基于多特征字典学习的害虫图像自动分类方法

AUTOMATIC CLASSIFICATION METHOD FOR PEST IMAGE BASED ON MULTI-FEATURE DICTIONARY LEARNING

为提高农田害虫图像识别分类的准确率,提出一种基于多特征字典学习的害虫图像自动分类方法。首先,利用监督字典学习的方式,对每一类害虫图像构建多特征过完备字典。为进一步增强计算机在复杂情况下对害虫图像的辨识能力,应用构造的过完备字典对害虫图像进行多特征稀疏表示。最后,通过最小化害虫图像的重构误差实现自动分类。实验结果表明,与其他方法相比,该方法提高了害虫图像识别的准确率。

An automatic classification method for pest image based on multi-feature dictionary learning is developed to improve the accuracy of pest classification. With supervised dictionary learning, each pest image can be constructed as multi-feature overcomplete dictionary. In order to further enhance the abilities of identifying pest image in complex back- ground,multi-feature of pest image are sparsely represented using the constructed overcomplete dictionary. Finally, the classification of pest is achieved by minimizing reconstruction error. Experimental results show that the proposed method performs well on the classification of insect species, and outperforms several state-of-the-art methods in insect categoriza- tion.