Navigating Landscapes through AI: A Comparative Study of EfficientNet and MobileNetV2 in Image Classification

This study presents a comparative analysis of two state-of-the-art deep learning models, EfficientNet, and MobileNetV2, fine-tuned for the task of intel image classification into four categories: urban areas, trails, outdoors & seas. The aim was to rate and contrast these models' accuracy, precision, recall, and F1 scores to identify which will be appropriate for those who are analyzing images to classify geographical and natural environments. In this study, the datasets, which were extracted from an online repository known as Kaggle, were used to train both models and they were evaluated using the structured methodology that entails data preprocessing, model training including hyperparameter optimization and performance evaluation. The outcome projected that EfficientNet was better than MobileNetV2 in all the metrics, providing the highest accuracy but with a balanced precision-recall and f1-scores criteria. Nevertheless, MobileNetV2's computational efficiency instead of resource limitations represents a promising possibility and viability of its application in these settings. This research gives a contribution to the literature being a development of knowledge encompassing abilities and failures of the models applied to the work on classification of intel images, showing the significance of a model selection based on the project peculiarities and computational resource availability. This study individually contributes to improving the deeply learned models derived for environmental and urban landscape analysis, but also creates a guide to future ambitious research and practical work in the same field.