OBJECT-BASED BUILDINGS DETECTION & CHANGE ANALYSIS USING MULTI TEMPORAL HIGH RESOLUTION REMOTE SENSING DATA

With urbanization and urban sprawl, land-use and land-cover change studies became an area of keen interest for researchers. Cause of rapid urbanization is peoples’ responses to economic opportunities, and government policies. New housing societies and business model adopted overall globe has provided opportunities for new land uses.

Image by piviso from Pixabay 

“Image Classification is the process of sorting pixels into a finite number of individual classes, categories of data based on their DN (Digital Number) values”. 

With advent of modern high resolution era of satellite images, PIXEL based, Object based and many other machine learning algorithms have been developed.

Currently the prospects of a new classification concept, Object-Based Classification, are being investigated. Recent studies have proven the accuracy of object-based classification over traditional classifiers. The object-based classification approach relies on basic principle to use other information such as shape, texture and 3D information from DEM (Digital Elevation Model), and ancillary GIS data.

Change detection is defined as detecting spatial and spectral change in an object over a time span probably using temporal data sets. With advent of 21^st century and rapid urbanization, accurate and timely updated information regarding LULC (land use land cover land cover) change and its spatial variations is significant for urban planning and management.

Our research domain for this project is based on Object-Based Classification and post-classification change detection technique.  Data set used for the study was multi-temporal high-spatial resolution imagery acquired by IKONOS sensor. Area of Interest was Lafayette-west from 2005 to 2018. Object-Based Classification techniques including, Multi-resolution segmentation was done on three images, then an object-based fuzzy approach was applied to implement the land cover classification separately for each of the three years i.e. 2005, 2013 & 2018. The object based and fuzzy approach was selected because it provides accurate and better results for change detection by incorporating spatial information and user defined expert knowledge. After rule set definition buildings of AOI were extracted and refined for object based post-classification comparison. Results were graphically and statistically compared for year 2005, 2013, & 2018. E-cognition ArcMap by Esri, and Erdas Imagine by Intergraph were used for our research project.

1.1  Objective

To make use of high resolution remote sensing data by incorporating object-based fuzzy classification and post-classification change detection to detect buildings of Lafayette-west for year 2005, 2013 & 2018.

1.2  Study Area

Lafayette is a city in and the parish seat of Lafayette Parish, Louisiana located along the Vermilion River in the southwestern part of the state. The city of Lafayette is the fourth-largest in the state, with a population of 127,657 according to 2015 U.S. Census estimates, It is the principal city of the Lafayette, Louisiana Metropolitan Statistical Area, with a 2015 estimated population of 490,488. The larger trade area or Combined Statistical Area of Lafayette-Opelousas-Morgan City CSA was 627,146 in 2015. Its nickname is The Hub City.

Figure 1.2 Study Area Zone-1

1.3  Data set used

High Resolution Imagery provided by

·         IKONOS imagery of  2013  and 2018

3D Information extracted using

·         DEM and DSM of study area 2005

Reference Data

·         2005 building footprints

Ancillary Data

·         Address data

·         Zoning Data

·         Roads Data

1.4  Software used

·         Ecognition by Trimble

·         ArcGIS by ESRI

·         Erdas Imagine by Intergraph

·         MS Excel by Microsoft

Image Source: eCognition Image

Image Source: ESRI Image

Image Source: Microsoft 

Image Source: Intergraph 

METHODOLOGY

Data was acquired, based on data quality no digital image processing was required. Image was further treated on ecognition software. Object based Classification was done on 2013 and 2018 imagery. Use of DEM and DSM provided subjected further ease in building detection.

Approach was made to distinguish shadow and buildings using NIR band threshold, Water was classified from shadow class by applying area threshold. Vegetation was classified by applying NDVI threshold. Buildings were then classified on the basis of height threshold and sub classified into child classes by applying area threshold as we have different area ranges in building class. Building class was classified by selecting samples and using nearest neighbor classifier. Most of the area at first was classified into building. Roads were separated from the imagery and building class by applying length threshold, vegetation using the same NDVI approach and child classes of building class was classified from buildings by applying area and length to width ration threshold.

