Scientific and accurate access to the classification of land cover in Qinghai Lake area is of great significance to the study of the ecological environment changes in this region. In this study, we use the 30 meter resolution LandSat 8 OLI remote sensing image data of Qinghai Lake to carry out the related research. The 30 m resolution is of medium resolution. The methods for classification of medium resolution remote sensing image still have defects of difficult feature extraction and low classification accuracy. In this study, using the GoogLeNet inception structure, a Convolutional Neural Network (CNN) model for feature extraction and classification is designed and proposed. We analyzed the effect of the neighborhood window size used for sample generation on the classification results, and compared it with the maximum likelihood classification and SVM classification method. The results show that when the window size is 9×9, the overall classification effect of the CNN model is the best, and the classification results of CNN are obviously better than that of maximum likelihood classification and SVM.