Object-based Image Analysis Techniques Research Articles

Mapping vegetation in a late Quaternary landform of the Amazonian wetlands using object-based image analysis and decision tree classification

Fan-shaped morphologies related to late Quaternary residual megafan depositional systems are common features over wide areas in northern Amazonia. These features were formed by ancient distributary drainage systems that are in great contrast to tributary drainage networks that typify the modern Amazon basin. The surfaces of the Amazonian megafans constitute vegetacional mosaic wetlands with different campinarana types. A fine-scale-resolution investigation is required to provide detailed classification maps for the various campinarana and surrounding forest types associated with the Amazonian megafans. This approach remains to be presented, despite its relevance for analysing the relationship between stages of plant succession and sedimentary dynamics associated with the evolution of megafans. In this work, we develop a methodology for classifying vegetation over a fan-shaped megafan palaeoform from a northern Amazonian wetland. The approach included object-based image analysis (OBIA) and data-mining (DM) techniques combining Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) images, land-cover fractions derived by the linear spectral mixing model, synthetic aperture radar (SAR) images, and the digital elevation model (DEM) acquired during the Shuttle Radar Topography Mission (SRTM). The DEM, vegetation fraction, and ASTER band 3 were the most useful parameters for defining the forest classes. The normalized difference vegetation index (NDVI), ASTER band 1, vegetation fraction, and the Advanced Land Observing Satellite (ALOS)/Phased Array type L-band Synthetic Aperture Radar (PALSAR) transmitting and receiving horizontal polarization (HH) and transmitting horizontal and receiving vertical polarization (HV) were all effective in distinguishing the wetland classes campinarana and Mauritia. Tests of statistical significance indicated the overall accuracies and kappa coefficients (κ) of 88% and 0.86 for the final map, respectively. The allocation disagreement coefficient of 5% and a quantity disagreement value of 7% further attested the statistical significance of the classification results. Hence, in addition to water, exposed soil, and deforestation areas, OBIA and DM were successful for differentiating a large number of open (forest, wood, shrub, and grass campinaranas), forest (terra firme, várzea, igapó, and alluvial), as well as Mauritia wetland classes in the inner and outer areas of the studied megafan.

High-Throughput 3-D Monitoring of Agricultural-Tree Plantations with Unmanned Aerial Vehicle (UAV) Technology.

The geometric features of agricultural trees such as canopy area, tree height and crown volume provide useful information about plantation status and crop production. However, these variables are mostly estimated after a time-consuming and hard field work and applying equations that treat the trees as geometric solids, which produce inconsistent results. As an alternative, this work presents an innovative procedure for computing the 3-dimensional geometric features of individual trees and tree-rows by applying two consecutive phases: 1) generation of Digital Surface Models with Unmanned Aerial Vehicle (UAV) technology and 2) use of object-based image analysis techniques. Our UAV-based procedure produced successful results both in single-tree and in tree-row plantations, reporting up to 97% accuracy on area quantification and minimal deviations compared to in-field estimations of tree heights and crown volumes. The maps generated could be used to understand the linkages between tree grown and field-related factors or to optimize crop management operations in the context of precision agriculture with relevant agro-environmental implications.

Gully Erosion Mapping Using Object-Based and Pixel-Based Image Classification Methods

Gully erosion mapping is a crucial step to monitor the erosion process and to study its current and future local impacts. Gully erosion mapping through field-work is difficult, time-consuming, and costly. This article compares various pixel-based image classification (PBC) algorithms, such as ISODATA, Maximum Likelihood Classification, and Support Vector Machine, with the object-based image analysis (OBIA) technique for gully erosion mapping on IRS-P6 images. Six models defined by classification types, classifiers, and feature spaces were built for comparison. The results show that OBIA classification performed better than PBC in terms of accuracy. We also found that the improvement of OBIA was primarily due to employing textural and shape features and optimized feature space, while the use of standard feature space did not improve OBIA. In addition, OBIA significantly reduced the salt-and-pepper effect that obscures the features on the output maps compared to the PBC maps (which had more salt-and-pepper effects). It seems that object-based techniques have yielded better results because of their focus on the shape of gully networks rather than on their spectral heterogeneity. In order to improve the accuracy, a priority may be gained by fully exploring the use of membership function and hierarchical approach with multi-scale segmentation for gully mapping. In future studies we propose to determine how these factors can affect the performance of OBIA in terms of gully mapping. This study provides information on the location of gullies, gully dynamics over a period of time, and the degree of land degradation (gully density) for developing and implementing soil conservation measures.

