光学学报, 2019, 39 (6): 0610002, 网络出版: 2019-06-17 复制并引用该论文  本文已被引 2 次  被引通知

图 & 表

图 1. 所提算法的整体框架图

Fig. 1. Whole framework of proposed method

下载图片 查看原文

图 2. 所提算法运算流程图

Fig. 2. Flow chart of proposed algorithm

下载图片 查看原文

图 3. UCM数据集中的若干类样本。(a)农田;(b)飞机;(c)棒球场;(d)密集住宅;(e)高速公路;(f)海港;(g)储罐;(h)网球场;(i)立交桥;(j)高尔夫球场

Fig. 3. Images of several UCM classes. (a) Agricultural; (b) airplane; (c) baseball diamond; (d) dense residential; (e) freeway; (f) harbor; (g) storage tanks; (h) tennis court; (i) overpass; (j) golf course

下载图片 查看原文

图 4. AID数据集中的若干类样本。(a)机场;(b)贫瘠地;(c)海滩;(d)桥梁;(e)商业区;(f)运动场;(g)池塘;(h)火车站;(i)体育场;(j)立交桥

Fig. 4. Images of several AID classes. (a) Airport; (b) bareland; (c) beach; (d) bridge; (e) commercial; (f) playgound; (g) pond; (h) railway station; (i) stadium; (j) viaduct

下载图片 查看原文

图 5. 所提算法在UCM上融合时的各unseen类的分类准确度。(a)高层图像特征融合;(b)中层图像特征融合;(c)低层图像特征融合;(d)不同层图像特征融合

Fig. 5. Test accuracies of unseen classes of our algorithm's fusion on UCM dataset. (a) High-level feature fusion; (b) middle-level feature fusion; (c) low-level feature fusion; (d) different-level feature fusion

下载图片 查看原文

图 6. 所提算法在AID上融合时的各unseen类的分类准确度。(a)高层图像特征融合;(b)中层图像特征融合;(c)低层图像特征融合;(d)不同层图像特征融合

Fig. 6. Test accuracies of unseen classes of our algorithm's fusion on AID dataset. (a) High-level feature fusion; (b) middle-level feature fusion; (c) low-level feature fusion; (d) different-level feature fusion

下载图片 查看原文

图 7. 所提算法在UCM场景集上融合时的总体损失曲线和测试准确度折线。(a)高层图像特征融合;(b)中层图像特征融合;(c)低层图像特征融合;(d)不同层图像特征融合

Fig. 7. Overall loss and test accuracy of our algorithm's fusion on UCM dataset. (a) High-level feature fusion; (b) middle-level feature fusion; (c) low-level feature fusion; (d) different-level feature fusion

下载图片 查看原文

图 8. 所提算法在AID场景集上融合时的总体损失曲线和测试准确度折线。(a)高层图像特征融合;(b)中层图像特征融合;(c)低层图像特征融合;(d)不同层图像特征融合

Fig. 8. Overall loss and test accuracy of our algorithm's fusion on AID dataset. (a) High-level feature fusion; (b) middle-level feature fusion; (c) low-level feature fusion; (d) different-level feature fusion

下载图片 查看原文

表 1UCM数据集上ZSC算法的相同层次图像特征融合效果OA值

Table1. OA values of different ZSC algorithms on UCM dataset for fusion of the same level features%

