Research On High-resolution Remote Sensing Image Classification Based On Multi-feature

High-resolution remote sensing images can display ground information in more detail.As a result,high-resolution remote sensing images are widely used in the fields of geological prospecting,precision agriculture,urban planning,land survey and disaster analysis.At the same time,the requirements for the accuracy and efficiency of image classification are also increasing.With the improvement of the spatial resolution of remote sensing images,the spectral variance of similar objects increases and the spectral variance of different objects decreases.Only the use of spectral feature information cannot effectively distinguish different objects with similar spectral characteristics.Compared with the underlying feature information such as texture,spectrum and spatial structure,the deep feature information extracted by the deep learning model has stronger feature representation ability and can show better separability in the feature space.Deep learning plays its unique advantages in the fields of image classification and target recognition,which can automatically learn deep image feature information and make classification decisions.In this paper,I use some low-level feature information such as texture,color and shape,combining with deep feature information extracted by deep learning model,and use open data set as experimental data to study the influence of different features on image classification results,and the main research results are as follows:(1)In view of the fact that high-resolution remote sensing image features information is highly detailed and cannot make full use of the rich feature information of image data,this paper proposes a classification method based on the fusion of multiple features and convolutional neural network.By constructing a two channel full convolutional neural network(T-FCN),the remote sensing image data and feature information of texture,color,shape are both input into the model for training,prediction and classification accuracy evaluation.The influence of different features on image classification results is analyzed from qualitative and quantitative aspects,and the optimal feature combination is obtained.Finally,Using the public dataset Gaofen Image Dataset(GID)as experimental data and the classification results of the proposed method are compared with those of U-net and Seg Net methods.The experimental results show that the proposed method is better,the overall accuracy is improved by 1.91% and 3.73% and the kappa coefficient is improved by 0.0212 and0.0512 compared with U-net and Seg Net methods.(2)In order to verify the applicability and migration of this method,the classification method proposed in this paper is applied to Changchun City.The GF-2remote sensing image data of Changchun City is selected as the experimental data,and the sample dataset is constructed by preprocessing and data enhancement to train and predict the model.Compared with the u-net and Seg Net methods,the experimental results show that the overall accuracy and kappa coefficient of the proposed method are higher than those of the u-net and Seg Net methods,the overall accuracy is improved by 4.2% and 3.82%,and the kappa coefficient is improved by0.0576 and 0.0534.