Research On Recognition And Classification Of Hyperspectral Remote Sensing Images Based On CNN

Hyperspectral remote sensing images have rich spectral and spatial information,so they have been widely used in many fields such as agriculture,industry and military.The recognition and classification of hyperspectral images is an important part of the research of hyperspectral remote sensing technology.Many researchers apply the latest technology in the field of identification and classification of hyperspectral images.In order to obtain an excellent classification effect,a large number of labeled samples are often needed.However,hyperspectral remote sensing images have difficulties such as large spectral dimensions,time-consuming and expensive labeling,resulting in a small amount of labeled data,which greatly affects the classification effect of the image.Based on the method of deep learning,this paper uses convolutional neural network to identify and classify hyperspectral remote sensing images.According to the characteristics of hyperspectral images,the network model is optimized and the parameters are adjusted.Research on the limited samples of hyperspectral image labeling and network degradation,the main work is as follows:（1）Aiming at the problem of insufficient hyperspectral remote sensing image data and data imbalance,this paper introduces virtual samples,and by mixing original samples and virtual samples to increase the amount of hyperspectral remote sensing image data,the imbalance of data is alleviated to a certain extent.Achieved the effect of data enhancement.（2）Aiming at the high dimensionality of hyperspectral image data and the difficulty of feature extraction,this paper introduces the principal component analysis method when preprocessing hyperspectral remote sensing data,and optimizes the band of the image data to achieve data optimization and noise removal.And the purpose of data dimensionality reduction.（3）Aiming at the difficulty of low-dimensional convolutional neural networks in processing high-dimensional data of hyperspectral images,this paper introduces 3D-CNN into the recognition and classification of hyperspectral remote sensing images,and proposes a model based on 3D-CNN.This model breaks through the limitations of traditional algorithms of single feature extraction and insufficient use of spectral information,and ultimately improves the classification accuracy of hyperspectral images.（4）Aiming at the problem of network degradation of convolutional neural networks due to the increase in the number of network layers,this paper uses a three-dimensional residual convolutional neural network,and uses a grouped dilated convolutional layer in the model to replace the original model.On the basis of the convolutional layer,the GHC-3DRes Net model is proposed,which effectively alleviates the problem of network degradation,and can also extract image features well,and finally achieves excellent classification results.