基于界标的机器视觉自适应无人管理域算法

Machine visual self-adaptive unmanned management domain algorithm based on landmark

在基础为子空间对齐的域自适应算法框架下,推测源分布与目标分布其间的转化,可用线性函数纠正。对非线性映射函数优化使源和目标2个子空间对齐。结合首次选取的新界标的无人管理域自适应法中两观点,通过非线性投影创造一个空间,在此空间中,两种域变得更近,实现子空间对齐。针对域自适应算法需要所有源案例及目标案例的问题,提出一种基于界标的无人管理域自适应算法。

Under the framework of the domain self-adaptive algorithm whose base is the subspace alignment,the transformation between source distribution(D S)and target distribution(D T)can be corrected by linear function.The nonlinear mapping function is optimized so as to align the two subspaces of the source and target.In combination with the two viewpoints of the first selected new boundaries in the unmanned management domain self-adaptive method,it is possible to create a space by the nonlinear projection.In this space,the two domains become closer and achieve the subspace alignment.In view of the problem that the domain self-adaptive algorithm needs all source cases and target cases,an unmanned management domain self-adaptive algorithm based on landmark boundaries is proposed.