A Dilated CNN Model for Image Classification

Xinyu Lei,Hongguang Pan,Xiangdong Huang

doi:10.1109/access.2019.2927169

Abstract

The dilated convolution algorithm, which is widely used for image segmentation, is applied in the image classification field in this paper. In many traditional image classification algorithms, convolution neural network (CNN) plays an important role. However, the classical CNN has the problem of consuming too much computing resources. To solve this problem, first, this paper proposed a dilated CNN model which is built through replacing the convolution kernels of traditional CNN by the dilated convolution kernels, and then, the dilated CNN model is tested on the Mnist handwritten digital recognition data set. Second, to solve the detail loss problem in the dilated CNN model, the hybrid dilated CNN (HDC) is built by stacking dilated convolution kernels with different dilation rates, and then the HDC model is tested on the wide-band remote sensing image data set of earth’s terrain. The results show that under the same environment, compared with the traditional CNN model, the dilated CNN model reduces the training time by 12.99% and improves the training accuracy by 2.86% averagely, compared with the dilated CNN model, the HDC model reduces the training time by 2.02% and improves the training and testing accuracy by 14.15% and 15.35% averagely. Therefore, the dilated CNN and HDC model proposed in this paper can significantly improve the image classification performance.

Highlights

Image classification is one of the most basically and widely used field in computer vision [1]
THE DESIGN AND TESTING FOR DILATED convolution neural network (CNN) MODEL we mainly focus on the structure design of dilated CNN model, and to verify the effectiveness, the performance of dilated CNN and the traditional CNN is compared under the same circumstances
THE DESIGN AND TESTING FOR hybrid dilated CNN (HDC) MODEL we mainly focus on the design and performance testing of HDC model, and to highlight the effectiveness of the HDC model, the traditional CNN model and dilated CNN model are compared with the HDC model

Summary

INTRODUCTION

Image classification is one of the most basically and widely used field in computer vision [1]. The dilated convolution is used as an optimization method to expand the receptive field and obtain more information, it is not a main method to extract image features and other algorithms are needed to improve the model performance. Kudo et al proposed a dilated CNN model for image classification and object localization, which replaced the traditional convolutional kernels in ResNet with dilated convolutional kernels and verified on the ImageNet data set. The dilated CNN model is proposed through replacing the convolution layers in traditional CNN by the dilated convolution layers, and expands the receptive field without increasing parameters, so that it can improve the network performance without increasing the network complexity. The HDC model proposed in this paper is simple in structure and easy to be repeated, and retains pooling operations to further enhance the performance, the reliability of the proposed model is higher and the accuracy is kept at a high level

CNN AND DILATED CNN

THE DILATED CNN MODEL DESIGN

THE DESIGN AND TESTING FOR HDC MODEL

Findings

CONCLUSIONS