基于深度学习的交通路口自动识别

基于车辆轨迹数据提取道路信息已经成为数字地图更新的一种有效而廉价的方法，引起了广泛的研究。交通路口是道路信息的重要组成，如何快速而准确提取交通路口具有一定的挑战。深度学习的快速发展为解决这一问题提供了一种新的思路。本文针对轨迹数据在交通路口的分布特性，首先提出一种基于LSTM深度学习的转弯轨迹模式自动提取方法，通过统计航向变化规律对路口转弯轨迹进行建模分类，然后采用LSTM深度模型学习轨迹时间序列，从而实现转弯轨迹模式的自动识别，训练好的模型可以对新的未知轨迹数据实现自动快速地提取转弯轨迹。其次，针对因路口转弯点的稀疏性而影响准确定位路口中心这一难点，本文提出一种联合补偿点计算和转弯轨迹航向变化幅度的方法选取路口候选点，然后通过聚类路口候选点识别道路交叉口。为验证和评估该方法的有效性，采用福州市鼓楼区出租车实际轨迹数据进行测试。结果表明，该方法道路交叉口识别准确率达到96.7%，为电子地图实时更新及无人驾驶自动导航应用等提供关键技术支持，同时我们开放该方法源代码和实验数据，以便于其他研究者开展相关研究。     

Recent Advances in Floating Photovoltaic Systems

In recent years, floating photovoltaic (FPV) technologies have gained more importance as a key source of clean energy, particularly in the context of providing sustainable energy to buildings. The rise of land scarcity and the need to reduce carbon emissions have made FPV systems a cost-effective solution for generating electricity. This review article aims to explore the rapidly growing trend of floating PV systems, which can be a practical solution for regions with limited land areas. The article discusses the structure of the PV modules used in FPV plants and key factors that affect site suitability choice. Moreover, the article presents various techniques for cooling and cleaning FPV to keep optimal performance and discusses feasible trends and prospects for the technology. Finally, this paper proposes the potential integration of FPV systems with other technologies to enhance energy generation efficiency and discusses other research aimed at the advancement of the technology. By examining the various features of FPV systems, this review article contributes to understanding the advantages and challenges associated with using this sustainable energy technology in different regional contexts.