Performance Comparision of VGG-16 and ResNet-34 Algorithms for Supervised Classification of Landsat Images

The rapid change in urban landscape across the globe demands accurate landscape feature extraction techniques. Remote sensing images obtain terrain information which can be used to extract various urban features of land use and land cover to identify the changes. These images enable us to perform feature extraction and classification. Several powerful and successful mechanisms have been proposed to perform supervised satellite image categorization of remotely sensed satellite multispectral data. In supervised satellite image classification, machine learning algorithms have become quite popular due to their good accuracy.This paper aims to study the performance of the Convolution Neural Network's VGG-16 and ResNet-34 algorithms for identifying Land Use Land Cover Change (LULCC) using supervised classification.Based on the remotely sensed multispectral Landsat 5 and Landsat 8 satellite images the LULCC is found by performing supervised image classification on the study region which is Mumbai Metropolitan Region (MMR) for the years 2000, 2010 and 2020. Four classes as Water, Built-up, Vegetation and Barrenland were identified to perform classification.The overall accuracy, the producer's accuracy, the user's accuracy, and standard kappa coefficient were also calculated. The training loss, validation loss and training accuracy were also studied. The various LULCC thematic maps and accuracy assessment results were obtained for the year 2000, 2010 and 2020.