Evaluation of state and community/private forests in Punjab, Pakistan using geospatial data and related techniques

Abstract Background Forests are fundamental in maintaining water supplies, providing economic goods, mitigating climate change, and maintaining biodiversity, thus providing many of the world’s poorest with income, food and medicine. Too often, forested lands are treated as “wastelands” or “free” and are easily cleared for agricultural and infrastructure expansion. Methods In this paper, the sustainability of two forest ecosystems (state and community/private owned) was evaluated using SPOT-5 satellite images of 2005 and 2011. This study was conducted in a sub-watershed area covering 468 km2, of which 201 km2 is managed by the state and 267 km2 by community/private ownership in the Murree Galliat region of Punjab Province of Pakistan. A participatory approach was adopted for the delineation and demarcation of forest boundaries. The Geographic Object-Based Image Analysis (GEOBIA) technique was used for identification and mapping of ten Land Cover (LC) features. Results The results show that between the years 2005 to 2011, a total of 55 km2 (24 km2 in state-owned forest and 31 km2 in community/private forest) was converted from forest to non-forest. The conclusion is that conservation is more effective in state-owned forests than in the community/private forests. Conclusions These findings may help to mobilize community awareness and identify effective initiatives for improved management of community/private forest land for other regions of Pakistan.

Pixel-based and object-based classifications using high- and medium-spatial-resolution imageries in the urban and suburban landscapes

With the increasing availability of high-spatial-resolution remote sensing imageries and with the observed limitations of pixel-based techniques, the development and testing of geographic object-based image analysis (GEOBIA) techniques for image classification have become one of the main research areas in geospatial science. This paper examines and compares the classification performance of a pixel-based method and an object-based method as applied to high- (QuickBird satellite image) and medium- (Landsat TM image) spatial-resolution imageries in the context of urban and suburban landscapes. For the pixel-based classification, the maximum-likelihood supervised classification approach was employed. And for the object-based classification, the pixel-based classified maps were integrated with a set of image segments produced using various calibrations. The results show evidence that the object-based method can produce classifications that are more accurate for both high- and medium-spatial- resolution imageries in the context of urban and suburban landscapes.

Extraction of complex plantations from VHR imagery using OBIA techniques

The study aimed to extract complex plantation using Object Based Image Analysis (OBIA) techniques. GeoEye-1 image covering Pa Khlok sub-district, Phuket Thailand was used, and thirteen vegetation indices calculated and analyzed with the aim of exploring plantations coverage in the area. Five plantation classes were identified including young coconut, mature coconut, young rubber, mature rubber and oil palm, with another five non-plantation classes assigned to water, built-up land, bare ground, mangrove forest and all other, using rule based techniques. Results support also the idea of mixed plantations in heterogeneous patterns with mixed and missing classes, as experienced in traditional pixel based classification. OBIA techniques can be used successfully to classify complex plantation structures in the study area, with values of 88% and 79% for overall accuracy and kappa coefficients of 0.85 and 0.75 in empirical (development) rules set images and validation images, respectively.

Monitoring land use changes associated with urbanization: An object based image analysis approach

Land use/land cover (LULC) change occurs due to natural and anthropogenic causes. In developing countries, rapid industrialization and urbanization imposes a major threat to natural environment. Present study was carried out to monitor the LULC changes due to urbanization in a rapidly changing river basin, India. The purpose of choosing the river basin was to analyze past changes and predict possible consequences within a defined natural boundary. Multi-temporal images acquired from Landsat and Indian Remote Sensing (IRS) satellites between 1992–2009 as well as a digital elevation model was used to generate historical and current LULC pattern in the basin. An object based image analysis technique was employed for precise classification of multi-temporal images followed by GIS-based change detection studies. The study reveals that the built up area has increased significantly and added 288 km2 between 1992–2009. Increase in built up area is attributed to decrease in wastelands and agricultural land. The expansion of built up area along major transportation networks, specifically after the year 2000 shows the rapid rate of urbanization in the basin.

