Figure 1.
Framework of the proposed method. The objects were obtained from principal components (PC) of two temporal images, and the CD objects are used as label data for the deep learning network. The result map of the deep learning network is integrated with the object boundary to update the CD objects, which involve change or no-change classes over a specific percentage for the entire region. This process is repeatedly performed until all the objects of the entire image are classified into change or no-change classes.
Figure 1.
Framework of the proposed method. The objects were obtained from principal components (PC) of two temporal images, and the CD objects are used as label data for the deep learning network. The result map of the deep learning network is integrated with the object boundary to update the CD objects, which involve change or no-change classes over a specific percentage for the entire region. This process is repeatedly performed until all the objects of the entire image are classified into change or no-change classes.
Figure 2.
Generation of CD objects. The pixels were classified as belonging to the same class in more than four methods among the five unsupervised CD methods and were selected to constitute the initial CD map, which was reconstructed in object units with threshold percentages, i.e., specific percentages of pixels in an object; in this study, 50%, 60%, and 70% were used to generate CD objects.
Figure 2.
Generation of CD objects. The pixels were classified as belonging to the same class in more than four methods among the five unsupervised CD methods and were selected to constitute the initial CD map, which was reconstructed in object units with threshold percentages, i.e., specific percentages of pixels in an object; in this study, 50%, 60%, and 70% were used to generate CD objects.
Figure 3.
The CD network architecture. The network comprises 3D convolutional layers to extract spatial and spectral features and convolutional LSTM layers to analyze the temporal relation between two features. Finally, two more 2D convolutional layers are considered to calculate the score map. Two temporal images and CD object were used as the input data. After the training step, the network produces a binary CD map. w, h, and represent the width, height, and number of spectral bands, respectively. and are the change and no-change classes, respectively. , , and are the change, no-change, and no-value classes of the initial CD map, respectively.
Figure 3.
The CD network architecture. The network comprises 3D convolutional layers to extract spatial and spectral features and convolutional LSTM layers to analyze the temporal relation between two features. Finally, two more 2D convolutional layers are considered to calculate the score map. Two temporal images and CD object were used as the input data. After the training step, the network produces a binary CD map. w, h, and represent the width, height, and number of spectral bands, respectively. and are the change and no-change classes, respectively. , , and are the change, no-change, and no-value classes of the initial CD map, respectively.
Figure 4.
The process of updating CD objects. CD objects are fed into the CD network, and the network produces a binary CD map, which can be divided into “noncandidate objects for updating” and “candidate objects for updating”. After uncertainty analysis, the selected objects were added to the previous CD object. and are the change and no-change classes, respectively. , , and are the change, no-change, and no-value classes of the initial CD map, respectively.
Figure 4.
The process of updating CD objects. CD objects are fed into the CD network, and the network produces a binary CD map, which can be divided into “noncandidate objects for updating” and “candidate objects for updating”. After uncertainty analysis, the selected objects were added to the previous CD object. and are the change and no-change classes, respectively. , , and are the change, no-change, and no-value classes of the initial CD map, respectively.
Figure 5.
VHR satellite images of two study sites and ground truth map. The Site 1 images were acquired from WorldView-3 over Gwanju city, South Korea on: (a) 26 May 2017; and (b) 4 May 2018. The Site 2 images were acquired from KOMPSAT-3 over Sejong city, South Korea on: (d) 16 November 2013; and (e) 26 February 2019. (c,f) The ground truth maps for Sites 1 and 2, respectively. and are the change and no-change classes, respectively.
Figure 5.
VHR satellite images of two study sites and ground truth map. The Site 1 images were acquired from WorldView-3 over Gwanju city, South Korea on: (a) 26 May 2017; and (b) 4 May 2018. The Site 2 images were acquired from KOMPSAT-3 over Sejong city, South Korea on: (d) 16 November 2013; and (e) 26 February 2019. (c,f) The ground truth maps for Sites 1 and 2, respectively. and are the change and no-change classes, respectively.
Figure 6.
