Latest Updates

Reviewers are welcome to join IRJTAS JOURNAL.
Authors are invited to publish their manuscript.
First edition is in process for publication.



An Enhanced Convolutional NeuralNetwork on Image Classification overCIFAR-10 Dataset

Megha SharmaProf. Amit Ganguli Prof. Ajit Kumar Shrivastava

MTech Scholarsiasharma815@gmail.comM. Tech. Co-ordinatoramitganguli@sistec.ac.inHead of Dept, CSESISTecR.hodcs@sistec.ac.in

ABSTRACT

Visual beholding is of great importance for humans to interact with each other and thus the wildlife. We possess a huge ability of visual recognition as we'll almost effortlessly recognize objects encountered in ourlife like animals, faces, and food. Especially, humans can easily recognize an object albeit it's getting to vary in position, scale, pose, and illumination. Such ability is known as core beholding , and is run through the ventral stream within the human sensory system. Yet, since pastfew years, machine performance has been dramatically improved due to the reemergence of convolutional neural networks (CNN) and deep learning, and thus even surpasses human performance. Like customary neuralorganizations, which are enlivened by natural neural frameworks, the design of CNNs for viewing is feed forward and comprises of a few layers during a various leveled way. Particularly, some works reveal hierarchicalcorrespondence between CNN layers and other people within the human beholding system. beholding performance of deep neural networks is typicallymeasured on datasets commonly utilized within the sector like CIFAR100, and CIFAR10. In this dissertation we have taken convolution neural network with different layers configuration and apply on CIFAR10 dataset and we found that our modified CNN model takes more time but perform better as compared toTraditional CNN.

Keywords : artificial neural networks; cifar-10;classification; image; convolutional neural networks; keras; python;jupyternotebook;machine learning.