Image Classification Using Convolutional Neural Network Based on Feature Selection for Edge Computing

In the field of image classification, convolutional neural networks are widely used because of its efficient feature extraction ability. However, some of the features extracted by convolutional neural networks have insignificant difference among classes, which not only have little contribution to classification, but also cause the increasing of complexity of classifier. And this will cause some difficulties in embedded deployment and application in many fields such as edge computing. Feature selection methods have the ability to select features that are helpful to classification, reduce the dimensions of features and improve the classification accuracy. In view of the existence of labels in training of image classification, and considering the potential role of labels information in feature selection, we propose an image classification method using convolutional neural network based on fuzzy clustering to select features. On the basis of image features extracted by convolution neural network, this method reconstructs the classifier according to the feature selection algorithm by means of fuzzy clustering, so as to select the valuable features to participate in the image classification. Specifically, we divide the features extracted by convolutional neural network from the image of training sets according to labels, and use Fuzzy C-Means to obtain the distribution of each feature on each class. Then we design a measure of the distribution difference between classes according to the cluster centers and membership degree matrices, to select the features with obvious inter-class difference to participate in classification. The experimental results on several image datasets verify the effectiveness of the proposed method.