Perceptual Metric Development Dataset

The Perceptual Metric Development Dataset (PMDD) is a publicly available benchmark dataset developed for training and evaluating feature extraction models used in multispectral perceptual quality assessment for remote sensing images. The dataset was specifically designed to support the development of perceptual metrics for six-band multispectral imagery.The dataset was generated using Landsat Collection 2 Level-2 imagery extracted from Google Earth Engine. All images correspond to 2024 and contain six spectral bands (Red, Green, Blue, NIR, SWIR1, and SWIR2) normalized to the range [0, 1]. Each sample consists of a 512 × 512 pixel multispectral image and its corresponding nine-class Land Use/Land Cover reference map obtained from the ESRI Global LULC dataset.The images were collected from Iran, Türkiye, Azerbaijan, Armenia, Georgia, Bulgaria, and Greece to provide geographically diverse samples and a balanced distribution of LULC classes. After preprocessing, the dataset contains 14,848 samples, divided into 12,620 training, 742 validation, and 1,486 testing samples.PMDD is intended for research on multispectral image quality assessment, perceptual metric development, remote sensing image analysis, and deep learning applications.