ELFGMNet object detection algorithm, improved based on EfficientDet network

To address challenges in detecting dense, occluded, and background-interfered targets, we propose a new detection network called the Efficient Local Filter with Gate Mechanism Detection Network, which improves upon EfficientDet. ELFGMNet enhances feature extraction by replacing simple attention mechanisms with a channel-sensitive mechanism that updates weights more effectively, improving multi-scale feature fusion and feature expression. Additionally, it introduces a lightweight multi-scale gate feature enhancement module for class/box processing, which adapts to interference and occlusion by learning to prioritize important features and suppress noise, thus boosting detection performance.On the VOC2007 dataset, ELFGMNet achieves a best [email protected] of 83.31%, a 2.23% improvement over EfficientDet, and outperforms other advanced models.