激光与光电子学进展, 2018, 55 (2): 021001, 网络出版: 2018-09-10 复制并引用该论文  被引通知

图 & 表

图 1. DBN结构示意图

Fig. 1. Schematic of DBN structure

下载图片 查看原文

图 2. RBM结构示意图

Fig. 2. Schematic of RBM structure

下载图片 查看原文

图 3. DBN分类的流程图

Fig. 3. Flow chart of DBN classification

下载图片 查看原文

图 4. 最优分割尺度计算流程图

Fig. 4. Flow chart of optimal segmentation scale calculation

下载图片 查看原文

图 5. DBN不同层数的分类精度(实验1和3)

Fig. 5. Classification accuracy of different DBN layers (experiments 1 and 3)

下载图片 查看原文

图 6. DBN不同层数的分类精度(实验2和4)

Fig. 6. Classification accuracy of different DBN layers (experiments 2 and 4)

下载图片 查看原文

图 7. 4种不同的基于DBN的实验方法的分类结果图。(a) DBN+9×9像素邻域;(b) DBN+9×9像素邻域+光谱-纹理特征;(c) DBN+最优尺度+9×9像素邻域;(d) DBN+最优尺度+光谱-纹理特征

Fig. 7. Classification maps by four different experimental methods based on DBN. (a) Classification map of DBN+9×9 neighborhood pixels; (b) classification map of DBN+9×9 neighborhood pixels+spectral-texture features; (c) classification map of DBN+optimal scale+9×9 neighborhood pixels; (d) classification map of DBN+optimal scale+spectral-texture features

下载图片 查看原文

表 1不同植被提取方法的各种精度

Table1. Various accuracies of different vegetation extraction methods

Extractionmethod Accuracy Kappa coefficient Overall accuracy PA UA Type DBN+9×9neighborhoodpixels 0.8793 0.8987 0.9436 0.8856 Tree 0.8922 0.9381 Grass 0.8333 0.7653 Bare soil 0.9328 0.9680 Road 0.8435 0.8671 Others DBN+9×9neighborhoodpixels+spectral-texture features 0.872 0.8624 0.9685 0.8865 Trees 0.8681 0.9864 Grass 0.5432 0.8800 Bare soil 0.9911 0.8793 Road 0.7632 0.8286 Others DBN+optimalsegmentationscale+9×9neighborhood pixels 0.8677 0.9192 0.9416 0.9644 Trees 0.9309 0.8686 Grass 0.5714 0.6667 Bare soil 0.9150 0.9790 Road 0.6818 0.5000 Others DBN+optimalsegmentationscale+spectral-texture features 0.6152 0.7296 0.7973 0.8732 Tree 0.7353 0.7453 Grass 0.4000 0.2128 Bare soil 0.8909 0.8167 Road 0.5641 0.2619 Others

查看原文

刘祖瑾, 杨玲, 刘祖涵, 段琳琳, 乔贤贤, 龚娇娇. 一种结合深度置信网络与最优尺度的植被提取方法[J]. 激光与光电子学进展, 2018, 55(2): 021001. Zujin Liu, Ling Yang, Zuhan Liu, Linlin Duan, Xianxian Qiao, Jiaojiao Gong. Method of Vegetation Extraction Based on Deep Belief Network and Optimal Scale[J]. Laser & Optoelectronics Progress, 2018, 55(2): 021001.