Region-based automatic building and forest change detection on Cartosat-1 stereo imagery
Introduction
Cartosat-1 (also called IRS-P5) was launched by the Indian Space Research Organisation (ISRO) in May 2005. The camera system on board the satellite acquires at the same time two 2.5 m spatial resolution stereo panchromatic images using a forward view camera with + 26° viewing angle and a backward view with −5° viewing angle (Kumar et al., 2006). The main purpose of this system is to acquire along-track stereo data for generating DSMs in large areas. It acquires images from the entire earth during a 126 day revisit cycle. The swath width is approximately 30 km (http://www.nrsc.gov.in). These characteristics of Cartosat-1 allow the monitoring of land cover changes for large areas with a relatively high repetition rate. However, Cartosat-1 has not yet been widely used for change detection until now due to the lack of multi-spectral channels.
Using the stereo imagery together with Rational Polynomial Coefficients (RPCs) from the provider, Digital Surface Models (DSMs) can be generated using photogrammetric techniques. The generated DSMs provide height information for land cover analysis. DSMs from Cartosat-1 with 25 m × 25 m lateral resolution were generated by Rao et al. (2007) for updating 1:25,000 or 1:10,000 scale topographic maps, in this research the RPCs were refined by using manually collected Ground Control Points (GCPs). Martha et al. (2010) analysed landslide volumetric changes based on 10 m × 10 m resolution DSMs, which were generated with SAT-PP photogrammetric software from ETH Zürich. With a newly developed dense matching method (Hirschmüller, 2008), DSMs with 5 m × 5 m resolution can be obtained (d’Angelo et al., 2008). Therefore change detection for small forest patches and industrial buildings becomes possible.
Numerous change detection methods using remote sensing with different kinds of images have been proposed by many researchers (Coppin and Bauer, 1996, Mas, 1999, Lu et al., 2004, Berberoglu and Akin, 2009, Blaschke, 2010, Klonus et al., 2011, Chen et al., 2012). The investigations can mainly be divided into pixel-based and region-based methods. In the first case, the change features from two images are compared for each pixel independently. In the second case, the images are segmented into disjoint and homogeneous regions, and then change features are extracted and compared for these regions (objects, segments). Literature research shows that the region-based remote sensing image analysis is attracting more interest approximately since the year 2000 (Blaschke, 2010). Many researches have been performed in comparing region and pixel-based change detection methods for remote sensing data. Walter (2004) used region-based classification results from two dates to generate a land cover change map. The original regions were obtained from existing GIS data, different land cover classes were better separated with region-based features. Desclée et al. (2006) used the region-based approach for forest change detection, the region-based change detection method exhibited a much higher Kappa Index of Agreement (KIA) (KIA = 60%) than the pixel-based method (KIA = 49%) using the same features from multi-spectral satellite images. Im et al. (2008) compared region and pixel-based change detection methods, the results showed that the region-based change detection method could also reach a higher KIA (about 90%) than pixel-based methods (KIA = 80–85%). Duveiller et al. (2008) applied the forest/non forest classification method from Desclée et al. (2006) in region-based deforestation detection from 571 image pairs, where an initial image selection was performed to exclude all bad quality images and image pairs with no forest change. Aguirre-Gutiérrez et al. (2012) compared pixel-based, region-based methods and their combination in land cover classification. A higher accuracy was achieved based on a combined classification method along with a region-based change detection analysis. A brief look at the above comparisons between the pixel-based and region-based change detection reveals that the region-based method performs generally better.
However, all of these papers were focusing on using multi-spectral information. Due to the lack of multispectral information not many tests have been performed on using single channel Cartosat-1 stereo data and no literature has been found for combined evaluation for automatic change detection including the derived DSMs. The research using Cartosat-1 images for automatic and semi-automatic change detection is still limited to large scale topographic monitoring or manual interpretation (Martha et al., 2010, Prabaharan et al., 2010, Kamini et al., 2006). Kamini et al. (2006) chose onscreen digitization to obtain land cover maps from Cartosat-1 data to monitor the land use in frequently flooded areas. Kumar et al. (2007) merged 2.5 m resolution Cartosat-1 data to 5.8 m multi-spectral Resourcesat-1 images in a visual interpretation procedure. Prabaharan et al. (2010) also used visual image interpretation to get land use and land cover (LULC) classification maps in the year 2008 from Cartosat-1 images, and compared them with the land cover maps from images of other sensors.
In this paper, we develop a change detection method using a combination of height changes from DSMs, optimized region boundaries and spectral change from panchromatic images. Firstly, DSMs with 5 m × 5 m resolution are generated with semi-global matching (SGM) from Cartosat-1 stereo imagery as described in Section 2. After orthorectification of the near-nadir images the initial regions are obtained from the segmentation results of these orthorectified images of two dates. The change analysis regions are generated through a combination of these segmentation results. A rule based region merge approach is provided to reach a better consistency between the homogeneous regions of the two dates. Several height and spectral features are then extracted at region level. Additionally to the above mentioned methods, DSMs of the two dates are used in a newly developed combined change detection procedure as described in Section 3. This approach is tested for two different tasks. The first task is forest change detection, performed in a test area in Bavaria, Germany, and the other task is building change detection in an industrial area, where we choose Istanbul, Turkey as our test area since several changes can be monitored.
