基于迁移学习的GOCI超分辨率重建与海洋漂浮藻类探测
GOCI super-resolution reconstruction based on transfer learning and detection of marine floating algae
朱红春 1朱国灿 1李金宇 2张怡宁 1芦智伟 1杨延瑞 1刘海英3
作者信息
- 1. 山东科技大学 测绘与空间信息学院,山东 青岛 266590
- 2. 山东省海洋工程咨询协会,山东 济南 250013
- 3. 山东科技大学 计算机科学与工程学院,山东 青岛 266590
- 折叠
摘要
遥感技术是进行海洋漂浮藻类目标识别与变化监测的重要手段.GOCI遥感卫星影像具有高时间分辨率、低空间分辨率的特点,其低空间分辨率影响了海洋漂浮藻类遥感探测的效果.本研究通过对具有较高空间分辨率的Sentinel-2遥感卫星影像结构特征的迁移学习,应用ESRGAN超分辨率重建技术,将GOCI影像的空间分辨率提升至125 m;在此基础上,构建了基于超分辨率重建GOCI遥感影像的U-Net深度学习语义分割网络,实现了海洋漂浮藻类的较高精度探测.实验结果表明:超分辨率重建的GOCI影像显著提升了影像的空间细节清晰度,基于此实现的海洋漂浮藻类探测结果取得了较高的精度,其中面积相对误差下降了51.87%,F1 值提高了2.41%.本研究是应用GOCI遥感影像进行海洋漂浮藻类高精度探测的一次成功实践,为实现海洋目标的动态精细化监测提供有益的参考.
Abstract
Remote sensing technology is an important means for detecting and monitoring changes in floating algae in the ocean.GOCI remote sensing satellite images have the characteristics of high-temporal and low-spatial resolution.Its low spatial resolution affects the effect of remote sensing detection of marine floating algae.In this paper,through the transfer learning of the structural characteristics of Sentinel-2 remote sensing satellite images with high-spatial resolution,the spatial resolution of the GOCI images were enhanced to 125 m by using ESRGAN super-resolution reconstruction technology.On this basis,a U-Net deep learning image segmentation network based on super-resolution reconstructed GOCI remote sensing images was constructed.This network was used to achieve higher-precision detection of marine floating algae.The experimental results showed that the super-resolution reconstructed GOCI images significantly improved the spatial detail clarity of images and the detection results of marine floating algae achieved high accuracy,with a reduction of 51.87%in the area relative error and an increase of 2.41%in the F1 value.As a successful practice in enhancing the accuracy of detecting marine floating algae targets using GOCI remote sensing images,this study provides a valuable reference for achieving the dynamic and fine-grained monitoring of marine targets.
关键词
GOCI影像/数据融合/超分辨率重建/海洋漂浮藻探测/深度学习Key words
GOCI image/data fusion/super-resolution reconstruction/marine floating algae detection/deep learning引用本文复制引用
基金项目
国家自然科学基金(41971339)
山东科技大学科研创新团队支持计划(2019TDJH103)
出版年
2024
NETL
NSTL
万方数据