Image classification for different land use and land covers using artificial neural network for higher accuracy

Neural network for land classification

Abstract

In the present study, supervised maximum likelihood (ML) classification method has been used to classify images of West Champaran district of Bihar of the years 2010 and 2020 for land use and land cover (LULC) and then the same images have been classified using Artificial Neural Network (ANN) for getting better classified images and higher accuracies.Eight land use and land cover classes are taken in to consideration viz. crop land, fallow land, dense builtup, low builtup, river wetland, lakes ponds wetland, barren land and natural vegetation.Different accuracies such as producer’s, user’s, overall accuracies and the value of kappa coefficients have been calculated for each classified images for both ML and ANN methodologies. LULC classified images are prerequisite for better agricultural planning and getting more crop production with minimum input cost. After that comparative analysis among producer’s accuracy, user’s accuracy, overall accuracy and value of kappa coefficients of the various categories/ classes determined from the classification of the image 2010 and 2020 for both methodologies using confusion matrix. The result shows that crop land, fallow land, wetland, barren land exhibit better producer and user accuracy than dense built-up and low built-up area. It is also found thatoverall accuracies of the year 2010 and 2020 of ML classified images are 87.46% and 88.53% and ANN classified are 92.37% and 93.76%. The value of kappa coefficients of respective years are 0.85 and 0.87 of ML classified whereas 0.90 and 0.92 are for ANN classified images. It is finally observed that overall accuracy and kappa coefficient of ANN classified images are higher than the ML classified images. So ANN is better image classification technique than maximum likelihood technique.