机电工程技术2025，Vol.54Issue(6)：70-77,8.DOI:10.3969/j.issn.1009-9492.2024.00127
基于优化Faster-RCNN遥感影像飞机目标检测算法
Aircraft Target Detection Algorithm Based on Optimized Faster-RCNN Remote Sensing Images
摘要
Abstract
Aiming at the problem of small dataset size of aircraft target detection in remote sensing imagery at this stage,the dataset is expanded by using the data enhancement methods of level-flipping and grayscale transformation,which can,to a certain extent,improve the accuracy of aircraft target detection and alleviate the phenomenon of overfitting.To solve the problem of shallow VGG16 network layers and insufficient feature extraction in Faster-RCNN,ResNet50 is used as the feature extraction network,which can refine deeper and abstract target features,and the residual structure in ResNet50 is beneficial to solve the problems in the case of deepening of the network depth,gradient eruption,and insignificant network performance enhancement.In order to solve the region mismatch problem caused by the two quantization of ROI Pooling,the ROI Align bilinear interpolation method is used to cancel the two quantization operations,obtain more accurate pixel coordinates,and transform the whole feature aggregation process into a continuous operation.The final optimized Faster-RCNN achieves98.72%aircraft target detection accuracy on RSOD dataset,and also has good generalization performance on UCAS-AOD dataset,which verifies the effectiveness of the optimized model.关键词
遥感影像/数据增强/飞机目标检测/Faster-RCNN/深度学习Key words
remote sensing images/data augmentation/aircraft target detection/Faster-RCNN/deep learning分类
信息技术与安全科学引用本文复制引用
刘裕芸,刘春阳,周绍鸿,张永城,王德金..基于优化Faster-RCNN遥感影像飞机目标检测算法[J].机电工程技术,2025,54(6):70-77,8.基金项目
安徽省自然科学基金面上项目(2108085MD130) （2108085MD130）
矿山采动灾害空天地协同监测与预警安徽普通高校重点实验室(安徽理工大学)开放基金资助(KLAHEI202203) （安徽理工大学）
