved KNN algorithm. The kappa coefficient of the improved KNN algorithm is also highest among the contrasting algorithms with the value of 0.93. In addition, the improved KNN algorithm can apply to the classification of moderate resolution remote sensing images.
作者
王佃来
宿爱霞
刘文萍
WANG Dian-lai;SU Ai-xia;LIU Wen-ping(Department of Information Engineering,Shougang Insititute of Technology,Beijing 100144,China;China Software Testing Center,Beijing 100048,China;College of Information,Beijing Forestry University,Beijing 100083,China)
出处
《科学技术与工程》
北大核心
2020年第23期9464-9471,共8页
Science Technology and Engineering
基金
北京市科技计划(Z171100001417005)
中央高校基本科研业务费专项(2015ZCQ-XX)。
关键词
KNN算法
土地覆盖分类
遥感图像
BP神经网络
k-nearest neighbor(KNN)algorithm
land cover classification
remote sensing images
BP neural network
