Abstract:Tremendous strides have been made in machine learning, one of the remaining open challenges is to achieve real-time speed as well as to maintain high performance, A convolutional neural network is designed for MNIST handwritten digit recognition, where the first convolutional layer has 32 feature images, the second convolution layer has 64 feature images. Next is the full connection layer, where the first full connection layer has 2048 neurons, and the second has 784 neurons. And the output layer has 10 neurons, which is corresponding to the 10 digit labels. In the experiment TensorFlow is adopted to setup network according to the above technology. When the training of network is completed after 500 epochs, the accuracy of the recognition is 96.1% for training set, and 95.7% for test set.

Design of High Accuracy Detector for MNIST Handwritten Digit Recognition Based on Convolutional Neural Network