Section snippets
Input data
Cartosat-1 stereo pairs in Orthokit format are used as input data for the DSM generation. The described process utilizes the Rational Polynomial Coefficients (RPCs) sensor model, which is used to transform the three-dimensional object-space coordinates into two-dimensional image-space coordinates (Grodecki et al., 2004). The RPC provided with the Cartosat-1 Orthokit, exhibit a much lower absolute accuracy than the ground resolution of Cartosat-1 data (2.5 m). Correcting the RPC for high quality
Change detection
A two-step region-based change detection procedure is proposed according to the characteristics of Cartosat-1 images in this paper (as shown in Fig. 1). In the first step, after co-registration of the two DSMs and orthorectified images, segmentation on co-registered Cartosat-1 images is performed to obtain the initial regions. Then the regions from two dates (date1 and date2) are combined to get an initial segmentation map. To cover the over segmentation produced from the region combination, a
Study area and data
Six study sites are chosen for the experiments. Three of them are located in forest areas in Bavaria, Germany, and the remaining three study sites are located in an industrial area in Istanbul, Turkey. The Cartosat-1 datasets of Bavaria were acquired in the years 2008 and 2009 respectively. Since they were acquired with a time difference of only one year, which is not enough for a considerable increase in tree height, we are mostly interested in deforestation in this area. The Cartosat-1 images
Summary and discussion
In this paper, a novel approach to change detection using Cartosat-1 stereo images is presented. The approach is aiming to improve the change detection result by fusing all information that can be extracted from stereo images. Firstly, DSMs with 5 m resolution from two dates are generated with the semi-global matching method. Secondly, a region-based change detection workflow is proposed. Each region is characterized by a vector, which contains the change features from both images and DSMs. The
Conclusion
The main contribution of this research is to explore the applicability of Cartosat-1 stereo data for change detection even for objects like industrial buildings and minor forest regions. To achieve that, we have jointly used information from panchromatic images and height information coming from the DSMs derived from stereo imagery, and proposed a workflow which relies on a high quality DSM generation method, an efficient region merge procedure and a robust change detection method.
Without
References (34)
- et al.
Optimizing land cover classification accuracy for change detection: a combined pixel-based and object-based approach in a mountainous area in Mexico
Applied Geography
(2012) - et al.
Assessing different remote sensing techniques to detect land use/cover changes in the eastern Mediterranean
International Journal of Applied Earth Observation and Geoinformation
(2009) Object based image analysis for remote sensing
ISPRS Journal of Photogrammetry and Remote Sensing
(2010)- et al.
Forest change detection by statistical object-based method
Remote Sensing of Environment
(2006) - et al.
Deforestation in Central Africa: estimates at regional, national and landscape levels by advanced processing of systematically-distributed Landsat extracts
Remote Sensing of Environment
(2008) Object-based classification of remote sensing data for change detection
ISPRS Journal of Photogrammetry and Remote Sensing
(2004)A multilevel parcel-based approach to change detection in very high resolution multitemporal images
IEEE Geoscience and Remote Sensing Letters
(2009)- et al.
An adaptive parcel-based technique for unsupervised change detection
International Journal of Remote Sensing
(2000) Unsupervised change detection in satellite images using principal component analysis and k-means clustering
IEEE Geoscience and Remote Sensing Letters
(2009)- et al.
Object-based change detection
International Journal of Remote Sensing
(2012)
Mean-shift: a robust approach toward feature space analysis
IEEE Transactions on Pattern Analysis and Machine Intelligence
Digital change detection in forest ecosystems with remote sensing imagery
Remote Sensing Reviews
Towards automated DEM generation from high resolution stereo satellite images
International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences (Part B4)
Filling SRTM voids: the delta surface fill method
Photogrammetric Engineering and Remote Sensing
Cited by (85)
Extracting unstructured roads for smart Open-Pit mines based on computer vision: Implications for intelligent mining
2024, Expert Systems with ApplicationsSeasonal effects on aboveground biomass estimation in mountainous deciduous forests using ZY-3 stereoscopic imagery
2023, Remote Sensing of EnvironmentChange detection of urban objects using 3D point clouds: A review
2023, ISPRS Journal of Photogrammetry and Remote SensingA co-learning method to utilize optical images and photogrammetric point clouds for building extraction
2023, International Journal of Applied Earth Observation and GeoinformationSemantic feature-constrained multitask siamese network for building change detection in high-spatial-resolution remote sensing imagery
2022, ISPRS Journal of Photogrammetry and Remote SensingCitation Excerpt :Thus, multitask learning has become a focus for SCD (Liu et al., 2021; Papadomanolaki et al., 2021; Yang et al., 2022; Tsutsui et al., 2021). In the field of urban remote sensing, building change detection plays an important role in urban management, municipal planning, and disaster assessment (Qin et al., 2016; Bouziani et al., 2010; Sublime and Kalinicheva, 2019; Zheng et al., 2021; Xiao et al., 2016; Tian et al., 2013). For example, illegal construction can be partially found by building change detection; this technique extracts and locates newly emerged buildings by comparing historical and current remote sensing images to find potential illegal constructions.
A novel framework for seamless mosaic of Cartosat-1 DEM scenes
2021, Computers and Geosciences