Hyperspectral Image Classification Using CNN12
Mr. Ruhikesh Dhande · Bioscience Biotechnology Research Communications · 2020
Convolutional neural networks (CNNs) have showed their dominance in hyperspectral image classification (HIS) where spectral embedded with spatial features have provided means for excellent database training when the application is image classification.Different scaling with context of number of layers in efficient neural networks has been the recent topic of research where attention aided and not aided machine learning algorithm have being the elementary topic of vigilance.We have in the paper highlighted an attention seeking CNN 12 layer algorithm that is utilized for efficient labeling the prime aspect of image classification.The raw image provided as input to the network extracts the relevant features in a way that joining feature map created provides better results in terms of accuracy.The frame work created utilizes the available sets of images and compares the parameters that will help in making the algorithm robust and fine tuned with respect to required hyper parameters.The linear transformation that is the main advantage of CNN architecture has made the system more reliable unlike the rule based feature extraction and classification frameworks.In this paper the three dimensional input utilization helps us solve the overfitting problem that is a byproduct of exhaustive layering during training.The way the three dimensional input is handled in terms of CNN 12 layer convoluting is the simple nonlinear function that is the key aspect of this research paper to improve accuracy.