CD result maps of Site 1 generated using various pixel-based unsupervised CD methods: (a) ; (b) ; (c) CVA; (d) IR-MAD; (e) PCA; and (f) initial CD map. and are the change and no-change classes, respectively. , and are the change, no-change, and no-value classes of the initial CD map, respectively.
Figure 6.
CD result maps of Site 1 generated using various pixel-based unsupervised CD methods: (a) ; (b) ; (c) CVA; (d) IR-MAD; (e) PCA; and (f) initial CD map. and are the change and no-change classes, respectively. , and are the change, no-change, and no-value classes of the initial CD map, respectively.
Figure 7.
CD result maps of Site 2 generated using various pixel-based unsupervised CD methods: (a) ; (b) ; (c) CVA; (d) IR-MAD; (e) PCA; and (f) initial CD map. and are the change and no-change classes in CD result map, respectively. , , and are the change, no-change, and no-value classes of the initial CD map, respectively.
Figure 7.
CD result maps of Site 2 generated using various pixel-based unsupervised CD methods: (a) ; (b) ; (c) CVA; (d) IR-MAD; (e) PCA; and (f) initial CD map. and are the change and no-change classes in CD result map, respectively. , , and are the change, no-change, and no-value classes of the initial CD map, respectively.
Figure 8.
Colored infrared images with the boundary of segmentation for at (a) Site 1 and (d) Site 2 and at (b) Site 1 and (e) Site 2. Colored composites of PCs with the segmentation results for (c) Site 1 and (f) Site 2.
Figure 8.
Colored infrared images with the boundary of segmentation for at (a) Site 1 and (d) Site 2 and at (b) Site 1 and (e) Site 2. Colored composites of PCs with the segmentation results for (c) Site 1 and (f) Site 2.
Figure 9.
CD objects overlapped with the boundaries of segments at different uncertainty levels: Level 3 at (a) Site 1 and (d) Site 2; Level 2 at (b) Site 1 and (e) Site 2; and Level 1 at (c) Site 1 and (f) Site 2. , , and are the change, no-change, and no-value classes of the initial CD object, respectively.
Figure 9.
CD objects overlapped with the boundaries of segments at different uncertainty levels: Level 3 at (a) Site 1 and (d) Site 2; Level 2 at (b) Site 1 and (e) Site 2; and Level 1 at (c) Site 1 and (f) Site 2. , , and are the change, no-change, and no-value classes of the initial CD object, respectively.
Figure 10.
CD results of traditional approaches using a CD network for Sites 1 and 2, respectively: (a,e) Case 1 (original images + initial CD map); (b,f) Case 2 (original images + CD objects); (c,g) Case 3 (adding object band images + initial CD map); and (d,f) Case 4 (postprocessing after Case 1). and are the change and no-change classes, respectively.
Figure 10.
CD results of traditional approaches using a CD network for Sites 1 and 2, respectively: (a,e) Case 1 (original images + initial CD map); (b,f) Case 2 (original images + CD objects); (c,g) Case 3 (adding object band images + initial CD map); and (d,f) Case 4 (postprocessing after Case 1). and are the change and no-change classes, respectively.
Figure 11.
The CD results of the proposed method for each epoch ( and epoch 0 and is epoch 200) with different uncertainties (the uncertainty level maintains one value in the training phase) at Site 1: (a) Level 3; (b) Level 2; and (c) Level 1. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 11.
The CD results of the proposed method for each epoch ( and epoch 0 and is epoch 200) with different uncertainties (the uncertainty level maintains one value in the training phase) at Site 1: (a) Level 3; (b) Level 2; and (c) Level 1. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 12.
The CD results of the proposed methods for each epoch with different uncertainties (the uncertainty level maintains one value in the training phase) at Site 2: (a) Level 3; (b) Level 2; and (c) Level 1. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 12.
The CD results of the proposed methods for each epoch with different uncertainties (the uncertainty level maintains one value in the training phase) at Site 2: (a) Level 3; (b) Level 2; and (c) Level 1. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 13.
The CD results of the proposed methods for each epoch. The uncertainty level increased at Site 1: (a) from Level 1 to Level 3; and (b) from Level 2 to Level 3. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 13.
