Deep Convolution Neural Network in Clustering Explanation

This paper analyze the effectiveness of feature extraction of deep convolutional neural networks in image classification from the perspective of the basic statistical pattern recognition method, K-means clustering. This clustering works on those features extracted from each convolution layer of the VGG16 network, as well as the visualization of those features. The visualization graphs show the changes IOR of feature extraction from CNN, while the clustering graphs show the distribution of features extracted by CNN for different class samples. This work provides an interpretive perspective on the statistical learning methods of deep learning, and hope to provide values for readers.