Unlike single geospatial objects extraction, the task of road extraction faces many challenges, including its narrowness, sparsity, diversity, and class imbalance. In order to solve the above problems, this paper proposes a modified convolution neural network with transfer learning (MCNNTL)for road extraction from remote sensing imagery. The techniques of data augmentation, transfer learning, data preprocessing, and backpropagation algorithm are used in order to get better performance. The Massachusetts roads dataset is chosen as the dataset to carry out the experiment of road extraction, and the result shows that this model outperforms traditional methods of road extraction from remote sensing imagery in precision, recall rate and composite accuracy.