Abstract:Under complex sea conditions, the ship detection using remote sensing image is easily affected by sea clutter, thin clouds and islands, which results in low reliability of detection. In this study, an endtoend deep semantic segmentation method is proposed, which combines the deep convolution neural network with the fully connected conditional random field. Based on ResNet architecture, the remote sensing image is roughly segmented by deep convolution neural network. Using the method of Gaussian pairwise and mean field approximation, the conditional random field is established as the output of the recurrent neural network. In this way, the endtoend connection is achieved. On the dataset provided by Google Earth and NWPURESISC45, the comparison between the proposed method and other models is implemented. Experimental results show that the proposed method can improve the accuracy of target detection and the ability of capturing fine details of images. The mean intersection over union is 832%, which has obvious advantage than other models. And it can also run fast, which meets the requirements of ship detection in remote sensing images.