中文摘要：

为了提高物体分类性能,提出了一种神经网络池特征分类方法,并结合SIFT特征实现物体的可靠分类。该方法首先提取样本的SIFT特征向量,并从特征向量集合中随机选取样本子集;然后采用径向基神经网络为每一个样本子集构建基元分类器;接着通过重复迭代方式得到许多基元分类器集合,再结合增强技术组建神经网络池;最后采用朴素贝叶斯模型对神经网络池中的各个基元分类器集合的分类结果进行融合,预测特征的最终分类结果。实验结果表明,新方法的运算效率高,对VOC-2007数据集的分类正确率高。

英文摘要：

In order to improve the performance of object classification, a feature classification method based on neural network pool is proposed, to achieve reliable object classification by combing with SIFT features. First, this method extracts SIFT features of sampies, and randomly selects sub-collections from feature vectors of samples. Then, it builds baseclassifiers for every sub-collection by using radial basis function neural network. And then ,it obtains many base-classifiers collections through repeat iteration process, and constructs neural network pool by combining boosting technology. Finally, it uses naive Bayes neural network model to fuse the classification results of each base-classifiers collection in neural network pool, for predicting the final classification results of fea- tures. Experimental results show that, it has high efficiency and high rate of correct classification on VOC-2007 dataset.