论文部分内容阅读
当前热门图像分类方法大多侧重在分类能力,忽视识别新事物,然而人类认识事物时侧重认识,只在细小之处重视分类,这一点与人类记忆机制密切相关.尽管目前有许多记忆建模理论被相继提出,但大多以单词列表的形式学习,对自然图像列表的研究有限.基于此,本文提出了基于卷积神经网络与Bayesian决策的图像识别分类记忆建模方法,首先利用卷积神经网络提取图像特征,并采用二进制形式存储特征向量;然后进行视觉图像的表达,存储与提取记忆建模,将测试图像特征向量与所有已存储特征向量进行匹配对比,计算似然率值;最后在所有似然率基础上计算测试图像是新类别的几率,若该几率大于某个阈值则判别其为新类别;反之,利用Bayesian决策规则进行.图像分类.在Caltech-101与Caltech-256数据库上的实验表明所提方法能很好地应用于图像识别分类任务中.其击中率比目前代表性的稀疏表达分类(SRC)以及极限学习机(ELM)方法高,且虚报率比其他两种方法低的多.
At present, most of the popular image classification methods focus on the ability to categorize, ignoring the recognition of new things, however, human understanding of things, focusing only on the classification of small things, which is closely related to human memory mechanism.Although there are many memory modeling theory Have been put forward one after another, but most of them are studied in the form of word list, so the research on natural image list is limited.Based on this, a method of image recognition classification memory modeling based on convolutional neural network and Bayesian decision-making is proposed. Firstly, Image features, and the binary form of the eigenvectors stored; and then the visual image representation, storage and extraction memory modeling, the test image feature vectors and all stored eigenvector matching comparison, the value of the likelihood ratio; Finally, all On the basis of the probability, the probability of the test image being a new category is calculated, and if the chance is greater than a certain threshold, it is judged as a new category; on the contrary, Bayesian decision rules are used for image classification. Experiments on the Caltech-101 and Caltech-256 databases It shows that the proposed method can be applied to image recognition classification tasks well. Sparse representation classification of (SRC) and Extreme Learning Machine (ELM) High methods, and false rate lower than the other two methods and more.