基于视觉注意机制的遥感图像显著性目标检测

Saliency remote sensing image object detection model based on visual attention mechanism

显著性目标检测是遥感图像处理的重要研究领域,传统的方法通过逐个像素点的计算来实现目标检测,难以满足遥感图像大面积实时处理的要求。将视觉注意机制应用到遥感图像的显著性目标检测中,在训练阶段,将所有的目标融合成目标类,所有的背景融合成背景类,目标类的显著性均值与背景类的显著性均值的比值得到一个权重向量;在检测阶段,所有的特征图乘以权重向量得到自顶向下的显著性图;自顶向下和自底向上的显著性图融合生成全局显著性图。实验结果表明当目标和背景不是总成对出现时,该方法的检测结果优于Navalpakkam模型和Frintrop模型的检测结果。

Saliency object detection is one of the important research fields in remote sensing image processing. The traditional methods achieve object detection by calculating the point-by-pixel, which is difficult to meet the requirement of large-scale real-time remote sensing image processing. This paper proposes a remote sensing image object detection model based on visual attention mechanism. In the training phase, all objects are fused into an object class and all backgrounds are fused into a background class. Weight vector is calculated as the ratio of the mean object class saliency and the mean background class saliency for all the features. In the detection phase, all feature maps are combined into a top-down saliency map multiplied by the weight vector, then top-down and bottom-up saliency maps are fused into a global saliency map.Experimental results indicate that when the object and background do not always appear in pairs, this method is excellent to Navalpakkam＇s model and Frintrop＇s model.