As a significant application in RS images, road detection is still a challenging task due to the presence of complex surroundings and multiple false objects. To achieve a satisfying result, a road detection method based on residual learning and saliency sampling is developed in this paper. First, a multistrapdown module is designed with double residual learning blocks that have low computational complexity and time consumption. Second, to improve the classification accuracy and learning ability of the method, a saliency sampling set is established by fusing brightness, orientation and texture maps. The sampling set is imported and merged into the model through the pooling layer. Finally, the cross entropy loss function is carried out in a Softmax classifier. Extensive experiments show that our proposed integrated method is much better than the state-of-the art methods in detection accuracy.