Fig 1.
The JLU UGV.
Fig 2.
The hunting process of grey wolves.
Fig 3.
The general structure of BRPN.
Fig 4.
The network of the synthesized feature sampling strategy.
Fig 5.
The enhanced pooling strategy.
Fig 6.
The framework of the EPN.
Fig 7.
The relationship between the GA and the parameters.
Fig 8.
The process of SVM parameter optimization based on the GA.
Fig 9.
The graph of with different values of η.
Fig 10.
The relationship between the GWO and loss function coefficients.
Fig 11.
Flowchart of the parameter optimization process with GWO.
Table 1.
Dataset information.
Table 2.
GA parameters.
Table 3.
GWO parameters.
Table 4.
BRPN parameters.
Table 5.
Detection results on the PASCAL VOC 2007 test set.
Table 6.
Detection results on PASCAL VOC 2012 test set.
Fig 12.
Small-object detection results on the PASCAL VOC 2007 and VOC 2012 test sets.
Fig 13.
Recall versus IoU threshold on the VOC 2007 test set.
Left: 150 region proposals. Middle: 450 region proposals. Right: 850 region proposals.
Fig 14.
ROC curves of the compared methods.
Table 7.
Experimental results over the novel enhanced pooling network.
Fig 15.
The optimization process for the different coefficients.
Table 8.
Runtime data for the different coefficients.
Table 9.
Experiment results of the training error.
Table 10.
Experimental results over the novel loss function optimized by the GWO.
Table 11.
Comparison between the softmax, SVM and GA-SVM classifiers.
Table 12.
Detection results on the KITTI dataset.
Table 13.
Detection frame rate of the different methods on the PASCAL VOC 2007 test set.