Research On Semantic Segmentation Of Remote Sensing Images Based On Multimodal Feature Fusion

In recent years,with the advent of the era of big data,and the continuous improvement of hardware computing level,deep learning has also entered the field of image processing and analysis.With its excellent algorithm performance,it breaks through the limitations of traditional computer vision methods,and opens a new way for remote sensing image processing and analysis.However,due to the problems of different target sizes and shadow occlusion in remote sensing image,the semantic segmentation method based on common image has low neutral performance in remote sensing image.With the rapid development of remote sensing technology,multimodal data such as depth map data has been widely collected,and it can be predicted that additional depth data will improve the accuracy of semantic segmentation.Therefore,it is feasible to use multimodal remote sensing data effectively to improve the semantic segmentation performance of remote sensing image.In this paper,we study the semantic segmentation of remote sensing image based on multimodal feature fusion,mainly including the following aspects:We propose a multimodal network which integrates multimodal information into the framework of semantic segmentation.This algorithm proposes a dual stream network architecture based on encoder decoder.The encoder with two branches extracts features from multimodal images,and fuses them in the network.The attention mechanism of decoder is used to learn the parameters of fusion features.The comprehensive experimental results show that the use of multimodal data sets can improve the accuracy of semantic segmentation of remote sensing images.Then,a semantic segmentation algorithm based on multimodal attention and adaptive feature fusion is proposed.The main idea of the algorithm is to connect the self-adaptive and feature fusion,and then extract the importance of images through multimodal attention,and introduce data enhancement to small-scale multimodal remote sensing data set.The experimental results show that the algorithm is effective.