基于高分辨率遥感图像的道路车辆定位研究  

Research on Road Vehicle Location Based on High Resolution Remote Sensing Image

作　　者：王嘉怡 田瑶 徐昇[1] WANG Jiayi;TIAN Yao;XU Sheng(College of Information Science and Technology,Nanjing Forestry University,Nanjing 210037)

机构地区：[1]南京林业大学信息科学技术学院,南京210037

出　　处：《计算机与数字工程》2025年第3期795-802,共8页Computer & Digital Engineering

基　　金：国家自然科学基金青年科学基金项目(编号:62102184);江苏省自然科学基金青年科学基金项目(编号:BK20200784);中国博士后科学基金面上项目(编号:2019M661852);江苏省高等学校大学生创新创业训练计划项目“针对遥感图像的改进生成对抗网络超分辨率重建”(编号:202210298042Z);南京林业大学大学生实践创新训练项目“基于深度学习的树木图像识别技术研究”(编号:2021NFUSPITP0264)资助。

摘　　要：随着卫星发射技术的不断进步与成熟,高分辨率遥感图像技术随之不断进步,遥感图像数据的获取也变得愈发便捷,从遥感图像中提取到的道路车辆的特征也愈发清晰。基于高分辨率遥感图像对道路车辆定位可以快速识别相对广阔范围内车辆的数量,是智能交通控制和解决交通突发状况的重要手段。文章根据南京市玄武区紫金山区域的高分辨率遥感图像,基于TensorFlow机器学习框架,使用SSD目标检测算法对道路车辆的定位展开研究。考虑到SSD网络框架涉及的参数量较大,用Mobilenet轻量级卷积神经网络替换SSD网络框架中使用的部分VGG网络。采用深度可分离卷积,在大幅度减少所需参数量的同时,缩短获取结果的时间。最后对实验结果进行多指标定量和定性分析后表明,该方法在光照条件良好且视野开阔的区域的道路车辆定位检测准确率高,光照、植被、建筑物干扰时道路车辆定位检测准确率较高,实验结果较YOLO V2和YOLO V3有优势。With the continuous progress and maturity of satellite launch technology,high-resolution remote sensing image technology has also continued to improve,the acquisition of remote sensing image data has also become easier and more convenient,then the features of road vehicles extracted from remote sensing images are becoming clearer.Road vehicle positioning based on high-resolution remote sensing images can quickly identify the number of vehicles in a relatively wide range.This paper provides an important solution for intelligent traffic control or traffic emergencies.Based on the provided remote sensing images of Xuanwu district,under the framework of TensorFlow based machine learning,this paper uses the SSD object detection algorithm to conduct research on the localization of road vehicles.However,considering the large amount of parameters involved in the network structure when using the SSD network framework,part of the VGG network designed and used in the SSD network framework is replaced by the depthwise separable convolution,which greatly reduces the amount of parameters required by the algorithm and saves the time required to obtain the result.Finally,the multi-index analysis of the experimental results shows that the proposed method has high accuracy in the area with good illumination conditions and wide vision.The accuracy of road vehicle positioning detection is slightly lower under the interference of light,vegetation and buildings.The experimental results are superior to those of YOLO V2 and YOLO V3.

关 键 词：遥感技术 深度学习 卷积神经网络 目标检测 图像处理 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]