A dual-branch remote sensing object detection model with parameter-free attention

To address the problem of imbalanced category distribution in remote sensing image target detection tasks, an efficient dual-branch remote sensing image object detection method based on simple, parameter-free attention is proposed. The method first designs a sampling strategy based on positive example category weighting. This allows the model to initially learn minority-category features and then gradually transition to majority-category features, thereby increasing its focus on minority-category features. Then, the Simple Parameter-Free Attention mechanism is incorporated into the backbone network to further improve feature extraction for minority categories. Finally, a dual-branch detection output network and an SPPF_ReLU module are designed to enable the model to effectively distinguish between regression and classification subtasks. Simultaneously, an Im_Ghost model was designed to replace the C3 module in the dual-branch detection output structure, thereby reducing the number of parameters and enhancing feature fusion for minority categories. Experimental results show that the model achieves state-of-the-art performance on the highly imbalanced DIOR dataset, but on the NWPU VHR-10 dataset, due to its small sample size and balanced class distribution, performance is not optimal. Comparative experiments with mainstream models demonstrate the superiority of the proposed method. The trained model was applied to disaster remote sensing images, achieving good detection results.