基于和积网络的花朵图像分类研究

Flower Image Classification Research Based on Sum-Product Networks

通过和积网络的结构学习,将其与新颖的图像特征提取方法相结合,提出了一个新的花朵图像分类方法。所提出的分类方法在考虑了自然场景图像的特点下,利用图像小块的信息进行特征提取,提取到的特征向量具有可辨别性、独立性和鲁棒性;对特征向量构成的矩阵使用和积网络结构学习算法,有利于将相似的实例聚为一类,不同的变量分为不同类。实验结果表明,提出的基于和积网络的花朵图像分类算法有着更理想的分类效果。

This paper proposes a new flower image classification method, throughout studying a new sum-product network structure, and combining it with the novel image feature extraction method. The new classification method, which takes the characteristics of natural scene images into consideration, and uses the information of image patches to extract the feature, makes the extracted feature vectors distinguishable, independent and robust. The use of sum-product structure learning algorithm for matrix is propitious to group similar instances into the same cluster and segment distinct variables into different categories. Experimental results show that the method has a better performance in flower image classification.