基于注意力机制和特征融合的SSD目标检测算法

为提高SSD算法对于小目标的检测能力和定位能力,本文提出一种引入注意力机制和特征融合的SSD算法。该算法在原始SSD模型的基础上,通过将全局池化操作作用于高层的不同尺度的特征图上,结合注意力机制筛选出需要保留的信息。为提高对小目标的检测精度,本文引入反卷积和特征融合的方式,提高对小目标的检测能力。通过在PASCAL VOC数据集上的实验表明,该算法有效的提升了对小目标识别的准确率,改善了漏检的情况,大幅度提升了检测精度和算法的鲁棒性。

SSD Target Detection Algorithm Based on Attention Mechanism and Feature Fusion

In order to improve the ability of SSD algorithm to detect and locate small objects,this paper proposes an SSD algorithm that introduces attention mechanism and feature fusion.On the basis of the original SSD model,the algorithm combined with the attention mechanism to screen out the information that needs to be retained by applying the global pooling operation on the feature graphs of different scales at the high level.In order to improve the detection accuracy of small targets,deconvolution and feature fusion are introduced in this paper to improve the detection ability of small targets.Experiments on PASCAL VOC data set show that the algorithm can effectively improve the accuracy of small target recognition,improve the situation of missing detection,and greatly improve the detection accuracy and robustness of the algorithm.