Extraction of built-up areas of Gujranwala District from Landsat-8 OLI data based on spectral-textural information and feature selection using support vector machine method

Project Title

Extraction of built-up areas of Gujranwala District from Landsat-8 OLI data based on spectral-textural information and feature selection using support vector machine method

Project Area of Specialization Artificial IntelligenceProject Summary

Nowadays rapid globalization is a major issue worldwide, cities are expanding day by day and the population is rising significantly. In this case we require some data that can tell us the current situation of a city that how much of that city area is built up. Now this built up data will help us in planning and controlling the future growth and expansion of the city and can also help us manage the required resources of the city. As our study area Gujranwala is the fifth most populated city of Pakistan and has a population growth rate of 3.0%. This growth rate causes expansion of the city and we know that Pakistan is an underdeveloped nation with no surplus supply of resources. So, it is more important in Pakistan to have built up data which we can use to manage our city resources and control and monitor the expansion. Furthermore, Gujranwala is a flood prone area with a population of 2.229 million. So, resources and disaster management is more of a concern. With the built up area information we can forecast and identify the areas that are effected or are going to be affected in future because of natural hazards like flood. So, we can prioritize those areas which are more likely to be affected in terms of rescue and disaster team response that can alternatively help us manage our resources and make our disaster management authorities response better and quicker.

So, knowledge of the built-up area is necessary for various applications, including proper urban planning and development along with coping with flood vulnerability. Landsat 8 OLI will be used for the extraction of the data. Now it is observed from previous studies that using only optical data for built up area extraction is difficult because of the resulting spectral confusion as the built-up class gets mixed with other classes such as river sand, bare land, etc. In this project, a more automated methodology will be used including a combination of spectral and textural features along with feature selection methods to avoid redundancy of information. For this purpose, eight textural features based on the Grey-Level Co-Occurrence Matrix (GLCM) will be selected and combined with multispectral data. By applying feature selection methods, features containing the most information will be chosen. After that, Support Vector Machine (SVM) classifiers will be trained on selected samples by using the chosen optimal features. The results from SVM classifiers will then be compared with k-Nearest Neighbor (k-NN) and Back Propagation-Neural Network (BP-NN). The comparison results will show whether this approach results in an increase in the overall accuracy or not.

Project Objectives

Built up area mapping a key ingredient in the development, planning and management of urban cities. The built up data can enhance our resource and rehabilitation measures in times of disaster and can also help the concern authorities to speed up the rescue responses. Furthermore, built up area mapping can be used for future development of the city and can be used to limit the expansion of the city boundaries as too much expansion of a city causes the rise in slump population around the city perimeter causing a rise in the demand of daily life resources which leads to many social issues and most importantly over population.

We will collect textural feature sample from google earth of Gujranwala city and use it as an input to get spectral features from Landsat 8 OLI. To avoid the redundancy and any unwanted information we will use feature selection techniques. After that, Support Vector Machine (SVM) classifiers will be trained on selected samples by using the chosen optimal features resulting from the feature selection techniques. After training the SVM will give us the final mapping of the built up area of Gujranwala.

The objective of the project was not just to extract BUA but was also to refine and improve the classifier algorithm of Support Vector Machine (SVM) to help us improve the overall accuracy of the model. Also, this study will help to establish a baseline for embarkation of build-up areas in this region that may be used for the flood hazard modeling since our study area (Gujranwala) usually receives high amount of rainfall in monsoon season and is prone to flooding. The built-up area data can help in identifying the affected areas in case of floods. Thus, helping us focus the rehabilitative measures on those areas that are predicted to be more effected in case of any such scenarios.

Project Implementation Method

The proposed method for extraction of built-up areas using Landsat-8 OLI imagery consists of 3 major steps. Initially, textural features will be extracted from the available data and normalization of these textural features will be done along with that of the spectral features. This step of overall process, which also involves feature normalization can be used as a preprocessing step while providing inputs in a classifier for machine learning purposes. After this step, optimal features are found by using techniques like MI (Mutual Information), SFFS (Sequential Forward Floating Selection), and SFS (Sequential Forward Selection). Optimal features can be defined as the features with most information in both terms, quantity wise as well as their usefulness in the Built-up area extraction process. The purpose of selecting and filtering these features is that in the selection of many features, there is a chance that although they may give accurate results but it also causes problems for classifiers like redundancy in data and irrelevant information. So, keeping that in mind a minimum number of features is selected, which gives sufficient information without causing an alteration in the original data, using the feature selection method. After this step, these small numbered but optimal features will be used to extract built-up area through classifiers including SVMs, BP-NN, and k-NN. At the end, for the implementation and preparation of maps, Python libraries and ArcMap 10.5 will be used respectively. The python libraries to be used include gdal/ogr (ver. 1.11.1), scikit-learn (ver. 0.20), matplotlib (ver. 1.4.3), tens or flow (ver. 1.11), etc.

