ABSTRACT
Landslide identification poses a significant challenge in ensuring the safety of vulnerable regions. Accurate detection is crucial for timely mitigation efforts. In this study, we propose a convolutional neural network (CNN) model based on transfer learning for classifying landslide-prone areas using a diverse dataset. The dataset comprises satellite images of landscapes categorized into distinct classes. Addressing class imbalance, we employ preprocessing techniques and oversampling methods. The images are resized to a standardized 32 × 32 pixel format to enhance model efficiency. The model leverages a pre-trained CNN architecture and incorporates additional layers for fine-tuning. Training utilizes the Adam optimizer and a suitable loss function. These strategies are vital for optimizing the model’s performance and ensuring precise classification of landslide-prone areas. Evaluation is conducted based on accuracy metrics, showcasing the model’s proficiency in capturing essential features of landscapes prone to landslides. Our proposed approach holds promise for geologists and environmental experts, offering a high-accuracy solution for identifying landslide-prone regions and facilitating effective mitigation strategies (MDPI, 2023)
