基于深度学习的多模型遥感影像检索

Multi-model Remote Sensing Image Retrieval Based on Deep Learning

遥感数据进行全局哈希检索的性能随着遥感数据量的增多而下降,通过引入无监督整数索引的方法实现对局部最近邻值进行检索,将有望提高遥感数据检索的可扩展性。本文提出了结合有监督哈希编码,无监督整数索引的多模型遥感影像检索方法,从而实现对遥感数据的高精度哈希检索,以及实现对遥感数据检索的可扩展性。另外本文结合Google Earth Engine(GEE)平台,构建了一个遥感影像检索管道,将有助于实现通过各种检索算法对大范围遥感数据中感兴趣区域的定位和识别。

The performance of global hash retrieval of remote sensing data decreases with the increase of remote sensing data volume. It is expected to improve the scalability of remote sensing data retrieval by introducing unsupervised integer index to realize local nearest neighbor value retrieval. In this paper, a multi-model remote sensing image retrieval method combining supervised hash coding and unsupervised integer index is proposed to achieve high precision hash retrieval and scalability of remote sensing data retrieval. In addition, combined with Google Earth Engine(GEE) platform, this paper constructs a remote sensing image retrieval pipeline, which will help to realize the localization and recognition of regions of interest in a large range of remote sensing data through various retrieval algorithms.