Semantic Segmentation Method Of Remote Sensing Image And Its Application

Semantic segmentation of remote sensing image is of great importance in the field of image understanding.It has drawn extensive attention both from academy and industry for its wide range of applications,such as urban planning,urban change detection and geographic information system.Nevertheless,many complicated factors,such as complex background,shadows,objects with various scales,topological shapes and appearances in different regions,make this task quite challenging.Traditional image segmentation methods mainly based on the underlying features usually fail to meet the requirements in real-world applications.In general,these methods lack of robustness to complex environments and are of low effectiveness in recognition.In recent years,the deep learning methods have made great progresses in the task of semantic segmentation for natural images.However,it is difficult to obtain satisfactory performance by directly applying them to remote sensing images.Therefore,we have carried out a series of research works on the deep learning based semantic segmentation and its application in precipitation nowcasting.Our main contributions are sununarized as follows.This paper proposes a deep convolutional neural network model with multi-scale information fusion for semantic segmentation of remote sensing images.The structure of our model is composed of two parts:encoder and decoder.In the encoder part,a strategy is proposed to fuse multi-scale feature based on DenseNet network.Specifically,global average pooling is first used to extract regional semantic information of different sub-regions to make network understand complex background in remote sensing images;sub-region global average pooling and multiscale convolution are then used to deal with complex background areas.In the decoder part,we design a shorter decoder which can fuse features from different levels of convolution to accurately restore the image details.For the overall model construction,we design a hierarchical monitoring mechanism with multiple outputs.This trick allows our model to obtain supervised information from different levels,which can help guide the training of the network.Extensive experiments on ISPRS benchmark datasets and Beijing remote sensing dataset demonstrate the effectiveness of our approach.A novel method based on semantic segmentation is proposed to make multi-period structured precipitation forecasting.The network consists of two parts,multi-source coupling network and multi-sequence information fusion network.In the first part,a multi-level and multi-scale feature fusion strategy is designed to extract features from temperature,relative humidity and air velocity.In the second part,the ConvLSTM model was used to correlate historical data with current data in the timeline.During training,several regularization methods such as Dropout are used to alleviate overfitting caused by excessive parameters.Experiments on the ECMWF dataset show the validity of the proposed multi-period structured precipitation prediction model.