Segmentation is crucial for the quality of the final classification results. Therefore it is very important to evaluate the segmentation results using quantitative methods and to know what kind of input data is needed to obtain the best possible segmentation results. The impact of the segmentation algorithm, the parameter settings, as well as the spatial and spectral resolution of the data is investigated in the paper. Paper has been published in the publication with impact factor 1.183.
COBISS.SI-ID: 36528685
This paper investigated the performance of two proposed classifiers and analysed the complexity of pixel distribution within the segment which was addressed using multiple random sampling of segment pixels and multiple calculations of similarity measures – statistics of non-parametric Kolmogorov-Smirnov test and parametric Student’s t-test. The performance of both classifiers was assessed on a WorldView-2 image. Both proposed classifiers showed an improvement in the overall classification accuracies and produced more accurate classification maps. Paper has been published in the publication with impact factor 3.180 (first quartile).
COBISS.SI-ID: 37833517
The paper presents a completely automatic processing chain for orthorectification of optical pushbroom sensors. The procedure is robust and works without manual intervention from raw satellite image to orthoimage. It is modularly divided in four main steps: metadata extraction, automatic ground control point (GCP) extraction, geometric modeling, and orthorectification. The GCP extraction step uses georeferenced vector roads as a reference and produces a file with a list of points and their accuracy estimation. The physical geometric model is based on collinearity equations and works with sensor-corrected (level 1) optical satellite images. It models the sensor position and attitude with second-order piecewise polynomials depending on the acquisition time. The exterior orientation parameters are estimated in a least squares adjustment, employing random sample consensus and robust estimation algorithms for the removal of erroneous points and fine-tuning of the results. The images are finally orthorectified using a digital elevation model and positioned in a national coordinate system. The usability of the method is presented with three RapidEye images of regions with different terrain configurations. Several tests were carried out to verify the efficiency of the procedure and to make it more robust. Using the geometric model, subpixel accuracy on independent check points was achieved, and positional accuracy of orthoimages was around one pixel. The proposed procedure is general and can be easily adapted to various sensors.
COBISS.SI-ID: 38502189
In this paper, we give an overview of optical remote sensing imagery segmentation algorithms and the possibility of their application. Besides, an overview of object-based classification software with a focus on segmentation process is given. The intent of this paper is not to give a receipt for the best algorithm and software selection, but to support users with the necessary knowledge to be able to obtain the best possible segmentation results of the analysis in given circumstances.
COBISS.SI-ID: 39605549
In this paper, we evaluated the correlation between segmentation accuracy and subsequent image classification accuracy. We conducted an assessment of segmentation and classification by analyzing 100 different segmentation parameter combinations, 3 classifiers, 5 land cover classes, 20 segmentation evaluation metrics, and 7 classification accuracy measures. The results showed that all unsupervised metrics that are not based on a number of segments have a very strong correlation with all classification measures and are therefore reliable as indicators of land cover classification accuracy. On the other hand, correlation at supervised metrics is dependent on so many factors that it cannot be trusted as a reliable classification quality indicator.
COBISS.SI-ID: 40688685