Detailed step by step approach:

1.  Open the ecognition software.

2.  Click on “create new project” option.

3.  Import the “Study area image” from the workspace.

4.  Assign weights to NIR band

5.  Click on the layers and rename each layer. Layer 1 as blue, layer 2 as green, and 3rd as red and 4th NIR.

6.  Go to “Process” option on the top toolbar and open “ process tree” All the operations in ecognition are executed through the “ process tree”

7.  Right click in “process tree” window and open the option “Append new”. Uncheck the “Automatic option” and type the name “Segmentation” and click OK.

8.  The parent class “segmentation” has been added. Now click the ‘segmentation” option and select the “Insert child” option to add sub classes.

9.  Click the first segmentation algorithm” Multi Resolution Segmentation” and select the execute option.

10. Execute  next “Multiresolution, Spectral difference”

Note that Execute Multi Resolution segmentation Scale 70 value -average size of objects Shape (0.4) weight -Geometric form of the objects Compactness (0.6) Value- The higher the value, the more compact image objects maybe.

11.  Make classes in class hierarchy

12.  Create NDVI

13.  Apply threshold

14.  Define other rule set such as NIR, DN value. Length/width ratio etc….

15.  Export Classes

16.  Data Cleansing in ArcMap

RESULTS

High resolution imageries of 2013 and 2018 were classified by object based fuzzy classification and with the aid of building footprint, address point and zoning data and change detection and comparison was done. Results generated are

	Buildings Count
Entities	2005	2013	2018
Total Building	131	203	234
Medical-Related MR	2	2	2
Planned Residential Development	107	180	212
Single Family Residential	6	5	4
Single_Two and Multi Family Residential	16	16	16

Table 3.1 Statistics of Building Detection

Change Detection Maps

Conclusion

Object based fuzzy approach proved to be an efficient accurate and time saving image classification approach. Due to Urbanization buildings are increasing with the passing years. Using fuzzy rules by giving appropriate thresholds the accuracy of the classification increases. Change detection analysis showed growth in planned residential development both in 2013 and 2018. Small growth was also observed in single and multifamily residential area. Availability of very high spatial and temporal resolution remote sensing data facilitates mapping highly complex and diverse urban environments. Project concluded that

·         Incorporating high resolution aerial digital images and elevation data, and it’s processing using object-based image analysis through various rule set for mapping urban land covers and quantifying buildings is no doubt an efficient and accurate technique.

·         The Project presents and demonstrates an approach for formulating an optimal land covers classification rule set over small representative training urban area image, and its subsequent transfer to the same, and different multi-sensor, multi-temporal images.

·         There also was very rigorous selection of the object features for all of the land cover classes and iterative classification and rules updating

·         Shadows missclassified as buildings are corrected with additional segmentation and using the shape and context features re-applied to the buildings class only

·         This process picks up most of the shadow correctly, but if some shadow segments over the roofs still remain, the building can be split into parts

·         Overall, the height information helped capture most of the buildings, which otherwise are very diverse in their spectral and shape characteristics

References

•       K. Leukert, A. Darwish and W. Reinhardt, “Urban Land-cover Classification: An Object-based Perspective,” 2nd Joint Workshop on Remote Sensing and Data Fusion over Urban Areas, Berlin, May 2003.
•       P. Mather, Computer processing of remotely-sensed images, Chichester: Wiley, 1999. [3] B. Jähne, Digitale Bildverarbeitung, Berlin: Springer, 1993.
•       R. Haralick, L. Shapiro, Computer and robot vision, Vol. 1, Reading, Mass.: Addison-Wesley, 1992.
•       Definiens imaging, eCognition user guide [Online]. Available: http://www.definiens-imaging.com/documents/index.htm, April 2003.
•       NIMA, Geospatial Standards and Specifications [Online]. Available: http://164.214.2.59/publications/specs/index.html, April 2003.
•       Lillesand, T., Kiefer, R., “Remote Sensing and Image Interpretation”, New York, USA: John Wiley & Sons, 2000.
•       E. Mendoza, R. Dirzo Deforestation in Lacandonia (southeast Mexico): evidence for the declaration of the northernmost tropical hot-spot
•       Biodiversity and Conservation, 8 (1999), pp. 1621-1641