基于深度多特征融合的CNNs图像分类算法

A CNNs Image Classification Algorithm Based on Deep Fusion of Multi-Features

近年来,卷积神经网络在图像处理方面的良好性能得到了广泛关注。为了更好地提取图像内容信息,提高图像分类精度,提出了一种基于深度多特征融合的CNNs图像分类算法。算法有效深度融合了图像的多种特征,即使用k-means++聚类算法提取的主颜色特征和利用去噪卷积神经网络提取的空间位置特征。实验结果表明,提出的基于深度多特征融合的CNNs图像分类算法在图像分类方面提供了有竞争力的结果,分类精度比CNN提升了7个百分点。该算法通过深度融合图像的多种特征,可为后续图像处理提供更全面更显著的有用信息。

In recent years, the good performance of convolutional neural network in image processing has been widely concerned. In order to better extract image content information and improve image classification accuracy, a CNNs image classification algorithm based on deep fusion of multi-features is proposed. In this algorithm, the main color features extracted by k-means++clustering algorithm and the spatial position features extracted by the deconvolution neural networks are effectively fused. The experimental results show that the proposed CNNs image classification algorithm based on deep fusion of multi-features provides competitive results in image classification, and the classification accuracy is 7% higher than that of CNN;The algorithm can provide more comprehensive and significant useful information for the subsequent image processing by deeply fusing various features of the image.