Utilizing National Agriculture Imagery Program Data to Estimate Tree Cover and Biomass of Piñon and Juniper Woodlands

With the encroachment of piñon (Pinus ssp.) and juniper (Juniperus ssp.) woodlands onto sagebrush steppe rangelands, there is an increasing interest in rapid, accurate, and inexpensive quantification methods to estimate tree canopy cover and aboveground biomass. The objectives of this study were 1) to evaluate the relationship and agreement of piñon and juniper (P-J) canopy cover estimates, using object-based image analysis (OBIA) techniques and National Agriculture Imagery Program (NAIP, 1-m pixel resolution) imagery with ground measurements, and 2) to investigate the relationship between remotely-sensed P-J canopy cover and ground-measured aboveground biomass. For the OBIA, we used eCognition® Developer 8.8 software to extract tree canopy cover from NAIP imagery across 12 P-J woodlands within the Sagebrush Steppe Treatment Evaluation Project (SageSTEP) network. The P-J woodlands were categorized based on the dominant tree species found at the individual sites for the analysis (western juniper, Utah juniper, and mixed P-J community). Following tree canopy cover extractions, relationships were assessed between remotely-sensed canopy cover and ground-measured aboveground biomass. Our OBIA estimates for P-J canopy cover were highly correlated with ground-measured tree canopy cover (averaged across all regions r = 0.92). However, differences between methods occurred for western and Utah juniper sites (P < 0.05), and were more prominent where tree canopy cover was > 40%. There were high degrees of correlation between predicted aboveground biomass estimates with the use of remotely-sensed tree canopy cover and ground-measured aboveground biomass (averaged across all regions r = 0.89). Our results suggest that OBIA methods combined with NAIP imagery can provide land managers with quantitative data that can be used to evaluate P-J woodland cover and aboveground biomass rapidly, on broad scales. Although some accuracy and precision may be lost when utilizing aerial imagery to identify P-J canopy cover and aboveground biomass, it is a reasonable alternative to ground monitoring and inventory practices.

Determination of the age of oil palm from crown projection area detected from WorldView-2 multispectral remote sensing data: The case of Ejisu-Juaben district, Ghana

Information about age of oil palm is important in sustainability assessments, carbon mapping, yield projections and precision agriculture. The aim of this study was to develop and test an approach to determine the age of oil palm plantations (years after planting) by combining high resolution multispectral remote sensing data and regression techniques using a case study of Ejisu-Juaben district of Ghana. Firstly, we determined the relationship between age and crown projection area of oil palms from sample fields. Secondly, we did hierarchical classification using object based image analysis techniques on WorldView-2 multispectral data to determine the crown projection areas of oil palms from remote sensing data. Finally, the crown projection areas obtained from the hierarchical classification were combined with the field-developed regression model to determine the age of oil palms at field level for a wider area. Field collected data showed a strong linear relationship between age and crown area of oil palm up to 13years beyond which no relationship was observed. A user’s accuracy of 80.6% and a producer’s accuracy of 68.4% were obtained for the delineation of oil palm crowns while for delineation of non-crown objects a user’s accuracy of 65.6% and a producer’s accuracy of 78.6% were obtained, with an overall accuracy of 72.8% for the OBIA delineation. Automatic crown projection area delineation from remote sensing data produced crown projection areas which closely matched the field measured crown areas except for older oil palms (13+ years) where the error was greatest. Combining the remote sensing detected crown projection area and the regression model accurately estimated oil palm ages for 27.9% of the fields and had an estimation error of 1year or less for 74.6% of the fields and an error of a maximum 2years for 92.4% of the fields. The results showed that 6 and 11year old oil palm stands were dominating age categories in the study area. Although the method could be reliably applied for estimating oil palm age at field level, more attention is required in improving crown area delineation to improve the accuracy of the approach.

