This paper presents a novel deep learning approach for rain detection and removal from a single image, even in heavy rain and rain streak accumulation. The method introduces a new rain image model and a deep learning architecture that jointly detect and remove rain. The model includes a rain-streak binary map to locate rain regions, and captures rain streak accumulation and overlapping rain streaks. A recurrent rain detection and removal network is proposed to iteratively remove rain streaks and clear up rain accumulation. The network uses a contextualized dilated network to extract regional contextual information and improve rain detection. The method is evaluated on real images and outperforms state-of-the-art methods in terms of PSNR and SSIM. The method is also effective in joint rain and dehazing tasks, where it first removes rain streaks and then dehazes the image. The method is implemented on both CPU and GPU, with the GPU version being computationally efficient. The results show that the method significantly improves visibility and preserves details in rain-affected images. The code and data sets will be publicly available.This paper presents a novel deep learning approach for rain detection and removal from a single image, even in heavy rain and rain streak accumulation. The method introduces a new rain image model and a deep learning architecture that jointly detect and remove rain. The model includes a rain-streak binary map to locate rain regions, and captures rain streak accumulation and overlapping rain streaks. A recurrent rain detection and removal network is proposed to iteratively remove rain streaks and clear up rain accumulation. The network uses a contextualized dilated network to extract regional contextual information and improve rain detection. The method is evaluated on real images and outperforms state-of-the-art methods in terms of PSNR and SSIM. The method is also effective in joint rain and dehazing tasks, where it first removes rain streaks and then dehazes the image. The method is implemented on both CPU and GPU, with the GPU version being computationally efficient. The results show that the method significantly improves visibility and preserves details in rain-affected images. The code and data sets will be publicly available.