Research On Typical Geo-objects Classification Of Remote Sensing Image Based On HWP-FCN

The classification of remote sensing images is an information guarantee for aircraft navigation.Its classification accuracy directly affects the navigation reliability of the aircraft.However,since the uncertainty and complexity of spatial distribution of the types of geo-objects,diversity of the same geo-object,it is still a tough issue to classify geo-objects in different scenes.A typical geo-objects classification method of remote sensing image based on Fully Convolutional Networks(FCN)is proposed,due to the problem of the SLIC algorithm has the characteristics of low precision and low adaptability while the FCN algorithm can extract and describe the features of typical geo-objects features through training and learning.The algorithm includes two stages: learning and classification.The FCN model is trained with the remote sensing image data set by FCN.In the classification stage,the remote sensing image classification region is classified by the trained FCN model.Experiments show that the method can effectively improve the classification accuracy of typical geo-objects.In this paper,a typical geo-objects classification method of Weighted Penalty-FCN(WP-FCN)remote sensing image is proposed to solve the problem of imbalanced classes in remote sensing image datasets.Experiments show that the method solves the problem of the misclassification of the "small minority classes" such as road and grassland to a certain extent,and improves the accuracy of typical geo-objects classification.In view of the importance of the exact classification of the road for aircraft navigation and guidance,the spatial distribution of typical geo-objects and features of road misclassification in WP-FCN classification results are studied and the method of Hierarchical Weighted Penalty-FCN(HWP-FCN)is proposed for remote sensing image classification in this paper.The experimental results show that compared with WP-FCN and FCN algorithm,HWP-FCN algorithm can effectively improve the classification accuracy of remote sensing image,and the algorithm can adapt to scene classification with small noise and small contrast changes in remote sensing images.