论文部分内容阅读
对于高维特征空间的分类,Adaboost算法是一种有效的分类算法。然而,如果把Adaboost算法直接运用到红外目标的识别,就会面临高噪声下的Adaboost过拟合问题。采用正则化后的Adaboost算法,即AdaboostKL算法作为分类算法的学习模型,以NaiveBayes作为弱学习器,提出了基于正则化Adaboost的红外目标识别算法。正则化的目的是为避免在红外图像特征高噪声下分类器的过拟合,改善了在高噪声数据下目标识别的可靠性。在求取Adaboost的权重分布时,采用的是熵正则化的方法。通过实验,验证了此算法,则即使面对高噪声的红外数据,也能获得较好的识别效果。
For the classification of high-dimensional feature space, Adaboost algorithm is an effective classification algorithm. However, if Adaboost algorithm is applied directly to the recognition of infrared targets, it will face the Adaboost over-fitting problem under high noise. Using Adaboost algorithm, that is AdaboostKL algorithm as the learning model of classification algorithm and NaiveBayes as weak learner, an infrared target recognition algorithm based on regularized Adaboost is proposed. The purpose of regularization is to avoid the over fitting of the classifier under the high noise of infrared image features and to improve the reliability of target recognition under high noise data. In Adaboost weight distribution, we adopt the method of entropy regularization. Experiments show that the algorithm can get a good recognition even in the face of high-noise infrared data.