Deep CNN for Classification of Image Contents

Abstract

In recent years the classification of images has made great progress and has been used in many fields. However, it may not be possible to classify images perfectly through the CNN because of overfitting and gradient vanishing. Most existing CNNs have too many parameters, as a result, it will take a long time to train the CNN and then to classify images. In this paper, an improved CNN, with fewer parameters, can perfectly solve the problems such as overfitting, gradient vanishing was developed. The number of designed CNN's parameters is 13M, less than that of other CNNs. In order to check the performance of the designed CNN, the database such as MNIST and CIFAR-10 were used to test the CNNs. The test result was 99.467% and 91.167% respectively. These results are similar to test accuracy of other existing CNNs. Therefore, it was confirmed that the designed CNN not only has fewer parameters than the other CNNs but also shows high test accuracy.