基于改进Mask RCNN的遥感图像小目标检测算法研究  

Research on Remote Sensing Image Detection Algorithm Based on Improved Mask RCNN

作　　者：张艺博 赵加坤 陈攀 支杨丹 夏星浩 ZHANG Yibo;ZHAO Jiakun;CHEN Pan;ZHI Yangdan;XIA Xinghao(School of Software Engineering,Xi'an Jiaotong University,Xi'an 710049)

机构地区：[1]西安交通大学软件学院,西安710049

出　　处：《计算机与数字工程》2024年第3期880-885,共6页Computer & Digital Engineering

摘　　要：随着航空遥感领域的不断发展,针对该场景下小型目标的检测已经成为目前研究领域中的一项重要工作。论文基于航空遥感图像场景,提出了一种针对航空遥感领域中小目标检测的优化方法。为了提高算法在小目标检测方面的实用性和准确性,论文在Mask RCNN算法的基础上添加了空间注意力机制模块来对图像的背景做降噪处理,使用CIOU作为边界框回归损失函数进行优化,然后使用Kmeans聚类算法代替原始算法生成更加匹配小型目标的检测锚框。改进的Mask RCNN在航空遥感图像数据集下的检测精度达到61.89mAP,检测精度相对于目前主流的遥感图像检测算法R-FCN提升了17%,相对于Mask RCNN提高了2.4%,达到了当前条件下最好的检测效果。With the continuous development of the field of aerial remote sensing,the detection of small targets in this scenario has become an important work in the current research field.Based on aerial remote sensing images,this paper proposes an optimiza-tion method for small target detection in the field of aerial remote sensing.In order to improve the practicability and accuracy of the algorithm in small target detection,this paper adds a spatial attention mechanism module to the Mask RCNN algorithm to reduce the noise of the image background,and uses CIOU as the bounding box regression loss function to optimize.And then it uses the Kmeans clustering algorithm instead of the original algorithm to generate a detection anchor box that more matches the small target.The improved Mask RCNN has a detection accuracy of 61.89mAP under the aerial remote sensing image data set.Compared with the current mainstream remote sensing image detection algorithm R-FCN,the detection accuracy has increased by 17%,and compared with Mask RCNN by 2.4%,reaching the best detection effect under the current conditions.

关 键 词：改进Mask RCNN 航空遥感图像 注意力机制 CIOU 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]