Improved Mask-RCNN remote sensing image building extraction by fusion attention mechanism
This paper proposes an improved Mask-RCNN remote sensing image building extraction method to address the issues of incomplete building extraction,error detections,and missed detections caused by complex image backgrounds and dense building stacking. Using a dual channel attention mechanism to enhance the effective features of the target,and introducing a feature enhancement pyramid network to enhance the network's ability to extract contextual feature information from remote sensing images. Finally,combining the dual channel downsampling module to reduce feature loss and improve the accuracy and efficiency of model extraction. The experiment shows that the improved Mask-RCNN proposed in this paper is compared and validated with multiple methods on the building dataset and RSOD dataset. The Precision and F1 values are higher than the comparison method,and the target recognition results are more complete,with a lower target miss rate.
万方数据
维普
