October 2019 Layout Flipping Full - page 749

Table 7. The evaluation indexes of comparing supervised algorithms and ours.
No.
MKL+GC
(Tao et al. 2012a)
MMM
(Li et al. 2017a)
VW (Weizman and
Goldberger 2009)
Deeplab v3+
(Chen et al. 2018)
LMB-CNN +GC
(Tan et al. 2019) ResNet +FCN
LMB-CNN +FCN
(Ours)
User.Acc
1 0.1723 0.4850 0.0610 0.7270 0.7939 0.7529 0.7640
2 0.1332 0.2937 0.0597 0.8777 0.7926 0.7412 0.8094
3 0.3570 0.7390 0.1588 0.6924 0.7764 0.7431 0.7830
4 0.3740 0.6177 0.1999 0.7054 0.7881 0.7405 0.7851
5 0.7494 0.9155 0.5055 0.9952 0.9244 0.9011 0.9186
6 0.4307 0.7437 0.2641 0.8298 0.8365 0.779 0.8299
7 0.2166 0.4779 0.0981 0.2298 0.8298 0.7849 0.8155
8 0.3390 0.8598 0.0597 0.1258 0.8308 0.7742 0.7052
9 0.3005 0.6682 0.2176 0.7492 0.7668 0.6616 0.7767
10 0.3717 0.7423 0.1554 0.7809 0.8136 0.7758 0.8086
All 0.3672 0.6866 0.1908 0.6713 0.8353 0.7853 0.8285
Prod.Acc
1 0.9199 0.9509 0.4344 0.7552 0.9222 0.9047 0.9387
2 0.9925 0.9107 0.4960 0.6734 0.8631 0.9085 0.8669
3 0.9977 0.9429 0.5059 0.8905 0.9680 0.9813 0.9694
4 0.9950 0.9551 0.4474 0.7771 0.9722 0.9816 0.9771
5 0.9803 0.9641 0.6692 0.8539 0.9916 0.9969 0.9904
6 0.9983 0.9550 0.5928 0.8931 0.9720 0.9841 0.9749
7 0.9961 0.9861 0.5795 0.3038 0.8429 0.9188 0.8429
8 0.9802 0.8895 0.5550 0.7558 0.9170 0.9659 0.9431
9 0.9988 0.9875 0.6461 0.9781 0.9899 0.9943 0.9879
10 0.9976 0.9480 0.5104 0.9325 0.9789 0.9914 0.9797
All 0.9879 0.9577 0.5788 0.7813 0.9670 0.9797 0.9691
Overall.Acc
1 0.7337 0.9373 0.5708 0.9687 0.9812 0.9768 0.9792
2 0.6472 0.8756 0.5464 0.9771 0.9802 0.9777 0.9816
3 0.8129 0.9594 0.6700 0.9475 0.9677 0.9628 0.9689
4 0.7309 0.8976 0.6224 0.9118 0.9534 0.9416 0.9532
5 0.8528 0.9471 0.5824 0.9363 0.9621 0.9523 0.9587
6 0.7998 0.9433 0.6880 0.9561 0.9670 0.9553 0.9659
7 0.7931 0.9374 0.6701 0.9017 0.9811 0.9809 0.9800
8 0.9261 0.9902 0.6484 0.7895 0.9897 0.9879 0.9827
9 0.6195 0.9177 0.5620 0.9429 0.9491 0.9158 0.9516
10 0.7838 0.9512 0.5821 0.9579 0.9686 0.9622 0.9677
All 0.7700 0.9357 0.6143 0.9290 0.9700 0.9613 0.9690
Mean IOU
1 0.4440 0.7034 0.2893 0.7778 0.8621 0.8365 0.8528
2 0.3801 0.5773 0.2805 0.7959 0.8416 0.8331 0.8503
3 0.5740 0.8312 0.3516 0.7901 0.8605 0.8457 0.8648
4 0.5264 0.7396 0.3483 0.7429 0.8578 0.8305 0.8577
5 0.7432 0.8979 0.3889 0.8753 0.9259 0.908 0.9197
6 0.5973 0.8261 0.3831 0.8520 0.8890 0.8584 0.8863
7 0.4985 0.7042 0.3320 0.5252 0.8494 0.8569 0.8434
8 0.6299 0.8833 0.3142 0.4521 0.8810 0.8705 0.8293
9 0.4228 0.7822 0.3297 0.8344 0.8500 0.7793 0.8558
10 0.5617 0.8288 0.3201 0.8456 0.8818 0.8636 0.8792
All 0.5502 0.7963 0.3418 0.7491 0.8888 0.8641 0.8858
Table 8. The time consumption of comparing supervised algorithms and ours (seconds).
MKL+GC
(Tao et al. 2012a)
MMM
(Li et al. 2017a)
VW (Weizman and
Goldberger 2009)
Deeplab v3+
(Chen et al. 2018)
LMB-CNN +GC
(Tan et al. 2019)
ResNet
+FCN
LMB-CNN
+FCN (Ours)
Average time 7877.3 628.8 186.2 16 4.0 2.7 3.0
Table 9. The basic information of the WorldView-3 images.
No. Collection Time Longitude and Latitude Location
1 8 December 2014 E121.52°,S30.79° Kalgoorlie, Australia
2 10 September 2014 E46.59°, S23.54° São Paulo, Brazil
3 9 September 2014 E3.57°, N40.48º Madrid, Spain
4 25 September 2014 E77.24°, N28.60° New Delhi, India
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING October 2019 749