Segment-Based Land Cover Mapping of a Suburban Area—Comparison of High-Resolution Remotely Sensed Datasets Using Classification Trees and Test Field Points

Matikainen, Leena; Karila, Kirsi

doi:10.3390/rs3081777

Cited by 79 publications

(45 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Table 3 shows a comparison with similar works that have done per-pixel classification estimation with similar choices of categories, but that cover other land areas. A direct comparison of different studies is not possible, because of different study areas, data acquisition methods and resolutions, reference datasets and class definitions [48]. However, it can be concluded that the use of CNNs for per-pixel classification is capable of achieving comparable results to the typically-regarded state-of-the-art object-based image analysis (OBIA) methods.…”

Section: Classification Resultsmentioning

confidence: 99%

“…Furthermore, a DSM is invariant to lighting and color variations and can give a better geometry estimation and background separation [48]. However, the measurements in a DSM need to be normalized because they are measured from a global reference point and not from the local surrounding ground level.…”

Section: Data and Pre-processingmentioning

confidence: 99%

“…Most datasets that use classification on a pixel-level only use a few hundred reference points [31,48,[50][51][52]. In order to validate the classifier using a much larger pool of validation points and for supervised learning and fine-tuning, each pixel in the full map is manually labeled.…”

Section: Manual Labelingmentioning

confidence: 99%

See 2 more Smart Citations

Classification and Segmentation of Satellite Orthoimagery Using Convolutional Neural Networks

et al. 2016

View full text Add to dashboard Cite

Abstract:The availability of high-resolution remote sensing (HRRS) data has opened up the possibility for new interesting applications, such as per-pixel classification of individual objects in greater detail. This paper shows how a convolutional neural network (CNN) can be applied to multispectral orthoimagery and a digital surface model (DSM) of a small city for a full, fast and accurate per-pixel classification. The predicted low-level pixel classes are then used to improve the high-level segmentation. Various design choices of the CNN architecture are evaluated and analyzed. The investigated land area is fully manually labeled into five categories (vegetation, ground, roads, buildings and water), and the classification accuracy is compared to other per-pixel classification works on other land areas that have a similar choice of categories. The results of the full classification and segmentation on selected segments of the map show that CNNs are a viable tool for solving both the segmentation and object recognition task for remote sensing data.

show abstract

Section: Classification Resultsmentioning

confidence: 99%

Section: Data and Pre-processingmentioning

confidence: 99%

See 1 more Smart Citation

Classification and Segmentation of Satellite Orthoimagery Using Convolutional Neural Networks

et al. 2016

View full text Add to dashboard Cite

show abstract

“…If there is a point inside a segment, the entire segment is assumed to belong to the class defined by the point. This approach is flexible and allows the same point set to be applied to different images and segmentations (Matikainen and Karila, 2011).…”

Section: Methods For Change Detectionmentioning

confidence: 99%

The Challenge of Automated Change Detection: Developing a Method for the Updating of Land Parcels

Matikainen

Karila

Litkey

et al. 2012

ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

ABSTRACT:Development of change detection methods that are functional and reliable enough for operational work is still a demanding task. This article discusses automated change detection from the viewpoint of one case study: the Finnish Land Parcel Identification System (FLPIS). The objective of the study is to develop a change detection method that could be used as an aid in the updating of the FLPIS. The method is based on object-based interpretation, and it uses existing parcel boundaries and new aerial ortho images as input data. Rules for classifying field and non-field objects are defined automatically by using the classification tree method and training data. Additional, manually created rules are used to improve the results. Classification tests carried out during the development work suggest that real changes can be detected relatively well. According to a recent visual evaluation, 96% of changes larger than 100 m 2 were detected, at least partly. The overall accuracy of the change detection results was 93% when compared with reference data pixel-by-pixel. On the other hand, there are also missing changes and numerous false alarms. The main challenges encountered in the method development include the wide diversity of agricultural fields and other land cover objects locally, across the country, and at different times of the spring and summer, variability in the digital numbers (DNs) of the aerial images, the different nature of visual and automatic interpretation, and the small percentage of the total field area that has really changed. These challenges and possible solutions are discussed in the article.

show abstract

“…Impervious land cover have similar spectral properties especially when only limited number of bands were used; therefore, inclusion of spatial and texture information into urban area classification is necessary (Salehi;2012) Different researchers have applied various approaches to improve urban land use/cover accuracy with the inclusion of geographic data, textural features, contextual information, expert systems and fuzzy classification (Thapa and Murayama, 2009;Deng et al, 2009;Weng, 2012;Mohapatra, and Wu, 2011;Lu, and Weng, 2006;City Development Information. , 2014;Gamba et al, 2005;Matikainen and Karila, 2011). The importance of different requirements for spatial, spectral and temporal resolution and geometric characteristics of urban features to map impervious surfaces in the urban areas using remotely sensed data was emphasized (Herold et al, 2003).…”

Section: Introductionmentioning

confidence: 99%

High resolution mapping of urban areas using SPOT-5 images and ancillary data

Sertel

Akay

2015

International Journal of Environment and Geoinformatics

View full text Add to dashboard Cite

This research aims to propose new rule sets to be used for object based classification of SPOT-5 images to accurately create detailed urban land cover/use maps. In addition to SPOT-5 satellite images, Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) maps, cadastral maps, Openstreet maps, road maps and Land Cover maps, were also integrated into classification to increase the accuracy of resulting maps. Gaziantep city, one of the highly populated cities of Turkey with different landscape patterns was selected as the study area. Different rule sets involving spectral, spatial and geometric characteristics were developed to be used for object based classification of 2.5 m resolution Spot-5 satellite images to automatically create urban map of the region. Twenty different land cover/use classes obtained from European Urban Atlas project were applied and an automatic classification approach was suggested for high resolution urban map creation and updating. Integration of different types of data into the classification decision tree increased the performance and accuracy of the suggested approach. The accuracy assessment results illustrated that with the usage of newly proposed rule set algorithms in object-based classification, urban areas represented with seventeen different sub-classes could be mapped with 94 % or higher overall accuracy.

show abstract

Segment-Based Land Cover Mapping of a Suburban Area—Comparison of High-Resolution Remotely Sensed Datasets Using Classification Trees and Test Field Points

Cited by 79 publications

References 41 publications

Classification and Segmentation of Satellite Orthoimagery Using Convolutional Neural Networks

Classification and Segmentation of Satellite Orthoimagery Using Convolutional Neural Networks

The Challenge of Automated Change Detection: Developing a Method for the Updating of Land Parcels

High resolution mapping of urban areas using SPOT-5 images and ancillary data

Contact Info

Product

Resources

About