Author Affiliations
1Faculty of Geomatics, Lanzhou Jiaotong University, Lanzhou 730070, Gansu , China2National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring, Lanzhou 730070, Gansu , China3Gansu Provincial Engineering Laboratory for National Geographic State Monitoring, Lanzhou 730070, Gansu , Chinashow less
Fig. 1. Architecture of proposed object detection model
Fig. 2. Comparison between convolution and dilated convolution. (a) General convolution; (b) dilated convolution
Fig. 3. Feature fusion network. (a) FPN; (B) PANet
Fig. 4. Multi-dimensional attention module
Fig. 5. Detection results on RSOD dataset. (a) Aircraft; (b) oiltank; (c) overpass; (d) playground
Fig. 6. Detection results on DOTA dataset. (a) PL; (b) GTF and SBF; (c) BR; (d) BD; (e) LV and SV; (f) RA; (g) HA; (h) ST; (i) SH; (j) BC and TC; (k) SP; (l) HC
Fig. 7. Examples of miss detection and error detection. (a) Miss detection; (b) error detection
Dataset | Aircraft | Oiltank | Overpass | Playground | mAP |
---|
RSOD | 81.02 | 90.77 | 100 | 100 | 92.95 |
|
Table 1. Detection results on RSOD dataset
Dataset | PL | BD | BR | GTF | SV | LV | SH | TC | BC | ST | SBF | RA | HA | SP | HC | mAP |
---|
DOTA | 89.68 | 84.39 | 52.12 | 72.71 | 64.49 | 67.07 | 77.45 | 90.11 | 83.98 | 86.01 | 64.03 | 63.33 | 74.45 | 67.74 | 62.87 | 73.39 |
|
Table 2. Detection results on DOTA dataset
Type | Aircraft | Oiltank | Overpass | Playground | mAP |
---|
FPN | 78.86 | 90.48 | 99.73 | 100.00 | 92.26 | FPN++ | 81.02 | 90.77 | 100.00 | 100.00 | 92.95 |
|
Table 3. Experimental comparison of different FPN on RSOD dataset
Type | PL | BD | BR | GTF | SV | LV | SH | TC | BC | ST | SBF | RA | HA | SP | HC | mAP |
---|
FPN | 88.81 | 84.33 | 50.98 | 72.45 | 62.71 | 65.35 | 75.34 | 89.96 | 81.16 | 84.07 | 53.26 | 63.33 | 72.24 | 66.29 | 60.11 | 71.36 | FPN++ | 89.68 | 84.39 | 52.12 | 72.71 | 64.49 | 67.07 | 77.45 | 90.11 | 83.98 | 86.01 | 55.03 | 63.33 | 74.45 | 67.74 | 62.87 | 72.86 |
|
Table 4. Experimental comparison of different FPN on DOTA dataset
Algorithm | aircraft | oiltank | overpass | playground | mAP |
---|
R-FCN | 71.48 | 90.23 | 76.84 | 97.70 | 84.07 | Deformable R-FCN | 71.50 | 90.26 | 81.48 | 99.53 | 85.70 | Faster R-CNN | 71.90 | 90.90 | 100.00 | 100.00 | 90.70 | RFN | 79.10 | 90.50 | 100.00 | 99.70 | 92.30 | MDCF2Det | 81.02 | 90.77 | 100.00 | 100.00 | 92.95 |
|
Table 5. Accuracy comparison of different algorithms on RSOD dataset
Algorithm | PL | BD | BR | GTF | SV | LV | SH | TC | BC | ST | SBF | RA | HA | SP | HC | mAP |
---|
YOLO v2 | 76.90 | 33.87 | 22.73 | 34.88 | 38.73 | 32.02 | 52.37 | 61.65 | 48.54 | 33.91 | 29.27 | 36.83 | 36.44 | 38.26 | 11.61 | 39.20 | R-FCN | 79.33 | 44.26 | 36.58 | 53.53 | 39.38 | 34.15 | 47.29 | 45.66 | 47.74 | 65.84 | 37.92 | 44.23 | 50.64 | 50.64 | 34.90 | 47.24 | CenterFPANet | 88.74 | 71.52 | 48.95 | 52.06 | 48.55 | 73.37 | 61.14 | 90.53 | 57.63 | 84.06 | 66.64 | 62.71 | 73.33 | 57.63 | 42.76 | 65.29 | FPN | 88.70 | 75.10 | 52.60 | 59.20 | 69.40 | 78.80 | 84.50 | 90.60 | 81.30 | 82.60 | 52.50 | 62.10 | 76.60 | 66.30 | 60.10 | 72.00 | ICN | 90.00 | 77.70 | 53.40 | 73.30 | 73.50 | 65.00 | 78.20 | 90.80 | 79.10 | 84.80 | 57.20 | 62.11 | 73.45 | 70.22 | 58.08 | 72.45 | FMSSD | 89.11 | 81.51 | 48.22 | 67.94 | 69.23 | 73.56 | 76.87 | 90.71 | 82.67 | 73.33 | 52.65 | 67.52 | 72.37 | 80.57 | 60.15 | 72.43 | MDCF2Det | 89.68 | 84.39 | 52.12 | 72.71 | 64.49 | 67.07 | 77.45 | 90.11 | 83.98 | 86.01 | 55.03 | 63.33 | 74.45 | 67.74 | 62.87 | 72.86 |
|
Table 6. Accuracy comparison of different algorithms on DOTA dataset