论文部分内容阅读
在毫米波距离像识别过程中,选择用于训练分类器的样本点的策略往往非常简单,比如随机选取。这些简单的策略无法充分利用已知样本的信息,使得分类器的性能差别很大。针对这一问题,将局部线性重构的主动学习算法(LLRAL)引入到训练样本的选择过程中。算法从全局样本中选择信息量最大的样本,并且在选取的过程中,借鉴了流形学习中局部重构的思想,加入局部线性重构的约束条件,使得选出的样本不仅在全局范围内具有最大的信息量,同时能够保持样本原有的局部结构。使用k-NN和SVM对不同的样本选择算法进行了仿真实验。仿真实验结果表明,在使用相同的分类算法时,该算法不仅性能稳定,而且具有更高的正确识别率。
In the millimeter wave distance image recognition process, the strategy of selecting the sample points used to train the classifier is often very simple, such as random selection. These simple strategies do not make the best use of the information of known samples, making the performance of classifiers very different. In response to this problem, the local linear reconstruction active learning algorithm (LLRAL) is introduced into the training sample selection process. The algorithm chooses the samples with the largest amount of information from the global samples, and draws lessons from the idea of local reconstruction in manifold learning and adds the constraints of local linear reconstruction in the process of selection, making the selected samples not only in the global scope With the largest amount of information, while maintaining the original local structure of the sample. Different sample selection algorithms are simulated using k-NN and SVM. The simulation results show that the algorithm not only has stable performance but also has a higher correct recognition rate when using the same classification algorithm.