Geographic Object-Based Image Analysis Using Optical Satellite Imagery and GIS Data for the Detection of Mining Sites in the Democratic Republic of the Congo

Earth observation is an important source of information in areas that are too remote, too insecure or even both for traditional field surveys. A multi-scale analysis approach is developed to monitor the Kivu provinces in the Democratic Republic of the Congo (DRC) to identify hot spots of mining activities and provide reliable information about the situation in and around two selected mining sites, Mumba-Bibatama and Bisie. The first is the test case for the approach and the detection of unknown mining sites, whereas the second acts as reference case since it is the largest and most well-known location for cassiterite extraction in eastern Congo. Thus it plays a key-role within the context of the conflicts in this region. Detailed multi-temporal analyses of very high-resolution (VHR) satellite data demonstrates the capabilities of Geographic Object-Based Image Analysis (GEOBIA) techniques for providing information about the situation during a mining ban announced by the Congolese President between September 2010 and March 2011. Although the opening of new surface patches can serve as an indication for activities in the area, the pure change between the two satellite images does not in itself produce confirming evidence. However, in combination with observations on the ground, it becomes evident that mining activities continued in Bisie during the ban, even though the production volume went down considerably.

Mapping urban land cover types using object-based multiple endmember spectral mixture analysis

Spectral mixture analysis has been frequently applied in various fields to solve the mixed pixel problem in remote sensing. So far, all the research in mixture analysis has focused on the sub-pixel analysis, i.e., selecting endmembers and conducting mixture analysis at the pixel level. Research efforts in mixture analysis at the object level are very scarce, even though the object-based image analysis (OBIA) techniques have been well developed. In this study, we examined the applicability of object-based mixture analysis in an urban environment using a Landsat Thematic Mapper image. Informative and accurate object-based fraction maps (vegetation, impervious surface, and water) were produced by combining the OBIA and multiple endmember spectral mixture analysis (MESMA) techniques. A new approach to identifying the spectral representatives of a specific class for MESMA was developed. The accuracy of the object-based fraction maps were assessed using manual interpretation results of a 1-m digital aerial photograph. Object-based mixture analysis produced a higher accuracy than traditional pixel-based mixture analysis. This work illustrates the potential of object-based mixture analysis of moderate spatial resolution imagery in mapping heterogeneous urban environments.

Comparison of remotely-sensed surveys vs. in situ plot-based assessments of sea grass condition in Barnegat Bay-Little Egg Harbor, New Jersey USA

With the increasing appreciation that sea grass habitats are in global decline, there is a great need to be able to efficiently and effectively assess and characterize the status and trends of sea grass in our coastal ecosystems. This paper examines the utility of remotely sensed vs. in situ plot-based monitoring using the Barnegat Bay-Little Egg Harbor (BB-LEH), New Jersey, USA estuarine system as a case study. Eelgrass (Zostera marina) is the dominant species, while widgeon grass (Ruppia maritima) is also common in lower salinity regions of the BB-LEH. Aerial imagery collected during the months of July and August 2009 was interpreted and mapped using object based image analysis techniques, similar to techniques used in the 2003 mapping survey of this system. Boat-based in situ monitoring data were collected concurrently with the aerial photography to assist the image interpretation and for an independent accuracy assessment. We compared the remotely-sensed mapping of sea grass cover change (in 2003 vs. 2009) vs. in situ plot-based monitoring conducted from 2004 through 2009. Comparison of the remotely-sensed vs. the in situ plot-change analysis suggests that the two methodologies had broadly similarly results, with the percent area showing declines in sea grass cover greater than those that exhibited increases. In conclusion, the two studies provide corroborating evidence that sea grass has declined in percent cover in the BB-LEH system during the decade of the 2000’s. While remotely-sensed surveys provide synoptic information for a “big picture” view on sea grass distribution, site specific in situ sampling is required to determine other aspects of sea grass status, e.g. above vs. below-ground biomass, blade length, shoot density, epiphytic loading, etc. Either method alone gives an incomplete picture. As demonstrated in this study, to fully characterize the spatial extent, health, and density of sea grass meadows across the entire estuary, combining remote sensing surveys concomitantly with comprehensive in situ assessment provides the most robust approach.

