Research Article
Aircraft Detection for Remote Sensing Images Based on Deep Convolutional Neural Networks
Table 4
The comparison of multiple algorithms on AP for the other three classes.
| Method | Backbone | AP | FPS | Oil tank (%) | Overpass (%) | Playground (%) |
| DConvNet [31] | ResNet-101 | 90.30 | 89.50 | 99.80 | 6.7 | DSSD [32] | ResNet-101 | 72.49 | 72.10 | 83.56 | 6.1 | FFSSD [33] | VGG-16 | 73.24 | 73.17 | 84.08 | 38.2 | ESSD [34] | VGG-16 | 72.94 | 73.61 | 84.27 | 37.3 | DC-SPP-YOLO [35] | Figure 5 in [35] | 73.52 | 74.82 | 84.82 | 33.5 | UAV-YOLO [36] | Figure 1 in [36] | 74.20 | 76.32 | 85.96 | 30.12 | RFN [37] | ResNet-101 | 90.50 | 100.00 | 99.70 | 6.5 | SigNMS [38] | VGG-16 | 90.60 | 87.40 | 99.10 | 6.7 | Improved-YOLOv3 [39] | Figure 4 in [39] | 87.57 | 89.37 | 91.56 | 25.8 | MRFF-YOLO [40] | Figure 5 in [40] | 86.56 | 87.56 | 92.05 | 25.1 | MSRDN-M(Ours) | Figure 11 | 90.68 | 84.12 | 100.00 | 25 |
|
|