Smart Remote Sensing Image Recommendation Mechanism based on Spatial Ranking Indicators

Abstract

Following the rapid evolution of sensor technology and Internet service mechanisms in recent years an increasing number of image platforms or digital collection agencies are providing circulation services related to remote sensing images In various fields of spatial data application remote sensing images (RS-images) have become an indispensable resource for providing information about the continuously changing reality In particular the formation of the concept and structure of big data will contribute to RS-images being used at a greater scope and scale Compared with the spatial data of vector format RS-images are characterized by continuous coverage multiple versions and a high update frequency which can provide users with a more comprehensive and real-time ground truth for the areas of interest (AOIs) In the context of the explosive growth of the circulation environment for RS-images users face the challenge of finding and selecting suitable RS-images from a large number of images under given spatial constraints Similar to how Internet search engines face the problem of responding to large quantities of search information an effective ranking mechanism can considerably reduce the difficulty and burden for users in selecting RS-images Many items included in the ISO metadata standard (ISO 19115-2:2019 Geographic Information Metadata Part 2: Extensions for Acquisition and Processing https://www iso org/obp/ui/#iso:std:iso:19115:-2:en) for RS-images such as spatial temporal resolution spectral and cloud coverage can serve as the basis of ranking and recommendation Spatial aspect is considered the most crucial constraint as it determines if the image may qualify for providing information about the Area of Interests However because image acquisition cannot be customized in accordance with the AOIs such coverage relationships generally differ Thus an effective spatial ranking recommendation mechanism must be capable of presenting and quantifying these degrees of difference to provide users with subsequent access to the physical images’ order thereby resulting in significant time and resource savings As the number of available RS-images grows rapidly the importance of intelligent recommendation mechanisms for RS-image will increase accordingly In this study an innovative location-based RS-image finding engine (LIFE) was formulated Through its built-in image recommendation mechanism the system is capable of filtering ranking and recommending RS-images based on user-designated AOIs thereby making it easier for users to select images The LIFE framework comprises three major components The first component is the cluster-based RS-image index structure whose goal is to store and process the spatial coverage information parsed from RS-image metadata where the spatial index framework then provides a rapid search and filter of the qualified candidate images The second component is the quantitative indicators for spatial ranking and recommendation which includes two indicators that were modeled according to users’ selection behavior The available space (AS) indicator was based on the spatial extensibility of AOIs to the boundaries of RS-images and the image extension (IE) indicator was designed in accordance with consideration of centrality for the AOIs contained by the RS-images The final component is the INDEX indicator that integrates the ranking abilities of AS and IE indicators by means of principal component analysis The INDEX-based mechanism provides a one-dimension ranking and recommendation for RS-image selection considering both characteristics from AS and IE indicators In addition to the quantitative spatial ranking indicators and framework for RS-image recommendation a scoring platform to collect user ratings according to the preferences for RS-image selection was also developed to verify the recommendation results of the developed indicators and mechanism In this study the normalized discounted cumulative gain (NDCG) among other statistics were employed to specifically compare the performance of different indicators and the practicability of the developed mechanism was evaluated by analysis of collected user scoring results The analysis demonstrated that the spatial ranking and recommendation indicators AS and IE can both provide effective recommendation results based on their characteristics However the different ranking results provided by AS and IE suggested each method has its unique advantages and disadvantages Thus ranking by AS indicator may result in a situation of neglecting the characteristics of IE indicator even though such situation may favor certain type of RS-image for example RS-images with large coverage The INDEX indicator from the linear combination of AS and IE indicators can yield a synergistic effect that combines both perspectives and provide better spatial ranking recommendations more consistent with users’ preferences for RS-image data resources of hybrid types or single type Furthermore all experimental results were verified and validated by comprehensive user preference analysis based on collected user rating data it demonstrated the potential of the proposed mechanism in providing a feasible alternative approach in RS-image selection for the public lack of RS-image application expertise and experience

Date of Award	2020
Original language	English
Supervisor	Jung-Hong Hong (Supervisor)