Authors - N V Bharani Subramanya Kumar, C V Mahesh Reddy, CH. Samyana Reddy, Krishn Chand Kewat, Laxmi Narsimha Talluri, Shaik Mohammed, Rahil Sarfaraz, Sushama Rani Dutta Abstract - This work showcases an improvement over existing methods by developing a novel deep convolutional neural network (CNN) architecture for image classification specifically targeting the images in the CIFAR-10 dataset [4] which consists of 60,000 color images ( 32 x 32 pixels size) divided into 10 classes. So far, the model architecture incorporates a number of convolution and pooling layers which are then followed by the fully connected layers to better learn the complex structure existing within the input spatial configuration. The typical challenge of overfitting is addressed by employing various techniques such as data augmentation and dropout regularization strategy. Immediately from the experimental evidence, it is clear that the deep CNN performs superior to other traditional models in the case of image recognition classifying problems and therefore the model has proved to be robust in discerning the differences that exist in the categories in the images within the CIFAR-10 dataset.
Open Zoom
