IMAGE TERMINAL GUIDANCE BASED ON YOLO V2 FRAMEWORK

The traditional image terminal guidance algorithm can't adapt to the change of target size well and poor in generalization ability. Recently the object detection algorithm develops rapidly in the support of deep learning, the YOLO v2 network got the best detection effect in the field of object detection in 2016. In this paper, we proposed a new approach for image terminal guidance based on YOLO v2 framework (short for G-YOLO). First, we get the training data through the Internet and experiments, then using improved clustering methods and experiments to determine the size of Anchor-box, to get more accuracy. Second, on the basis of trained network using Pascal-VOC dataset, we simplify the network structure by experimenting, to improve speed of detection obviously. Third, we use the multi-scale detection training to adapt to the drastic changes of the target size during terminal guidance. Experiment results show that the proposed method is robust to rapid changes of target size, and the accuracy and stability is superior to traditional methods.