The CD results of the proposed methods for each epoch. The uncertainty level increased at Site 1: (a) from Level 1 to Level 3; and (b) from Level 2 to Level 3. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 14.
The CD results of the proposed methods for each epoch. The uncertainty level increased at Site 2: (a) from Level 1 to Level 3; and (b) from Level 2 to Level 3. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 14.
The CD results of the proposed methods for each epoch. The uncertainty level increased at Site 2: (a) from Level 1 to Level 3; and (b) from Level 2 to Level 3. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 15.
The CD results of the proposed methods for each epoch. The uncertainty level decreased at Site 1: (a) from Level 3 to Level 1; and (b) from Level 2 to Level 1. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 15.
The CD results of the proposed methods for each epoch. The uncertainty level decreased at Site 1: (a) from Level 3 to Level 1; and (b) from Level 2 to Level 1. Grey (, white (, and black () colors represent the change, no-change, and no-value classes, respectively.
Figure 16.
The CD results of the proposed methods for each epoch. The uncertainty level decreased at Site 2: (a) from Level 3 to Level 1; and (b) from Level 2 to Level 1. Grey, white, and black colors represent the change, no-change, and no-value classes, respectively.
Figure 16.
The CD results of the proposed methods for each epoch. The uncertainty level decreased at Site 2: (a) from Level 3 to Level 1; and (b) from Level 2 to Level 1. Grey, white, and black colors represent the change, no-change, and no-value classes, respectively.
Figure 17.
CD results using different segmentation scales overlapped with the segment boundaries at Sites 1 and 2: (a,f) scale 30; (b,g) scale 50; (c,h) scale 100; (d,i) scale 200; and (e,j) scale 300.
Figure 17.
CD results using different segmentation scales overlapped with the segment boundaries at Sites 1 and 2: (a,f) scale 30; (b,g) scale 50; (c,h) scale 100; (d,i) scale 200; and (e,j) scale 300.
Figure 18.
F1 scores and OAs with different scale factors at: (a) Site 1; and (b) Site 2.
Figure 18.
F1 scores and OAs with different scale factors at: (a) Site 1; and (b) Site 2.
Table 1.
Accuracy of the CD maps generated using various pixel-based unsupervised CD methods.
Table 1.
Accuracy of the CD maps generated using various pixel-based unsupervised CD methods.
Study Site | Methods | OA | Precision | Recall | F1-Score |
---|
Site 1 | | 0.8316 | 0.6147 | 0.7760 | 0.6860 |
| 0.7778 | 0.5107 | 0.6685 | 0.5790 |
CVA | 0.8298 | 0.6189 | 0.7676 | 0.6853 |
IR-MAD | 0.8200 | 0.5946 | 0.7520 | 0.6641 |
PCA | 0.8471 | 0.8177 | 0.7135 | 0.7620 |
Site 2 | | 0.8649 | 0.7797 | 0.6676 | 0.7193 |
| 0.7705 | 0.5417 | 0.4848 | 0.5117 |
CVA | 0.8716 | 0.7762 | 0.6863 | 0.7285 |
IR-MAD | 0.8033 | 0.7496 | 0.5410 | 0.6285 |
PCA | 0.8108 | 0.6693 | 0.5622 | 0.6110 |
Table 2.
The number of pixels with different classes and uncertainty levels.
Table 2.
The number of pixels with different classes and uncertainty levels.
Study Site | Uncertainty Level | The Number of Pixels in |
---|
| | |
---|
Site 1 | Level 1 | 185,626 | 938,376 | 315,998 |
Level 2 | 215,381 | 997,600 | 227,019 |
Level 3 | 263,464 | 1,043,117 | 133,419 |
Site 2 | Level 1 | 19,215 | 96,322 | 44,463 |
Level 2 | 22,066 | 108,230 | 29,704 |
Level 3 | 24,280 | 117,225 | 18,495 |
Table 3.
Accuracy of the CD objects with different uncertainty levels.
Table 3.
Accuracy of the CD objects with different uncertainty levels.
