SRS-Net: Training object detectors from scratch for remote sensing images without pretraining

作     者：Haining WANG Yang LI Yuqiang FANG Yurong LIAO Bitao JIANG Xitao ZHANG Shuyan NI Haining WANG;Yang LI;Yuqiang FANG;Yurong LIAO;Bitao JIANG;Xitao ZHANG;Shuyan NI

作者机构：Space Engineering UniversityBeijing 101416China Beijing Institute of Remote Sensing InformationBeijing 100192China 

出 版 物：《Chinese Journal of Aeronautics》 (中国航空学报（英文版）)

年 卷 期：2023年第36卷第8期

页      面：269-283页

核心收录：

学科分类：08[工学] 0825[工学-航空宇航科学与技术] 

基　　金：supported by the Natural Science Foundation of China(No.61906213) 

主　　题：Denseconnection Object detection Pretraining Remote sensing image Trainfrom scratch 

摘      要：Most of the current object detection algorithms use pretrained models that are trained on ImageNet and then fine-tuned in the network,which can achieve good performance in terms of general object ***,in the field of remote sensing image object detection,as pretrained models are significantly different from remote sensing data,it is meaningful to explore a train-fromscratch technique for remote sensing *** paper proposes an object detection framework trained from scratch,SRS-Net,and describes the design of a densely connected backbone network to provide integrated hidden layer supervision for the convolution ***,two necessary improvement principles are proposed:studying the role of normalization in the network structure,and improving data augmentation methods for remote sensing *** evaluate the proposed framework,we performed many ablation experiments on the DIOR,DOTA,and AS *** results show that whether using the improved backbone network,the normalization method or training data enhancement strategy,the performance of the object detection network trained from scratch *** principles compensate for the lack of pretrained ***,we found that SRS-Net could achieve similar to or slightly better performance than baseline methods,and surpassed most advanced general detectors.