Evaluation of object-based image analysis techniques on very high-resolution satellite image for biomass estimation in a watershed of hilly forest of Nepal

Methods of forest carbon estimation using remote-sensing data and techniques are evolving within a short timeframe as compared to traditional forest inventory methods. Object-based image analysis (OBIA) provided new opportunities to improve biomass and carbon stock estimation and mapping by delineating and classifying crown projection area (CPA) of individual trees. In this paper, image segmentation techniques of OBIA (region growing and valley following) are being applied on GeoEye-1 satellite data and compared in terms of accuracy in Ludhikhola watershed in the Gorkha District of Nepal. Accuracy assessment of tree crown delineation of both segmentation approaches were analyzed using accuracy measures of D and one-to-one (1:1) correspondence. The combination of over-segmentation and under-segmentation, D is interpreted as the “closeness” measure to an ideal segmentation result in relation to a predefined reference set. Region growing and valley following segmentation 68 and 58 % accuracy, respectively, was achieved and linear regression model was developed for carbon stock for Shorea robusta which resulted into a coefficient of determination value of 0.67 at 95 % confidence level and the coefficient of determination resulted into a value of 0.70 for other species. The research concluded that region growing approach showed better delineation as compared to valley follow approach, since it used both features of local maxima and local minima.

An Object-Based Image Analysis of Pinyon and Juniper Woodlands Treated to Reduce Fuels

Mechanical and prescribed fire treatments are commonly used to reduce fuel loads and maintain or restore sagebrush steppe rangelands across the Great Basin where pinyon (Pinus) and juniper (Juniperus) trees are encroaching and infilling. Geospatial technologies, particularly remote sensing, could potentially be used in these ecosystems to (1) evaluate the longevity of fuel reduction treatments, (2) provide data for planning and designing future fuel-reduction treatments, and (3) assess the spatial distribution of horizontal fuel structure following fuel-reduction treatments. High-spatial resolution color-infrared imagery (0.06-mpixels) was acquired for pinyon and juniper woodland plots where fuels were reduced by either prescribed fire, tree cutting, or mastication at five sites in Oregon, California, Nevada, and Utah. Imagery was taken with a Vexcel UltraCam X digital camera in June 2009. Within each treatment plot, ground cover was measured as part of the Sagebrush Steppe Treatment Evaluation Project. Trimble eCognition Developer was used to classify land cover classes using object-based image analysis (OBIA) techniques. Differences between cover estimates using OBIA and ground-measurements were not consistently higher or lower for any land cover class and when evaluated for individual sites, were within ±5% of each other. The overall accuracy and the K hat statistic for classified thematic maps for each treatment were: prescribed burn 85% and 0.81; cut and fell 82% and 0.77, and mastication 84% and 0.80. Although cover assessments from OBIA differed somewhat from ground measurements, they are sufficiently accurate to evaluate treatment success and for supporting a broad range of management concerns.

An Optical High and Medium Spatial Resolution Approach for Erosion-Prone Areas Assessment in Mustang, Nepal