Study Site | Uncertainty Level | Precision | NPV |
---|
Site 1 | Level 1 | 0.9420 | 0.9432 |
Level 2 | 0.9224 | 0.9331 |
Level 3 | 0.9183 | 0.9077 |
Site 2 | Level 1 | 0.9420 | 0.9432 |
Level 2 | 0.9153 | 0.9399 |
Level 3 | 0.8912 | 0.9339 |
Table 4.
Accuracy of CD maps generated via traditional approaches using a CD network.
Table 4.
Accuracy of CD maps generated via traditional approaches using a CD network.
Study Site | Methods | OA | Precision | Recall | F1-Score |
---|
Site 1 | Case 1 | 0.8623 | 0.6535 | 0.8726 | 0.7473 |
Case 2 | 0.8727 | 0.6335 | 0.9378 | 0.7562 |
Case 3 | 0.8436 | 0.5618 | 0.8982 | 0.6913 |
Case 4 | 0.8874 | 0.6878 | 0.9336 | 0.7921 |
Site 2 | Case 1 | 0.8787 | 0.5127 | 0.8883 | 0.6501 |
Case 2 | 0.8821 | 0.4933 | 0.9438 | 0.6479 |
Case 3 | 0.8772 | 0.5114 | 0.8801 | 0.6469 |
Case 4 | 0.8831 | 0.5068 | 0.9296 | 0.6560 |
Table 5.
Accuracy of the CD maps generated using the proposed method with different uncertainty levels.
Table 5.
Accuracy of the CD maps generated using the proposed method with different uncertainty levels.
Study Site | Uncertainty Level | OA | Precision | Recall | F1-Score |
---|
Site 1 | Level 1 | 0.8180 | 0.9446 | 0.4262 | 0.5874 |
Level 2 | 0.9174 | 0.9373 | 0.7847 | 0.8542 |
Level 3 | 0.8769 | 0.6444 | 0.9206 | 0.7581 |
Level 3 to Level 1 | 0.8735 | 0.6330 | 0.9193 | 0.7498 |
Level 2 to Level 1 | 0.9185 | 0.9411 | 0.7846 | 0.8558 |
Level 1 to Level 3 | 0.8611 | 0.9608 | 0.5780 | 0.7218 |
Level 2 to Level 3 | 0.9299 | 0.8158 | 0.9422 | 0.8745 |
Site 2 | Level 1 | 0.8998 | 0.6252 | 0.8908 | 0.7348 |
Level 2 | 0.9006 | 0.6220 | 0.8991 | 0.7353 |
Level 3 | 0.8802 | 0.9359 | 0.5159 | 0.6651 |
Level 3 to Level 1 | 0.8987 | 0.8908 | 0.6252 | 0.7348 |
Level 2 to Level 1 | 0.8957 | 0.9146 | 0.6766 | 0.7778 |
Level 1 to Level 3 | 0.8859 | 0.5223 | 0.9341 | 0.6700 |
Level 2 to Level 3 | 0.9012 | 0.6164 | 0.9092 | 0.7347 |
Table 6.
Accuracy of CD maps generated using the proposed methods with different scale factors.
Table 6.
Accuracy of CD maps generated using the proposed methods with different scale factors.
Study Site | Scale Factors | OA | Precision | Recall | F1-Score |
---|
Site 1 | 30 | 0.8795 | 0.6495 | 0.9256 | 0.7634 |
50 | 0.8969 | 0.7194 | 0.9185 | 0.8068 |
100 | 0.8965 | 0.7102 | 0.9267 | 0.8041 |
200 | 0.9299 | 0.8158 | 0.9422 | 0.8745 |
300 | 0.9097 | 0.7621 | 0.9228 | 0.8348 |
Site 2 | 30 | 0.8920 | 0.5627 | 0.9197 | 0.6982 |
50 | 0.9012 | 0.6164 | 0.9092 | 0.7347 |
100 | 0.8915 | 0.6063 | 0.8644 | 0.7127 |
200 | 0.8935 | 0.5621 | 0.9306 | 0.7009 |
300 | 0.8931 | 0.5632 | 0.9261 | 0.7004 |