成都理工大学毕业论文:基于Yolo算法和机器学习的遥感图片识别系统研究遥感图片识别技术研究 作者姓名:苏语稻香 指导老师:版权方要求不公开 相关源代码已经上传GitHub,需要源代码的朋友,请自行访问: 如果您有建议或者问题指正,请联系作者苏语稻香Q鹅:幺九八贰三幺九陆肆三 期待您的来信 摘要 遥感图像识别技术被广泛应用于军事、地理研究等专业领域,以及地图导航、自动驾驶等民用领域。当下,激增的遥感识别需求和庞大的遥感图像数据对遥感识别技术提出快和准的新要求,传统检测技术面临挑战。而快速发展的机器学习技术,又为遥感识别技术的改良提供了新思路。 本文通过机器学习技术配合目标检测算法,实现对遥感图像中指定对象的自动化机器识别。系统采用TensorFlow和Keras框架搭建机器学习部分,采用Yolo-v3算法对遥感图像进行目标检测。系统框架搭建完成后需要进行训练才能正常识别遥感图像。其过程为:使用自主开发的爬虫程序采集遥感图像;对图像中要识别的对象进行人工标记,制作成样本;将样本提供给系统进行有监督学习;最后系统经过迭代训练生成检测遥感图像的模型。当系统对其他遥感图像进行检测时,需要先载入该模型,再利用目标检测算法提取图像特征进行对比,得到遥感图像中指定检测对象的数量以及各个对象的坐标。本文的遥感识别系统检测遥感图片中的桥梁、港口、机场,要求平均查准率不低于30%,平均检测准确率不低于60%,主要进行了以下设计: 开发爬虫程序,采集更多的遥感图像样本来训练系统。开发独立的训练模块,让系统多次进行有监督学习,获取更精确的检测模型。 对Yolo目标检测算法和后端框架进行封装,以模块化的方式优化系统的结构。 最终,系统能够对遥感图像中桥梁、港口、机场三个对象进行检测,%,%。 关键词:遥感图像识别;目标检测;Yolo-v3;机器学习 abstract Remote sensing image recognition technology is widely used in military, geographic research and other professional fields, as well as map navigation, automatic driving and other civil fields. At present, the rapid growth of remote sensing recognition needs and huge remote sensing image data put forward new requirements of fast and accurate to remote sensing recognition technology, and the traditional detection technology is facing challenges. The rapid development of machine learning technology provides new method for the improvement of remote sensing recognition technology. In this paper, machine learning technology combined with target detection algorithm is used to realize automatic machine recognition of specified objects in remote sensing images. The machine learning part of the system is based on tensorflow and keras framework, and uses Yolo-v3 algorithm to detect remote sensing image. After the system is built,in order to recognize remote sensing images ,training is needed. The process is as follows: using self-developed crawler program to collect remote sensing images; The object to be identified in the image is manually marked and made int