Key points are not available for this paper at this time.
In the era of rapid advancements in artificial intelligence, the geospatial field is experiencing transformative changes. Traditional methods for land cover classification and anomaly detection have often been inconsistent and inaccurate, leading to significant real-world issues such as resource misallocation, unnoticed illegal activities like deforestation, unmonitored topographical changes such as unauthorized constructions, unattended forest fires, and border fence crises, all of which exacerbate climate change and urbanization challenges. This study systematically explores various machine learning (ML) techniques and their application to publicly available geospatial datasets. Specifically, it compares selected Convolutional Neural Networks (CNNs) and other ML models on these datasets to evaluate multiple performance metrics and conduct a comparative analysis. While numerous ML models have been previously employed for land cover classification and anomaly detection, this review seeks to enhance performance metrics and improve classification accuracy. Prior studies have employed techniques such as Random Forest on Sentinel-2 data (Gromny et al., 2019), multiple regression approaches on Landsat data (Wu et al., 2016), and Principal Component Analysis (PCA) on OpenStreetMap data (Feldmeyer et al., 2020). Our study introduces the application of advanced models like VGG16, UNet, and Isolation Forest to geospatial datasets, assessing their impact on enhancing land cover classification and anomaly detection. This research not only aims to achieve higher classification accuracy but also contributes to the field by providing insights into the effectiveness of these models and proposing future directions and opportunities.
Kolluru et al. (Tue,) studied this question.