论文部分内容阅读
针对传统的Co-training和Tri-training协同训练算法中基分类器独立性低、迭代过程中误差累积和整体泛化性能低的问题,将多视图理论、编码理论和万有引力公式引入协同训练分类算法中,提出了改进算法,算法有效地防止了迭代过程中的误差累积,同时提高了分类系统的泛化性能。在高光谱图像分类实验中,随机地从数据集中抽取5%、10%和20%样本作为已标记训练集时,码字匹配的协同训练分类算法对比Co-training和Tri-training算法,在分类精度上平均分别提高了12.38%和6.13%,在Kappa系数上平均分别提高了0.2和0.07。进一步加入引力筛选机制,对比Co-training和Tri-training算法,在分类精度上平均分别提高了21.30%和10.99%,在Kappa系数上平均分别提高了0.26和0.13,结果表明了本文算法的有效性。
Aiming at the problem of low independence of base classifier, error accumulation in iterative process and low overall generalization performance in traditional Co-training and Tri-training collaborative training algorithms, multi-view theory, coding theory and gravitation formula are introduced into collaborative training classification algorithm , An improved algorithm is proposed. The algorithm can effectively prevent the error accumulation in the iterative process and improve the generalization performance of the classification system. In the experiment of hyperspectral image classification, when we randomly select 5%, 10% and 20% samples from the dataset as the labeled training set, the co-training classification algorithm of codeword matching is compared with Co-training and Tri-training algorithms, On average, the accuracy increased by 12.38% and 6.13%, respectively, with an average increase of 0.2 and 0.07 respectively over the Kappa coefficient. Compared with Co-training and Tri-training algorithm, the classification accuracy was improved by 21.30% and 10.99%, respectively. Compared with Co-training and Tri-training, the average Kappa coefficient increased by 0.26 and 0.13, respectively. The results show that the proposed algorithm is effective .