针对现有的语义分割算法存在分割结果空间不一致的问题,提出一种基于加权损失函数的多尺度对抗网络语义分割算法。在DeepLab v3基本框架的基础上,引入Pix2pix网络作为生成对抗网络模型,实现多尺度对抗网络语义分割。同时,为增加模型的泛化能力与训练精度,提出将传统的多分类交叉熵损失函数与生成器输出的内容损失函数和鉴别器输出的对抗损失函数相结合,构建加权损失函数。大量定性定量实验结果表明,该算法能够识别并分割细小的物体,其语义分割性能超过现有的深度网络,在保证语义分割空间一致性的同时提高了分割效率。

With the popularization of electric vehicles, the maintenance of power batteries has gradually become a major research topic in the maintenance of electric vehicles.A battery aging detection method was proposed, based on the real-time conditions of electric vehicles.Changes of ohmic internal resistance of the power battery were observed by monitoring current changes.Furthermore, battery health status was estimated through state of charge(SOC) variation tendency of the electric vehicle, thus providing reference for vehicle maintenance on the part of the users.According to changes of SOC, state of health(SOH) and ohmic internal resistance of the power battery, it could be judged whether the power battery should be replaced or not.