In this paper we have built a deep neural network by using this we have achieved a accuracy of 98.61% of image recognition. In this we have explained the techniques of applying deep learning algorithm at every step.  With this progression of the deep learning Image and speech recognition is possible with good accuracy results.  It is possible due to the Keras Sequential classifier model and load the data set by preprocessing image data set from directory. We have implemented the data augmentation by the keras layers experimental preprocessing. We have included the others layers of the same also and run on the GPU. Firstly, we achieved the 60% accuracy on validating the data sets. Then we applied Dropout to a layer in random manner for we have set the activation to zero and with output units at training time. Results of the application time of dropping out 10%,20%, 30% randomly on this layer of application.[1][2]

Key Words CNN, Keras, Convolution, GPU, MNSIT