Abstract: In order to effectively extract and integrate multi-scale information with different depths in salient object detection, the salient object region detection model, M-IRNet, based on multi-scale interactive residual module was designed. A multi-scale interactive residual structure module, IRM, was built to extract multi-scale feature information, to suppress noise interference, to obtain detail information in low-level feature and deep context information. For effective information fusion, a modified bi-directional propagation strategy was adopted, which can effectively fuse context information of different depths, and at the same time, the semantic information of the upper level regulates the information of the lower level to realize the location and detection of the significant object of the image. The experimental results on five benchmark datasets show that the designed model of image salient target region detection has certain advantages, especially on the PASCAL-S datasets the MAE drops to 0.092, and on the DUT-OMROM datasets, the F-measure drops to 0.763. According to extracting more effective multi-scale features to increase the ratio of shallow and deep features, not only the ability to represent the detailed features for salient object is improved, but also the ability to locate salient object is increased. It is applied to the image salient target detection model, which can improve the detection accuracy.