基于Faster R-CNN的西藏牧区牦牛检测算法

YAK DETECTION ALGORITHM BASED ON FASTER R-CNN IN TIBETAN PASTORAL AREA

目前西藏牧区的牦牛养殖在我国畜牧业当中尚处于发展阶段,它的发展很大程度的影响我国牧区畜牧业的经济水平,尤其是对我国西藏牧区为主畜牧养殖业,彻底改革牧区的传统养殖方式非常必要,利用基于图像处理的牦牛目标检测就可以解决传统方式的耗时耗力问题。针对实际的牦牛放养场景,运用深度学习目标分类算法中具有代表性的Faster R-CNN网路结构,联合ImageNet中的牦牛数据集及采集的牦牛样本数据,把场景目标检测转换为目标二分类区别问题,进行牦牛目标检测。通过实际场景的实验结果及数据分析,牧区基于图像的牦牛目标检测方法在检测精度和执行效率上具有良好的检测效果。本文为解决西藏牧区牦牛目标检测提供了新技术,为改革牧区传统养殖方式上提出了新思路。

Yak breeding in Tibet pastoral areas is still in the development stage in China.Its development has a great impact on the economic level of animal husbandry in pastoral areas, especially for the main animal husbandry industry in Tibet.The traditional breeding methods in the pastoral areas is necessary to thoroughly reform.The time-consuming and labor-consuming problems of traditional methods could be solved based on the image processing to detect yak target.Aiming at the actual yak grazing scene, the representative Faster R-CNN network structure was applied, combined with the yak data set in the ImageNet and the collected yak sample data.The scene target detection was converted to the target two-class discrimination problem.Through the experimental results and data analysis of the actual scene, the image-based yak target detection method in pastoral areas had a good effect in the detection accuracy and efficiency.This paper provided a new technology for the yak detection in Tibet pastoral areas, and propose a new idea for reforming the traditional farming methods in pastoral areas.