Research On Algorithm And Application Of Deep Learning Based On Convolution Neural Network

Recently, the Deep learning method with multiple hidden layers was proposed to complete the task of learning as machine learning methods. It is a new step of the machine learning. Deep learning enhances the accuracy of prediction or classification for network models by building a machine learning model with multiple hidden layers and using a large number of training samples to learn more useful features. Convolutional Neural Network is an important network model for Deep learning. It has many characteristics, such as hierarchical structure, weight sharing, regional local sensing, feature extraction and the classification process with a global training and so on. It has been widely used in the image recognition field. Especially, Deep Convolutional Neural Network is the hotspot of current study, which has important value on its application in the study of different recognition task.The development and the results of Deep learning and Convolutional Neural Network in home and abroad was summarized in this paper, and the concept and algorithm of artificial Convolutional Neural Network was briefly introduced. The paper based on the classic convolution neural network model and used handwritten numbers as experimental samples, used two experiments that used the big sample and small sample respectively to find the impact that the depth of the network and the size and number of convolution kernel caused on the system performance. For the number of the sample are few in the practical application, we corrected network activation function and changed the traditional network only using the last stage of the characteristics of the sample to make target recognition, and then we can send all the levels sample features got from the network to the fully connected network to make target recognition. Thus we can build a new Convolutional Neural Network model, which can be applied to the face recognition and the shadowed object recognition. The experiment results show that the model has good recognition.