The input data that will be used will be taken from Landsat 8 OLI. The data from satellites is in form of continuous and multiple spectral channels enabling us to analyze vast earth surface areas. For processing that data to get the maps of the study area depicting the Built-up area density in the study area will be generated by using a tool of ArcGIS Software. The tool to be used is ArcMap 10.5. Now for every textural feature the Landsat 8 OLI gave us seven spectral features thus creating redundant, repetitive and unwanted data. So, to remove this redundancy and make our data more accurate Python libraries will also be used in the processing of the data received from Landsat 8 OLI.

Benefits of the Project

Urban land accounts for a small fraction of the Earth's surface area but has a disproportionate influence on its surroundings in terms of mass, energy and resource fluxes. Rapid globalization these days result in a continuous increase in build-up areas (BUA). Therefore, nowadays an essential part of territorial planning and study regarding the effects of land covered areas on our environment is the demarcation of build-up areas. The most important benefit of the built-up area information in that it helps in town planning and plays a vital role in this domain of work. For example, using the knowledge of Built-up area of a certain region, town planners can plan the resources of that town or region according to the portion in the town with dense population as those portions will require more resources. Examples of such scenarios can include deciding the capacity of water supply pipes while designing water supply network for the town. Similarly, the same information can help in designing the sewerage system as well. In addition, the dumping site for the waste disposal can also be selected based on the built-up area data. Another advantage of built-up area data is that it can help identify the areas with more population density. The areas with more population or more built-up area will also have more traffic density as well. Thus, these areas are more prone to traffic accidents. To avoid any such scenario, future road networks can be predicted or planned in order to disperse the traffic flow. Thus, enabling to reduce the traffic density in those areas and eventually reducing the probability of any traffic related mishap. Furthermore, this data can also play an important part in the domain of disaster management. The built-up area data can help in identifying the affected areas in case of any disasters like Earthquakes, and floods etc. Thus, helping us focus the rehabilitee measures on specific areas that are predicted to be more effected in case of any such scenario. Furthermore, hospitals, rescue services, and fire brigade centers can be designed in the areas that are more prone to accidents so that, the response time in case of any causality can be reduced. Moreover, mapping urban land in a timely and accurate manner is indispensable for watershed run-off prediction as well.

Technical Details of Final Deliverable

The technical details of the final deliverable are listed below:

a) Built-up Area Extraction:

The deliverables that are to be achieved as a target at the end of this project include Build-up area information of our study area. Using the Support Vector Machine model, that will be trained initially, we will find the areas with concrete cover in the study area. This data can be utilized for study regarding flood vulnerability of the region as our area of focus for this project is Gujranwala which receives large amount of rainfall in monsoon season. Thus, making it a flood prone area.

b) Final Mapping:

 In addition to this built-up area data, another deliverable is making final maps of the study area. These maps will depict the regions which have concrete cover and will help us differentiate between the built-up and non-built-up area. These maps will be generated by the help of ArcGIS’s software ArcMap 10.5 along with the use of python libraries.

c) Improving Accuracy:

Alongside these results related to built-up region in the study area, we will also get results of statistical results from the analysis of the efficiency of the 3 models used. i.e., Support Vector Machine (SVM), k - Nearest Neighbor (k-NN), and the Back Propagation - Neural Network (BP-NN). These results of the statistical analysis performed will show the increase in the overall accuracy of the output from Support Vector Machine model. This will be due to the improvement and the refinement in the algorithm. In addition to showing the improvement in the overall accuracy, this part will show the mutual result comparison among these classifiers as well.

Final Deliverable of the Project HW/SW integrated systemCore Industry ITOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Sustainable Cities and CommunitiesRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
PC with Graphic Card	Equipment	1	70000	70000
Local Travel	Miscellaneous	1	10000	10000