Small target detection based on faster R-CNN

In order to solve the problem of missing detection due to small targets and to improve the accuracy of small target detection, this paper proposes a small target detection algorithm based on Faster R-CNN. In order to overcome the problem of gradient disappearance and gradient explosion caused by the over-deep network, this paper uses the residual network ResNet50 instead of the VGG16 backbone feature extraction network, and additionally uses a soft non-maximum suppression method to improve the recognition rate of overlapping objects. The algorithm was trained and tested on the PASCAL VOC dataset, and experiments comparing various networks showed that the algorithm showed good detection and high accuracy in the presence of local occlusion of the target as well as in the presence of too small targets, with a detection accuracy of 83.26% on the test set, which was on average 8.45% higher than the traditional Faster R-CNN detection results.