Object detection in remote sensing images based on region mask contrastive distillation
To address the challenges of background interference and dense target distribution in remote sensing images,a target detection method based on region mask contrastive distillation was implemented.The method aims to improve target detection performance in remote sensing images.Initially,detailed feature masks were created by applying masking operations to specific target regions,distinguishing foreground from background and capturing intricate target textures.Subsequently,the contrastive distillation algorithm was utilized,facilita-ting a comparison-based learning approach between the region masks of teacher and student networks.This ap-proach allowed the student network to comprehensively absorb the teacher network's knowledge related to de-tecting target feature textures.Concurrently,a rotational positioning loss function was introduced during the detection phase.This algorithm estimated loss by measuring grid vector differences between ground truth and predicted bounding boxes,thus reducing rotational disparities between predicted and ground truth bounding bo-xes.The results demonstrate that the mean average precision of the improved algorithm is 3.57％and 5.22％higher than that of the traditional algorithm on the DOTA and HRSC2016 datasets,respectively.
remote sensing imageobject detectionregion mask contrastive distillationrotational positio-ning loss
万方数据
维普
