CLASSIFICATION OF HYPERSPECTRAL IMAGES USING CNN

The resilience, precision, and efficiency of spectral-spatial information-based algorithms have recently attracted increased attention. This work proposes an CNN- based classification algorithm that extracts features while taking into account both spectral and spatial information. The suggested method makes use of CNN to encode pixel’s spectral and spatial data and also employed for classification tasks. Investigated is a clear contrast between the relative gain obtained with the inclusion of spatial features and its spectral counterpart. Three benchmark datasets—Indian Pines, Pavia University, and Salinas—were used in the investigation. The proposed approach outperforms the classification methods K closest neighbours, linear discriminant analysis, Naive Bayes, and decision tree, according to experiments. The classification of hyperspectral images (HSIs) has grown in popularity in the remote sensing community. broadly speaking. Because of their reliability, precision, and efficiency, spectral- spatial information based algorithms are recently receiving greater attention. An CNN-based classification approach that extracts features while taking into account both spectral and spatial information has been developed in this study. The suggested technique uses CNN to encrypt the spectral and spatial information contained in each pixel and is also used to perform a classification task. It is also looked into how well the relative gain obtained by including spatial features compares to its spectral counterpart. The experiment was run using the Indian Pines, Pavia University, and Salinas benchmark datasets. It has been demonstrated through experiments that the suggested approach outperforms the classification techniques K closest neighbours, linear discriminant analysis, Naive Bayes, and decision tree. In the area of remote sensing, categorization of hyperspectral images (HSI) has gained popularity.