P. Swarnalatha - Big Data Analytics for Satellite Image Processing and Remote Sensing

• Dhanasekaran K. Pillai
  Jain College of Engineering, India

ABSTRACT

This chapter focuses on the development of new computational models for remote sensing applications with big data handling method using image data. Furthermore, this chapter presents an overview of the process of developing systems for remote sensing and monitoring. The issues and challenges are presented to discuss various problems related to the handling of image big data in wireless sensor networks that have various real-world applications. Moreover, the possible solutions and future recommendations to address the challenges have been presented and also this chapter includes discussion of emerging trends and a conclusion.

INTRODUCTION

The goal of developing new computational models is to enable creation of new big data based remote sensing infrastructure for analysing and mining image data. The system must include a data collection component to aggregate, integrate data and perform validation of image data. Then, the central component of the system performs tasks like filtering, analysis and extraction of relevant patterns from image data. The result of extraction and prediction can be used for agricultural monitoring, crop monitoring or for forecasting of weather and market values.

Most of the big data framework that uses image remote sensing involves the following steps:

• 1.Organizing and integrating data from scenario based models, satellite images, and other remote stations.
• 2.Developing data mining and correlation analysis techniques to perform time-series data mining, spatial data analysis or spatiotemporal analysis.
• 3.Developing classification methods to perform image data classification.
• 4.Evaluating fitness of the data models by comparing data with standard values or indices.
• 5.Developing methods for monitoring activity using images with low or high resolution.

The system architectural model in Figure 1 involves scenario based models for analysing and mining images, weather data, and pollution data.

Figure 1. Computational system model using different types of data

For data storage and management, DSpace can be used to store and maintain a large amount of heterogeneous data. The DSpace is an open source dynamic digital repository that can be used for image analysis while using big data. It enables free access to the data.

This chapter enables users to understand major issues and problems related to remote sensing in combination with big data handling for image data. After analysing solutions recommended for addressing the problems, users will be able to understand the process of developing a new framework, tools, or software systems to meet the current needs.

BACKGROUND

Mostly, remote sensing data is collected to analyse disease conditions, growth of plant, pollution, land use, road traffic congestions, and effects of disaster etc. One solution to address these problems is to develop possible computational models which represent several modules for the data analysis. The creation of thematic map for certain problems requires meaningful analysis which aims to show satisfactory results.

The analytical solutions to various engineering problems require robust analysis in a particular context. In this perspective, various issues related to scenarios should have to be addressed, because, complexity of different scenarios varies over time. For most of the existing developments, analysing and developing computational models have been the main motivation for remote sensing applications that use images.

The image data collected through multispectral image sensing can provide information at the element level. It can also provide information at the composite level via inter-pixel relationships. In some of the applications, the output information is used to assess the user belief or expert suggestions by the analyst. The software program validates the hypothesis developed by users. In most of the cases, the analysis fails due to the compatibility issues between user-defined performance measure that is used for optimization and the objective that is unlikely to produce the expected results. So, every analysis should need to be applied iteratively. The ordering and optimal selection of the objects involved in analysis may not be known. Hence, an effective approach must use suitable selection and ordering technique for objects in image remote sensing and analysis applications.

Mostly, practical engineering problems cannot be solved with hundred percent perfectness, because, the software system performance largely depends on the error-tolerance, resource availability, and time taken for the process. Engineering solutions require experimentation with iterative analysis of algorithms. The new computational models presented in this chapter will be helpful while proposing a new system to address various problems related to image remote sensing and big data.

Crop related mapping of soybean and corn has been conducted at regional scale focusing on the tropical and temperature plains (Arvor, Jonathan, Meirelles, Dubreuil, & Durieux, 2011). Most of the methods have used spectral features of land cover classes for classification either based on supervised learning methods or based on unsupervised learning methods.

Genetic Programming (GP) uses the principle of natural selection to discover information using software programs (Alavi & Gandomi, 2011). In fact, GP is a specialized version of genetic algorithms in which the encoded individual solutions are software programs rather than binary strings.

GP in some of the recent existing works has focused on the behaviour characterization and some of the studies have used GP as a tool for interpreting remote sensing data and can also be used to analyze ground movement patterns, object movement patterns, and change detection etc.

The crop classes may be identified from phonological information by deriving phonological metrics and by building classification rules based on crop calendar and stage-dependent crop conditions (Dong, Xiao, Kou, Qin, Zhang, & Li, 2015).

The effects of temperature, nutrients, and disease have been predicted by developing an efficient feature classification model (Kuttiyapillai & Rajeswari, 2014). This computational model was focused on information extraction considering tomato related feature analysis for classification using large margin k-nearest neighbour classifier.

A context-based method for extracting food safety information was discussed in a paper titled a method for extracting task-oriented information from biological text sources (Kuttiyapillai & Rajeswari, 2015). It was focused on information extraction and dynamic programming technique to find relevant genes from sequences.

In agricultural field, the detailed information of the planting area is required to improve the yield estimate, so, it can be incorporated into crop yield models (Lobell, Thau, Seifert, Engle, & Little, 2015). To find insights into planting area related and crop related information, image remote sensing that involves computational models would be a viable option.

Another approach based on machine learning has been developed to select and combine feature groups. It allows users to give positive and negative examples. This method improves the user interaction and the quality of queries (Minka, & Picard, 1997). The two methods discussed in the following paragraphs are based on the concepts of information mining.