基于改进YOLO V5s模型的遥感图像目标检测及应用

Application of Improved YOLO V5s Model for Regional Poverty Assessment using Remote Sensing Image Target Detection

利用改进YOLO V5s模型实现遥感图像目标检测并用于地域贫困评估。针对现有模型提出了三点改进:加强PAN结构、基于bounding box的RIOU_Loss回归损失函数、协同注意力机制。同时将遥感图像目标作为表征,计算连续时间节点内的贫困率变化。实验结果表明,改进模型的P、R、mAP@0.5、mAP@0.5:0.95值存在不同程度的提升,而Loss值有所下降。因此,与原模型相比,改进模型具备更精准的目标检测能力。同时,与传统的统计数据方法相比,改进模型为地域贫困评估提供了一种等效的无数据评估思路。

This study aims at applying the improved YOLO V5s model for the assessment of regional poverty using remote sensing image target detection.For this purpose,three improvements were made to the model.So,a new enhanced PAN structure was proposed.Accordingly,a new RIOU_Loss regression loss function of bounding box was proposed.Furthermore,a new collaborative attention mechanism was put forward.In addition,while objects in the remote sensing images were used as the representations of poverty status,the changes in the images were considered to evaluate the regional poverty rate in a continuous time interval.The results show that the values of P,R,mAP@0.5 and mAP@0.5:0.95 of the model are improved,while the Loss value is decreased.Therefore,compared with the original model,the improved model has more accurate object detection capabilities.Meanwhile,compared with traditional statistical data methods,the improved model provides an equivalent dataless evaluation approach for regional poverty assessment.