Research On Target Classification Of Hyperspectral Image Based On Deep Learning

With the advancement and development of the technology era,remote sensing technology has also developed rapidly.Remote sensing image is an important research object for people to obtain ground information.It is also a research hotspot in recent years.Remote sensing images occupy a heavy weight in the field of observation.Today,hyperspectral remote sensing technology has been applied to marine monitoring,military,and land resources.The resolution of hyperspectral images is getting higher and higher,and the images are getting clearer,which makes the information in the images more abundant,but it also brings new challenges,not only increasing the amount of data and the dimension of the data.Large,the requirements of some of the research algorithms are also higher.Hyperspectral image classification is a hot topic in todays research.Its premise of theoretical research and application is to ensure accurate target classification.Its key point is the extraction of features.Hyperspectral imagery is a data with a three-dimensional cube that contains a wealth of spectral information and spatial information.The feature information that was most commonly used to classify hyperspectral images is spectral information.In recent years,the extracted spatial information features have been frequently used for classification,and good classification results have been achieved.However,there are still many problems in the extraction of spatial information features.How to fully extract spatial information features to improve the classification accuracy of hyperspectral images It is still the focus of current research.In recent years,deep learning has received a lot of attention from researchers and researchers.Its powerful data analysis capabilities and feature extraction capabilities have made it popular among many scholars.Deep learning has also been applied to many fields.For the classification of hyperspectral images,in order to extract more useful feature information and improve its classification accuracy,a space-based information extraction algorithm based on mean square error is first proposed.Then,the spatial information features of the image are extracted with this algorithm,and then the spectral information in the image is extracted.The spatial information extracted based on the mean square error algorithm is fused with the extracted spectral information to form spatial spectrum information.Finally,different CNN models and different stack self-encoding neural network models are designed,and the designed models are used to train and classify spectral information,spatial information and spatial spectrum information.The experimental results show that the classification accuracy of the space-based information model based on the stack self-encoding neural network is the highest,and the parameter-adjusted space spectrum-2D in the SVM model,the hyperspectral image classification based on deep learning and the parameter setting research paper.The classification accuracy obtained by the-CNN model is high,which improves the classification accuracy.