Significance Object Detection Based on Global Feature Learning

Although significant progress has been made in salient object detection, there are still shortcomings in predicting the detailed edges of salient regions. Specifically, models should predict all areas of salient objects, including some smaller and more irregular regions. To promote the integrity of salient object detection, we designed a new global feature learning network (GFL), which improves in three aspects: first, unlike existing models that focus more on the discriminability of features, we introduce a diversified feature aggregation module (FAM) to aggregate contextual features, increasing the diversity of features. This diversity is the basis for exploring the overall salient object. second, on the basis of the feature aggregation module, we introduce a channel enhancement module (CEM) to enhance the feature maps of prominent overall salient objects while suppressing the feature channels of other scattered objects. finally, after extracting enhanced features, we use a feature refinement module (FRM) to further improve the salient regions, thereby enhancing the micro-level integrity of each salient object. To demonstrate the effectiveness of the proposed method, we conducted comprehensive experiments on multiple challenging datasets. The experimental results show that our method outperforms the baseline method in most metric.