Landslides are a critical natural hazard, which poses significant risk to human lives, infrastructure, and the environment. Precise detection and mapping of landslides are crucial for disaster management and mitigation. A deep learning-based approach is presented using satellite imagery, utilizing ResNet18 for feature extraction, autoencoders for anomaly detection, and attention mechanisms to focus on relevant information. The dataset comprises both landslide and non-landslide images, including some metadata. The models were assessed for their usefulness in landslide detection using accuracy, precision, and F1-score. The proposed approach outperforms the existing approaches with the best accuracy at 93.03%, followed by ResNet18 at 90.15%, and the model using an attention mechanism at 88.5%.