Object Detection for Rare Birds on the Plateau

Abstract

Object detection has always been a hot research direction in the field of computer vision. At present, most methods are supervised learning methods, but this algorithm requires a large amount of image labeled data, which not only takes time to manually label, but also takes a lot of time when training data. In this paper, the improved object detection network based on yolov3 network is studied, due to the fast inference speed of yolov3, high cost performance and strong versatility, the improved object detection network can identify and locate specific class objects by extracting features by algorithms. In order to improve the performance of detection, before training, the labeled pictures of rare birds on the plateau were augmented to expand the data, and attention mechanisms were added to the last three effective output layers of the backbone network. Finally, the experimental results show that the obtained model has a certain improvement in the picture detection effect of rare birds on the plateau.