<p>Unmanned aerial vehicle (UAV) imagery-based landslide inventory detection is widely used for hazard assessment and risk management in landslide-prone areas. Previous studies have primarily relied on image processing techniques and deep learning algorithms applied to high-resolution UAV imagery. However, in real-world conditions, UAVs often encounter challenging environmental factors such as low light, overexposure, and cloud cover, which can degrade image quality and significantly reduce the performance of deep learning-based landslide detection. This study proposes a novel approach that integrates pixel segmentation algorithms with generative artificial intelligence (AI) to enhance UAV image-based landslide detection in challenging lighting and atmospheric conditions. Low-quality landslide images were classified using Laplacian variance and brightness thresholds. Generative AI enhanced the quality of these images. Pixel segmentation identified landslide locations within the enhanced images, and these regions were mapped onto the original images. Results indicate that the generative AI approach effectively enhances low-light, overexposed, and cloud-covered images, producing high-quality outputs with strong structural similarity to real images. The structural similarity index values for testing images ranged from 0.577 to 0.921. This approach significantly improved detection accuracy, achieving F1 scores between 0.878 and 0.891 on enhanced images, compared to 0.029 to 0.616 on the original images. These findings contribute to the development of a robust, real-time UAV-based automatic landslide detection and mapping method.</p>