首页 >  , Vol. , Issue () : -

摘要

全文摘要次数: 98 全文下载次数: 197
引用本文:

DOI:

10.11834/jrs.20211042

收稿日期:

2021-01-25

修改日期:

2021-08-05

PDF Free   EndNote   BibTeX
基于Dense RFB和LSTM遥感图像舰船目标检测方法
摘要:

针对当前遥感图像舰船目标检测精度不佳问题, 本文构建舰船目标数据集STAR, 提出基于Dense RFB和LSTM多尺度舰船目标检测算法。 该算法首先在SSD网络基础上设计了浅层特征增强模块, 基于人眼视点图采用Dense RFB特征复用和膨胀卷积增大感受野的尺度和种类, 增强浅层网络对细节特征的提取能力; 其次设计了深层多尺度特征金字塔融合模块, 采用FPN和LSTM思想, 基于反卷积和残差网络对深层不同尺度特征进行融合, 增强网络结构非线性和特征层的表征能力; 最后加入聚焦分类损失函数进行联合训练, 有效避免了正负样本失衡问题。 在光学遥感图像数据集上实验表明, 舰船目标检测精度均值达到81.98%, 检测速度达到 29.6fps。此外, 遥感图像中成像模糊、被遮挡、部分被裁剪等舰船目标的检测效果也优于原有经典算法, 说明该算法对遥感图像舰船目标检测的泛化能力较强, 有效地提高了遥感图像舰船目标检测的精度。

Ship Detection Method in Remote Sensing Image Based on Dense RFB and LSTM
Abstract:

Aiming at the problem of low accuracy performance of commonly used target detection algorithms, the STAR ship data set is constructed, and a novel ship target detection method based on Dense RFB and LSTM is proposed to detect multi-scale ship targets, the proposed method can be divided into three parts. First, it inherits the network structure of Single Shot MultiBox Detector (SSD) and introduces the shallow feature enhancement module, which uses Dense RFB feature reuse and expansion convolution increase the receptive field and enhance the ability of shallow network detail feature extraction. Second, based on the Feature Pyramid Networks (FPN) and Long Short-Term Memory (LSTM), the multi-scale feature pyramid fusion module is designed by introducing the deconvoluion and residual feature fusion module to enhance expression of the network nonlinearityand ability. Finally, the focal loss function is introduced into the training strategy to address the problem of the imbalance of positive and negative samples in the training process. The experiments tested on the detection dataset demonstrate that the proposed method shows the Average Precision (AP) achieves 81.98%, at the speed of 29.6 frames/s. On the extended experiment, the performance of our method is better than others for the detection of fuzzy and occluded targets in remote sensing images. Experimental result shows that the method has the generalization ability for ship target detection, and effectively improves the accuracy of remote sensing image target detection.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群 分享按钮