ADSSD: Improved Single-Shot Detector with Attention Mechanism and Dilated Convolution

The detection of small objects is easily affected by background information, and a lack context information makes difficult. Therefore, object has become an extremely challenging task. Based on the above problems, we proposed Single-Shot MultiBox Detector with attention mechanism dilated convolution (ADSSD). In module, strengthened connection between in space channels while using cross-layer connections to accelerate training. multi-branch combined three expansion convolutions different ratios obtain multi-scale used hierarchical feature fusion reduce gridding effect. results show that PASCAL VOC2007 VOC2012 datasets, our 300 × input ADSSD model reaches 78.4% mAP 76.1% mAP. outperform those SSD other advanced detectors; effect some significantly improved. Moreover, performance factors such as dense occlusion better than traditional SSD.