In order to solve the problem that the existing algorithms pay less attention to the interactive selection between features from different sources and the extraction of cross modal features is insufficient,a RGB-D visual saliency detection network based on extracting bi-directional selection dense features is proposed.First,in order to filter out the features that can enhance the saliency areas of RGB images and depth images at the same time,a bi-directional selection module (BSM) is introduced. In order to solve the problem of insufficient cross modal feature extraction,which leads to redundant calculation and low accuracy,a dense extraction module (DEM) is introduced.Finally,the dense features are cascaded and fused through the feature aggregation module (FAM),and the recurrent residual refinement aggregating module (RAM) is combined with the deep supervision to achieve the continuous optimization of the coarse saliency maps,and finally the accurate saliency maps are obtained.Comprehensive experiments on four widely used datasets show that the proposed algorithm is superior to seven existing methods in four key indicators.