自适应控制深度学习和知识挖掘图像分类

Adaptive control deep learning and knowledge mining image classification

针对传统分类方法的局限性,提出了一种深度学习结合知识挖掘的零样本图像自适应控制图像分类算法.利用对图像属性的深度学习来实现图像深层次特征及属性的学习和预测,基于图像的属性-类别映射使分类器性能有较大差异,通过稀疏表示模型挖掘图像类别和属性之间的关系并设计自适应控制的属性分类器实现对图像的分类操作.结果表明,与DBN和SVM算法相比,在监督模式和零样本模式下,该算法具有较高的属性预测准确度.在零样本情况下对Shoes数据集进行分类时,该算法具有最高的准确分类识别率,比其他算法的分类识别率提高了15%.

Aiming at the boundedness of traditional classification methods,an adaptive image classification algorithm for zero sample images in combination with both depth learning and knowledge mining was proposed. With the deep learning of image attributes,the learning and forecast of deep-level features and attributes of images were realized. Based on the attribute-class mapping of the images,the classifier had the great performance differences. The relationship between the image categories and attributes was characterized by the sparse representation,and an attribute classifier with adaptive control was designed to realize the classification operation of images. The results showthat compared with both DBN and SVMalgorithms,the proposed algorithm has high attribute prediction accuracy under both supervised mode and zero sample mode. When the Shoes data set was classified under the condition of zero sample,the proposed algorithm has the highest accurate classification recognition rate,which is 15% higher than other algorithms.