A Review of Researches on Deep Learning in Remote Sensing Application

Abstract

In recent years, deep learning has been widely used in the field of image understanding and made breakthroughs research progress in image understanding. Because remote sensing application and image understanding are inseparable, researchers have carried out a lot of research on the application of deep learning in remote sensing field, and extended the deep learning method to various application fields of remote sensing. This paper summarizes the basic principles of deep learning and its research progress and typical applications in remote sensing, introduces the current main deep learning model and its development history, focuses on the analysis and elaboration of the research status of deep learning in remote sensing image classification, object detection and change detection, and on this basis, summarizes the typical applications and their application effects. Finally, according to the current application of deep learning in remote sensing, the main problems and future development directions are summarized.

Assessment of the Ecological Condition of Soil Cover Based on Remote Sensing Data: Erosional Aspect

“…10), the areas adjacent to most of the sampling points on both slopes of the basin do not differ in color from modal soils, although they are classified as eroded by the value of H. Thus, the continuous allocation of areas characterized by a "feathery" image structure as superficially eroded, is justified. The spectral samples collection of eroded lands, created during the research, will be further used to test the method of "computer vision" [21] for automated decoding of space images.…”

Deep Learning in Forest Structural Parameter Estimation Using Airborne LiDAR Data

“…It conducts complex tasks by automatically learning representations from raw input data and across multi-layer neural networks that the backpropagation algorithm is applied to automatically optimize internal parameters [37], [38]. Comparing with traditional machine learning, deep learning is a completely data-driven algorithm to generate the best ways for feature extractions [39]. It exclusively learned complex hierarchical structures from RS dataset [40], rather than relied on a pre-designed, specific algorithm [37].…”

Landslide Detection Using Densely Connected Convolutional Networks and Environmental Conditions

“…Remote sensing has the characteristics of being fast, macroscale, and all-weather, and can quickly obtain macrosurface information [10]. At the same time, the improvement of remote sensing image time, spectrum and spatial resolution provides a robust database for remote sensing applications [11,12]. Utilizing remote sensing technology to monitor and govern landslides has the advantages of economy and speed.…”

“…Currently, the popular machine learning algorithms applied to landslide susceptibility analysis, identification and monitoring are support vector machines (SVMs) [22,23], random forest (RFs) [24], rotation forests [25], and ensemble learning including bagging [26]. Traditionally, remote sensing image interpretation combined with machine learning uses statistical methods such as maximum likelihood and k-means clustering depending on spectral and texture features [11]. These methods are based on pixels.…”

“…In fact, landslides have other attributes (shape, texture, spatial structure, mutual relationship, etc.). In recent years, methods such as object-oriented analysis, artificial neural networks (ANNs), genetic algorithms, and SVMs have achieved good results [11,27]. The object-oriented method fully considers the characteristics of the landslide, which reduces the error of extracting information from only pixels and obtains higher accuracy [28,29].…”