The use of RGBD images for obtaining high-accuracy 6D pose estimation results has gained popularity due to the availability of inexpensive RGB-D sensors. Typically, state-of-the-art methods employ deep learning-based methods to process RGBD data for pose estimation. They either use different backbones to extract features from each type of image or combine preprocessed depth images with RGB images before feeding them into a network. However, with increased network layers, deep networks tend to lose detail, which may be crucial for object localization. Additionally, the influence of occlusion also becomes more pronounced, posing challenges in effectively focusing on target objects. To address these issues, we propose RDGE-6D, an intuitive and effective method that leverages reverse direction fusion to enhance RGB deep features using shallow features from depth images. Extensive experiments on benchmark datasets demonstrate the superior performance of our method, especially in scenarios with heavy occlusion.