Application Of Deep Learning In Semantic Segmentation Of Remote Sensing Images

With the continuous advancement of science and technology,deep learning has developed rapidly in the field of computer vision in the past two years.Neural networks have become an important means in the field of images.The amazing feature extraction capabilities shown in the field of computer vision can adaptively extract the Shallow features and deep features,especially with strong understanding of complex scenes,are very necessary for the semantic segmentation of remote sensing images.This paper sorts out the development and research results in the field of deep learning and semantic segmentation at home and abroad,introduces the basic structure and principles of neural networks,and describes several common network models in the field of semantic segmentation.This paper takes remote sensing image as the research object,uses public data set and performs image enhancement on the data set to obtain the training data set,and then uses TensorFlow to build a Deeplab v3+ model to train the remote sensing image data set and complete the network optimization.The main research work of this paper is as follows:(1)The construction of semantic segmentation data set.This article selects the public remote sensing image,uses batch processing method to crop and rotate the data set,add noise and other image enhancements,enhance the expressive power of the data set,and get suitable for the types of features Data set required for semantic segmentation.(2)Construct the Deeplab v3+ network structure,use the Xception backbone network and hollow space pyramid pooling to complete the construction of the decoder,and then perform the fusion of the low-level and high-level features to achieve the construction of the decoder.Use average pixel accuracy and average intersection ratio as the evaluation index of model accuracy to evaluate the effect of the model.(3)The optimization of the Deeplab v3+ network mainly changed the backbone network,replaced the Xception network with Res Net with a smaller parameter,and then changed the void ratio of the void space pyramid to make the network more suitable for the remote sensing data set of this article,and finally used The accuracy evaluation index indicates the optimization effect.In summary,this article uses Deeplab v3+ to realize the semantic segmentation of remote sensing image features,enhances the required data set,and obtains a more expressive training data set,and then implements the Deeplab v3+ network in the remote sensing image.The application of object types,combined with the effects of the model and existing problems,combined with remote sensing images to optimize the model,so that the model’s accuracy evaluation indicators on remote sensing images have been improved,making Deeplab v3+ network more suitable for remote sensing images.