ResUpNet is a deep learning–based defense framework designed to protect image classification models from adversarial attacks. It operates by transforming images into the YCbCr color space and selectively removing perturbations from the luminance (Y) channel using a ResNet-based upsampling architecture, while preserving color information.