With the rapid development of network in the large image, image hashing algorithm has attracted interests as an approach of approximate nearest neighbor algorithm in the image retrieval system. In this paper, we proposed the deep hash which based on deep learning models . The high dimensional global are extracted by deep convolutional neural network, then using stack autoencoder to get the parameters of the models by unsupervised learning to get the binary hash code. Finally using the hamming distance to compute the similarity of the images. The deephash proves the better results in image retrieval.