Remote sensing image classification is a specific application of the pattern recognition technology in the remote sensing field. This study proposes an atrous convolution model based on encoder-decoder (DeepLab-v3+) for performing remote sensing image classification with respect to the inaccurate edge classification and low classification accuracy problems encountered while processing remote sensing image classification using ordinary convolutional neural networks. First, the satellite image data are marked, and the DeepLab-v3+ model is trained using a calibration dataset. This model can extract edge features exhibiting considerable robustness from the remote sensing image. Finally, the classification results of the remote sensing image is obtained. When compared with other classification methods, the proposed method achieves higher classification accuracy, more robust edge features, and better classification results when applied on a remote sensing dataset.