Evaluation of Machine Learning Frameworks: VGG16, InceptionV3, ResNet50, and Custom CNN for Martian Surface Classification

Feature classification of the Mars surface is very crucial in planetary science as well as in missions and disentanglement of the history behind the Red Planet. In this paper, research will be done to detect the most effective and stable deep learning model for classifying Mars surface. In this experiment the models used, such as ResNet-50, VGG-16, Convolutional Neural Network CNN and Inception V3 were used to classify Mars surface from a dataset of high-resolution Mars Orbital Images classified into: Bright Dune, Crater, Dark Dune, Impact Ejecta, Slope Streak, Spider, and Swiss Cheese.The findings revealed that all four models performed extremely well, with InceptionV3 emerging as the most accurate model. It outperformed the others while maintaining high precision, recall, and F1 scores across all classes. ResNet-50 demonstrated strong training stability with minimal performance variations, while the custom CNN showcased an effective problem-specific approach. VGG-16 also provided competitive results, making all the models viable options for Mars surface classification tasks.