Every year during the rainy season, water-induced soil erosion poses serious spatial-environmental problems, causing heavy damage to agricultural lands, sedimentation in reservoirs, and water quality problems in nearby surface water bodies, from the plains to the mountain areas in Nepal. The goal of this study is to identify potential areas for soil erosion in sub and macro watershed in Mustang, Nepal using remote sensing (RS) and geographic information systems (GIS) techniques. The study examines the possibility of advanced mapping of soil erosion-prone areas using a high spatial resolution image of QuickBird satellite and medium spatial resolution of Landsat satellite. The satellite image was classified using object-based image analysis (OBIA) techniques, taking into account spectral, spatial, and context information as well as hierarchical properties. The resulting land cover classification was thereafter combined with additional data in ArcGIS, where the input layers were reclassified and all classes of the input layers were ranked according to their proneness to soil erosion. Soil erosion-prone areas were delineated in five classes ranging from “very high” to “very low”. Using high spatial resolution image the study revealed that 22% area categorized as “high erosion-prone” areas and 5% as “very high” or “extremely erosion-prone”. Using medium resolution image the study exposed that 27% area categorized as “high erosion-prone” areas and 6% as “very high” or “extremely erosion-prone”. Comparison between two analysed erosion results almost all the erosion zone area was very close excluding medium erosion-prone category. The study proved GIS modeling techniques can successfully identify soil erosion-prone areas. The soil erosion-prone map produced out of the exercise can be used in decision making, particularly in selecting conservation measures to reduce soil loss.

Object-based Vegetation Mapping in the Kissimmee River Watershed Using HyMap Data and Machine Learning Techniques

Accurate and informative vegetation maps are in urgent demand to support the Kissimmee-Okeechobee-Everglades ecosystem restoration project in South Florida. In this study, we evaluated the applicability of fine spatial resolution hyperspectral data collected from the HyMap sensor for both community- and species-level vegetation mapping. Informative and accurate vegetation maps were produced by combining machine learning methods (Support Vector Machines (SVM) and Random Forest (RF)), object-based image analysis techniques, and Minimum Noise Fraction (MNF) data transformation. An overall accuracy of 90% was obtained in discriminating 14 vegetation communities. Classification of a large number of species is also promising. An overall accuracy of 85% was achieved in identifying 55 species using a SVM model. The results indicate that fine spatial resolution hyperspectral data classification using such automated procedure has great potential to replace the manual interpretation of aerial photos for vegetation mapping in heterogeneous wetland ecosystems.

Requirements for labelling forest polygons in an object-based image analysis classification

The ability to spatially quantify changes in the landscape and create land-cover maps is one of the most powerful uses of remote sensing. Recent advances in object-based image analysis (OBIA) have also improved classification techniques for developing land-cover maps. However, when using an OBIA technique, collecting ground data to label reference units may not be straightforward, since these segments generally contain a variable number of pixels as well as a variety of pixel values, which may reflect variation in land-cover composition. Accurate classification of reference units can be particularly difficult in forested land-cover types, since these classes can be quite variable on the ground. This study evaluates how many prism sample locations are needed to attain an acceptable level of accuracy within forested reference units in southeastern New Hampshire (NH). Typical forest inventory guidelines suggest at least 10 prism samples per stand, depending on the stand area and stand type. However, because OBIA segments group pixels based on the variance of the pixels, fewer prism samples may be necessary in a segment to properly estimate the stand composition. A bootstrapping statistical technique was used to find the necessary number of prism samples to limit the variance associated with estimating the species composition of a segment. Allowing for the lowest acceptable variance, a maximum of only six prism samples was necessary to label forested reference units. All polygons needed at least two prism samples for classification. †Present address: School of Forestry and Environmental Studies, Yale University, New Haven, CT, USA Present address: Math and Science Division, Babson College, Babson Park, MA, USA.

LANDSLIDE MAPPING USING IMAGERY ACQUIRED BY A FIXED-WING UAV

Abstract. In Taiwan, the average annual rainfall is about 2,500 mm, about three times the world average. Hill slopes where are mostly under meta-stable conditions due to fragmented surface materials can easily be disturbed by heavy typhoon rainfall and/or earthquakes, resulting in landslides and debris flows. Thus, an efficient data acquisition and disaster surveying method is critical for decision making. Comparing with satellite and airplane, the unmanned aerial vehicle (UAV) is a portable and dynamic platform for data acquisition. In particularly when a small target area is required. In this study, a fixed-wing UAV that equipped with a consumer grade digital camera, i.e. Canon EOS 450D, a flight control computer, a Garmin GPS receiver and an attitude heading reference system (AHRS) are proposed. The adopted UAV has about two hours flight duration time with a flight control range of 20 km and has a payload of 3 kg, which is suitable for a medium scale mapping and surveying mission. In the paper, a test area with 21.3 km2 in size containing hundreds of landslides induced by Typhoon Morakot is used for landslides mapping. The flight height is around 1,400 meters and the ground sampling distance of the acquired imagery is about 17 cm. The aerial triangulation, ortho-image generation and mosaicking are applied to the acquired images in advance. An automatic landslides detection algorithm is proposed based on the object-based image analysis (OBIA) technique. The color ortho-image and a digital elevation model (DEM) are used. The ortho-images before and after typhoon are utilized to estimate new landslide regions. Experimental results show that the developed algorithm can achieve a producer's accuracy up to 91%, user's accuracy 84%, and a Kappa index of 0.87. It demonstrates the feasibility of the landslide detection algorithm and the applicability of a fixed-wing UAV for landslide mapping.

