Screen content images are a mixture of text content, natural content, etc., widely used in online meetings, remote desktops, and other scenarios. Due to the mixed nature of content, existing screen content coding methods are not efficient enough. When screen content images are compressed at ultra-low bit rates, texture details are severely damaged and text content is almost unreadable. To address this issue, this paper proposes a screen content compression method based on dual-modality for ultra low bitrate transmission. Specifically, we first propose a dual-modal screen content compression framework, where text contents are converted into document modality and the background image after removing the text is preserved as image modality. Then, we use binary encoding to compress the document and employ a downsampling-compression-upsampling process to compress the background image, where the upsampling utilizes our proposed super-resolution network to enhance quality. Finally, experimental results show that compared with state-of-the-art methods, our proposed method could improve the compression efficiency by up to 66.8% on public datasets.