Features OA LatEm BiDiLEL JLSE SSE DMaP SAE RKT Ours High-level features CaffeNet 18.96 35.81 31.28 30.81 28.52 36.60 32.02 42.63 VGGNet 20.06 32.83 35.04 26.23 28.61 45.60 34.24 46.82 GoogLeNet 15.68 37.04 35.98 34.24 25.44 44.20 28.44 48.04 ResNet 20.00 35.01 22.10 19.58 25.01 24.24 18.03 38.74 Fusion_CAT 20.20 34.44 31.88 28.02 22.42 43.20 32.01 44.42 Fusion_CCA 20.54 35.46 21.34 20.44 27.41 30.06 24.44 37.24 Fusion_ADL 22.86 31.84 29.54 44.83 26.81 36.40 29.62 45.63 Fusion_Ours 23.20 35.63 37.80 49.21 31.83 44.80 34.41 61.41 Middle-level features BoVW 20.80 36.83 20.72 26.64 18.84 25.80 29.83 37.24 IFK 20.74 47.04 27.34 19.24 26.76 39.20 26.04 49.22 LDA 21.92 38.03 27.56 27.83 31.22 29.40 33.59 39.23 LLC 20.82 45.37 26.66 18.21 33.44 28.20 27.18 47.42 pLSA 19.64 39.19 29.78 22.64 31.63 29.20 26.21 40.81 SPM 21.94 45.37 28.36 21.03 23.04 32.40 27.82 46.84 VLAD 20.38 42.63 26.62 24.04 30.02 39.00 29.83 44.81 Fusion_CAT 21.70 33.64 18.62 23.01 22.41 28.60 28.76 34.83 Fusion_CCA 21.48 35.04 20.40 29.62 20.63 34.80 28.37 37.61 Fusion_ADL 20.40 40.64 32.34 23.44 35.57 37.80 29.64 44.04 Fusion_Ours 23.70 46.02 33.28 40.61 37.19 39.80 34.03 59.41 Low-level features CH 19.76 25.84 21.24 21.64 15.60 20.60 16.21 26.21 SIFT 20.80 43.24 21.22 21.38 41.82 28.40 20.02 44.63 GIST 21.98 37.43 20.86 18.19 31.61 21.20 20.04 39.84 LBP 20.28 44.41 31.50 26.41 39.24 37.00 26.24 45.82 Fusion_CAT 20.20 47.04 22.64 21.43 41.03 39.40 20.03 47.81 Fusion_CCA 20.22 44.83 25.80 28.84 38.81 34.80 32.84 45.84 Fusion_ADL 23.06 47.04 30.96 27.81 41.61 37.60 30.03 47.43 Fusion_Ours 23.20 47.24 32.40 34.24 43.24 41.20 38.19 54.47

查看原文

表 2AID数据集上ZSC算法的相同层次图像特征融合效果OA值

Table2. OA values of different ZSC algorithms on AID dataset for fusion of the same level features%

Features OA LatEm BiDiLEL JLSE SSE DMaP SAE RKT Ours High-level features CaffeNet 19.48 51.30 40.63 49.05 41.54 36.30 45.55 52.23 VGGNet 19.93 49.94 44.37 44.56 41.30 33.40 44.73 52.09 GoogLeNet 20.12 51.59 44.80 48.76 45.21 43.30 51.18 53.27 ResNet 19.98 37.69 17.95 29.47 36.39 32.20 17.87 41.65 Fusion_CAT 19.99 52.99 39.74 29.76 35.80 38.90 49.64 53.41 Fusion_CCA 18.86 51.03 20.70 45.86 38.70 45.10 35.50 51.96 Fusion_ADL 20.42 52.77 43.83 55.15 46.21 43.90 52.60 55.52 Fusion_Ours 21.02 53.09 47.22 55.38 45.62 54.90 50.12 68.34 Middle-level features BoVW 20.04 36.04 40.33 51.83 29.70 44.00 35.56 52.76 IFK 19.75 50.01 45.22 30.89 34.67 43.10 28.17 51.89 LDA 20.01 36.12 42.35 47.10 34.44 38.90 36.21 48.67 LLC 19.72 44.58 37.61 44.85 30.00 41.30 47.99 48.51 pLSA 20.15 35.37 41.36 49.53 34.44 36.30 44.73 50.93 SPM 20.04 43.36 38.78 37.46 35.15 37.20 33.43 46.80 VLAD 20.13 36.39 35.47 41.60 29.17 35.60 33.37 46.56 Fusion_CAT 20.15 35.18 40.67 41.78 37.34 43.80 34.50 46.37 Fusion_CCA 21.76 36.35 19.12 15.74 23.61 27.40 31.66 46.04 Fusion_ADL 20.11 44.13 37.34 38.76 34.73 33.80 33.79 44.18 Fusion_Ours 20.91 45.17 38.28 42.49 40.53 36.90 32.43 66.05 Low-level features CH 20.00 40.87 35.00 30.53 45.09 26.00 18.82 46.07 SIFT 19.98 25.19 13.59 17.81 28.64 19.20 15.38 30.28 GIST 19.81 27.34 29.17 17.51 39.70 26.70 15.50 40.68 LBP 19.89 31.07 15.45 21.66 26.75 32.40 15.38 34.40 Fusion_CAT 19.76 36.87 38.65 31.30 40.00 35.50 15.38 43.19 Fusion_CCA 19.84 35.77 40.21 31.66 40.47 24.30 43.25 46.08 Fusion_ADL 20.03 34.72 40.04 28.17 38.11 36.50 36.39 44.15 Fusion_Ours 20.37 46.60 41.77 32.31 46.04 36.60 46.86 53.91

