A New Method of Image Detection for Small Datasets under the Framework of YOLO Network
Guanqing Li, Zhiyong Song, Qiang Fu · 2018
For the image detection problems under small-scale datasets, the detection rate of deep learning method is usually very low. This paper presents a new image detection method based on transfer learning and sample enhancement under the framework of YOLO network. This method takes full advantages of the real-time feature of YOLO, as well as the enhancement of the generalization ability brought by transfer learning and sample enhancements. The detection rate of 87.4% of the 6 targets under the small-scale datasets was achieved, and this method is more than 6 times faster than the Faster R-CNN at the same detection rate. The measured data verified the method. Furthermore, this paper quantitatively analyzes the relationship between sample scale and detection performance under small-scale datasets.