Deep learning based CIFAR-10 classification

Cifar-10 is a well-known dataset having a variety of images divided into specific classes for image classification using different models. Among all models of image classification, deep learning methods of image classification have achieved great popularity due to good results, ease of usage, and deep learning of features in shorter time. This paper proposed a Convolution Neural Network (CNN) model of increasing and decreasing sizes, VGG-16, VGG-19, K-Nearest Neighbors (K-NN), and Random Forest (RF) with 2GB GPU operating system memory. The results showed that K-NN and RF obtained least correctly classified images. However, the proposed CNN increasing filter size architecture classified 88% of images accurately, whereas VGG-16 and VGG-16 got accuracy around 61%. The results of experiments were also compared on various benchmarks. Moreover, for future work we will propose improvements in VGG-16 and VGG-19 to get more correctly classified instances.