查看原文

表 3不同ZSC算法在UCM数据集上的不同层次图像特征融合OA值

Table3. OA values of different ZSC algorithms on UCM dataset for fusion of different level features%

Method OA High-level feature Middle-level feature Low-level feature Fusion_CAT Fusion_CCA Fusion_ADL Fusion_Ours LatEm 20.06VGGNet 21.94SPM 21.98GIST 20.96 20.87 21.6 27.46 BiDiLEL 37.04GoogLeNet 47.04IFK 44.41LBP 36.82 32.84 39.42 47.83 JLSE 35.98GoogLeNet 29.78pLSA 31.50LBP 34.24 35.12 37.20 38.52 SSE 34.24GoogLeNet 27.83LDA 26.41LBP 31.64 34.80 38.05 39.83 DMaP 28.61VGGNet 33.44LLC 41.82SIFT 30.62 39.21 38.83 42.44 SAE 45.60VGGNet 39.20IFK 37.00LBP 45.60 47.60 48.90 49.20 RKT 34.24VGGNet 33.59LDA 26.24LBP 35.64 34.81 33.43 38.54 Ours 48.04GoogLeNet 49.22IFK 45.82LBP 46.44 47.59 48.24 61.43

查看原文

表 4不同ZSC算法在AID数据集上不同层次图像特征融合OA值

Table4. OA values of different ZSC algorithms on AID dataset for fusion of the different levels features%

Methods OA High-level feature Middle-level feature Low-level feature Fusion_CAT Fusion_CCA Fusion_ADL Fusion_Ours LatEm 20.12GoogLeNet 20.15pLSA 20.00CH 20.08 20.13 20.33 21.16 BiDiLEL 51.59GoogLeNet 50.01IFK 40.87CH 53.50 47.51 52.58 54.55 JLSE 44.80GoogLeNet 45.22IFK 35.00CH 45.80 43.61 47.57 49.40 SSE 49.05CaffeNet 51.83BoVW 30.53CH 38.39 33.24 42.31 45.27 DMaP 45.21GoogLeNet 35.15SPM 45.09CH 40.47 45.03 45.86 47.37 SAE 43.30GoogLeNet 44.00BoVW 32.40LBP 28.20 37.4 39.5 42.20 RKT 51.18GoogLeNet 47.99LLC 18.82CH 51.49 47.75 50.71 52.43 Ours 52.23GoogLeNet 52.76BoVW 46.07CH 50.85 53.58 55.48 66.82

查看原文

表 5各ZSC算法在AID上的GoogLeNet特征运算耗时

Table5. Computing time of different ZSC algorithms on AID dataset for GoogLeNet feature

Method LatEm BiDiLEL JLSE SSE DMaP SAE RKT Ours Time /s 85.65 252.22 83.39 172.19 73.68 75.46 485.57 71.59

查看原文

吴晨, 王宏伟, 袁昱纬, 王志强, 刘宇, 程红, 全吉成. 基于图像特征融合的遥感场景零样本分类算法[J]. 光学学报, 2019, 39(6): 0610002. Chen Wu, Hongwei Wang, Yuwei Yuan, Zhiqiang Wang, Yu Liu, Hong Cheng, Jicheng Quan. Image Feature Fusion Based Remote Sensing Scene Zero-Shot Classification Algorithm[J]. Acta Optica Sinica, 2019, 39(6): 0610002.