Research And Implementation Of Remote Sensing Image Classification System Based On Deep Learning

Hyperspectral remote sensing is an important research field of remote sensing science and technology.The classification and recognition of ground cover is the first step in the analysis of hyperspectral image(HSI).Hyperspectral image classification is to determine the category labels of each pixel in hyperspectral image.Deep learning has been widely applied to the field of HSI classification and has achieved great success.However,deep learning relies on a large amount of data for training.It is difficult to obtain HSI images and the cost of sample labeling is high.If sufficient training samples are missing,the performance of traditional deep learning methods is poor,and the sparse number of samples will also make the model overfitting problem more prominent.Aiming at the problem of poor performance of deep learning in the sparse number of HSI samples,tHSI paper designs a CNN-based model,and combines the model with meta-learning.The main work includes:1)A CNN-based dual-channel network model is designed.The model has two convolution channels,one channel is responsible for extracting the spatialspectral features of HSI,and the other channel is responsible for extracting HSI spectral features.Then the two channel features are fused,and the proportion of the two channels feature fusion is adjusted by adding the shaping layer,which enhances the spectral feature expression and improves the classification performance of the model in the case of sparse samples.2)Combining the proposed dual-channel network model with metalearning algorithm MAML,the proposed network model is trained by MAML,and the trained initial parameters are obtained.Then the parameters are applied to the new data set to explore the classification performance of the proposed model when the samples are more spares.3)Based on the dual-channel network model proposed in tHSI paper,a HSI classification system based on deep learning is designed and implemented.The system only needs few training samples to realize the classification of HSI data.Finally,the performance of the proposed model is tested on three datasets.The results show that the proposed model can get better classification performance when the samples are scarce.And the function test of the classification system based on the proposed model is completed.Users can use the system to realize HSI classification.