A Robust Approach for Multi Spectral Image Classification

Remote sensing images reveal valuable and sensitive information. The applications of satellite image processing are uncountable. Without classification of any satellite image, it is not possible to interpret into useful information. Different classification methods exist in optical remote sensing, microwave remote sensing, hyper-spectral, lidar, etc. Classification models like support vector machine method, k-means method, k-nearest neighbor method, maximum likely hood method, neural network method, fuzzy logic method, random forest method, etc. are used models in the satellite image classification. The number of classes in the optical image processing or multispectral image processing depends on the usage of the data, software used to simulate the data, processor used to classify the data, etc. The validation process is mandatory to crosscheck the classified result with ground points. More number of ground points gives a better result of a classification. Validation of classified image with ground truth says about the relevance of the points that leads to form a confusion matrix.

This paper presents an adaptive methodology for image classification.