Real-time continuous feature extraction in large size satellite images

Highlights

• Real-time continuous feature extraction.

• Remote sensing.

• Image processing.

• Satellite image.

Abstract

Remotely sensed imagery is being increasingly used for the development of the earth observation satellites to investigate human activities, to monitor environmental changes and to update existing geospatial data. The ordinary pictures are difficult to process automatically by computers but can be easily interpreted by humans. The most significant step is how to get anticipated information from the images and how to convert these images into useful data for further studies. The key objective is to satisfy an algorithm claiming to be efficient in large size image processing includ enhanced processing efficiency, finding correlation among data, and extracting continuous features. To achieve these objectives in the setting mentioned above, we propose a real-time approach for continuous feature extraction and detection in remote sensory earth observatory satellite images to find rivers, roads, and main highways. Deep analysis is made on the ENVISAT satellite missions datasets and based on this analysis the algorithm is proposed using statistical measurements, RepTree machine learning classifier, and Euclidean distance. The system is developed using Hadoop ecosystem to improve the efficiency of the system. The designed system consists of various steps including collection, filtration, load balancing, processing, merging, and interpretation. The system is implemented on Apache Hadoop system using MapReduce programming with higher efficiency results in a massive volume of satellite ASAR/ ENVISAT mission datasets.

Introduction

Digital image processing is becoming a hot topic these days because of its various applications in security, medical healthcare, agriculture, entertainment and fun, area monitoring, etc. Digital image processing is the use of computer algorithms on digital images to perform image processing. This technology is widely used for the image morphology, feature extraction, segmentation, rendering, and pattern recognition [1], [2], [3], [4] and many other digital image operations. Various research also works on image processing aspect of H264/AVC [5], [6], [7], [8], [9] such as, in edge detection, deblocking filter, and motion estimation in H264/AVC. Moreover, Feature extraction is the most widely used part of the image processing that can be used for many application such as, security and authentication, object detection, and pattern matching, etc. In practice, two types of feature extraction (feature selection) methods are used, i.e., type I and type II. Type I feature extraction methods mainly focus on the finding of original parameters from the scratch for feature extraction while type II feature extraction method is used to optimize the accuracy of a feature set by removing inconsistent features [10] by given set of features. Also, Type II also used to discover a subset of features associated with optimal identification accuracy [11]. Simpson et al. do well at this in the article Genetic & Evolutionary Type II feature extraction for periocular-based biometric recognition [12].

Remotely sensed imagery is being increasingly used for the development of the earth observation satellites to investigate human activities, to monitor environmental changes and to update existing geospatial data [1], [13]. The ordinary pictures are difficult to process automatically by computers but can be easily interpreted by humans. The most significant step is how to get anticipated information from the images and how to convert these images into useful data for different further studies. Moroever, the processing of larger size images or large datasets of thousands of satalite images in an efficient manner is also a key challenge [14].

The continuous features extraction such as roads, river, and highways detection through satellite image is very valuable and efficient for most of the urban planning application. Very few work has been done in the field of continuous natured feature extraction using satellite image processing. The painted lane markings that exist in the most urban roads, in campus sites or in the comparable environments of the theme parks, industrial estates and science parks may not be easily discernible by closed-circuit television (CCTV) cameras because of bad weather conditions, poor lighting and insufficient maintenance. Similar is the case with the river as well. The existence of pavements or curbs is the important feature of roads or rivers on either side defining the boundaries. For the implementation of autonomous navigation or driver assistance systems, the curbs that are parallel to the roads can be harnessed to extort useful features of the roads.

Due to the fact that the use of vision image data is a difficult task for the extraction of the curbs or features of the road edge as curbs are not perceptible in the vision image. Favorable and heuristic lighting and extensive image processing requires to extract the curbs from the camera image. A laser range measurement system is one of the favorable for obstacle detection and depth range measurement under poor lighting, bad weather condition with its best features of the low cast of an alternative to millimeter wave radar system. The significant rise has been observed in the use of laser range measurement system for an autonomous navigation task in the past several years [15], [16], [17], [18], [19], [20], [21], [22]. However, the major domain of their use has been in indoor environments [17], [18], [19], [20], [21], [22]. Laser range measurements systems have found some of the common tasks of obstacle detection [15], [16], map building [21], [22], navigation [17], [18], and localization [19], [20].

However, to keep the properties of rivers in mind, they are long in length and geometrically smooth. These particular attributes can give advantages to most algorithm to construct a river network. River finding, river tracking, and river linking are three typical stages of river extraction. To search the potential river pixels, this methodology is set with in a river window. When creating consecutive river points, the local properties like magnitude and direction are accounted. Continuous and smooth groups of river seeds are linked together to produce different lengths of segments once the river points are found. Finally, the river segments with longer length are selected as a piece of river in the river linkage stage to form a river network.

Therefore, based on the aforementioend needs, this paper presents an efficient mechanism that detects the continuous features in the images (such as river) using statistical computations, Euclidean distance, and machine learning approaches. To gain the more efficiency of the system, the system is implemented on the parallel environment of Hadoop server. The Hadoop has distributed file system, i.e., HDFS and distributed programming language MapReduce, which have the capability to process large size and a large amount of images using parallel tasking on the same dataset. Moreover, the proposed system divides the whole process into various steps to increase the efficiency of the detection mechanism, which includes collection and filtration, segmentation, processing, and merging.

The rest of the paper describes the background and related work in Section 2. Section 3 demonstrates the details of the datasets used for analysis and tested. Section 4 presented the analysis and discussion based on which the proposed system is developed. The proposed system details are given in Section 5. While the evaluation is done in Section 6. Finally, the conclusion is made in Section 7.