This study proposes a comprehensive learning-based technique, DescatterNet, for real-time, non-invasive, incoherent imaging through dynamic scattering media. The technique is designed to overcome the challenges posed by scattering media, such as biological tissues, fog, and turbid water, which can severely degrade image quality due to speckle noise. Extensive experiments demonstrate that DescatterNet outperforms existing methods in terms of image quality, inference speed, and memory consumption. The method is effective in both controlled laboratory settings and outdoor environments, including natural fog. Key contributions include the development of an e-ink display for realistic training dataset simulation, an effective preprocessing method to reduce domain gap, and a robust neural network architecture. The results suggest that DescatterNet holds significant potential for a broad spectrum of imaging applications.This study proposes a comprehensive learning-based technique, DescatterNet, for real-time, non-invasive, incoherent imaging through dynamic scattering media. The technique is designed to overcome the challenges posed by scattering media, such as biological tissues, fog, and turbid water, which can severely degrade image quality due to speckle noise. Extensive experiments demonstrate that DescatterNet outperforms existing methods in terms of image quality, inference speed, and memory consumption. The method is effective in both controlled laboratory settings and outdoor environments, including natural fog. Key contributions include the development of an e-ink display for realistic training dataset simulation, an effective preprocessing method to reduce domain gap, and a robust neural network architecture. The results suggest that DescatterNet holds significant potential for a broad spectrum of imaging applications.