AUTOMATING THE PHOTOGRAMMETRIC WORKFLOW IN A NATIONAL MAPPING AGENCY

Abstract. The goal of automating the process of identifying changes to topographic features in aerial photography, extracting the geometry of these features and recording the changes in a database, is yet to be fully realised. At Ordnance Survey, Britain's national mapping agency, research into the automation of these processes has been underway for several years, and is now beginning to be implemented in production systems. At the start of the processing chain is the identification of change – new buildings and roads being constructed, old structures demolished, alterations to field and vegetation boundaries and changes to inland water features. Using eCognition object-based image analysis techniques, a system has been developed to detect the changes to features. This uses four-band digital imagery (red, green, blue and near infra-red), together with a digital surface model derived by image matching, to identify all the topographic features of interest to a mapping agency. Once identified, these features are compared with those in the National Geographic Database and any potential changes are highlighted. These changes will be presented to photogrammetrists in the production area, who will rapidly assess whether or not the changes are real. If the change is accepted, they will manually capture the geometry and attributes of the feature concerned. The change detection process, although not fully automatic, cuts down the amount of time required to update the database, enabling a more efficient data collection workflow. Initial results, on the detection of changes to buildings only, showed a completeness value (proportion of the real changes that were found) of 92% and a correctness value (proportion of the changes found that were real changes) of 22%, with a time saving of around 50% when compared with the equivalent manual process. The completeness value is similar to those obtained by the manual process. Further work on the process has added vegetation, water and many other rural features to the list of features that can be detected, and the system is currently being evaluated in a production environment. In addition to this work, the research team at Ordnance Survey are working with the remote sensing (data collection) department to develop more efficient methods of DSM and DTM creation; more automated seamline-generation for the creation of orthoimage mosaics; and methods to automatically record simple building heights on buildings in the database. These are all methods that have been proven in a research environment – the challenge is to implement them within the working environment of the existing data collection process.

Object Based Image Analysis and Data Mining applied to a remotely sensed Landsat time-series to map sugarcane over large areas

The aim of this research was to develop a methodology for contributing in the automation of sugarcane mapping over large areas, with time-series of remotely sensed imagery. To this end, two major techniques were combined: Object Based Image Analysis (OBIA) and Data Mining (DM). OBIA was used to represent the knowledge needed to map sugarcane, whereas DM was applied to generate the knowledge model. To derive the image objects, the segmentation algorithm implemented in Definiens Developer® was used. The data mining algorithm used was J48, which generates decision trees (DT) from a previously prepared training set. The study area comprises three municipalities located in the northwest of São Paulo state, all of which are good representatives of the agricultural conditions in the Southern and Southeastern regions of Brazil. A time series of Landsat TM and ETM+ images was acquired in order to represent the wide range of pattern variation along the sugarcane crop cycle. After training, the DT was applied to the Landsat time series, thus generating the desired thematic map with sugarcane ready to harvest. Classification accuracy was calculated over a set of 500 points not previously used during the training stage. Using error matrix analysis and Kappa statistics, tests for statistical significance were derived. The statistics indicated that the classification achieved an overall accuracy of 94% and a Kappa coefficient of 0.87. Results show that the combination of OBIA and DM techniques is very efficient and promising for the sugarcane classification process.