object-oriented algorithms

Water erosion is one of the most important causes of soil destruction, and it is considered a serious environmental hazard all over the world. Recently, remote sensing is customarily used in conservation and erosion projects that most of them use air photography which, despite the many benefits, bot have limitations. The present study was aimed to detect and zoning soil erosion levels using high resolution of Sentinel-2 satellite image, its integration with aerial photographs, base maps, and implementation of various classification methods, including pixel-based and object-oriented techniques. After the staff operations, atmospheric and geometric corrections, pre-processing, and processing done on images of Sentinel-2 for detecting the area of erosion in the Lighvan watershed. In order to evaluate the correctness and accuracy of each method in this study, the criterions of user and producer accuracy, accuracy and kappa coefficient were compared. Based on the results, the supervised tuning method with user accuracy equal 77.78 and 33.33, has the highest and lowest accuracy for classification using spectral angle map algorithm and Mahalanubis distance, respectively. The maximum of overall accuracy and kappa coefficients, 72 and 62 percent, respectively, indicate the medium accuracy of the produced maps with pixel based algorithms. The results of object-oriented processing show that based on user and producer accuracy, object-oriented methods have increased accuracy (12 %) compared to pixel-based methods. Classification results with object-oriented algorithms and based on overall accuracy, 88 and 84 percent, respectively, for the brightness and the combination of brightness and slope, and the kappa coefficient for these two algorithms was 0.86 and 0.79, respectively. This result represents an acceptable increase in classification accuracy in the use of object-oriented algorithms compared to pixel base algorithms.

Due to the location of Iran in a dry and semi-arid region, it is always affected by sloping instability and erosion as the most severe type of erosion, gully erosion. This erosion pattern occurred in different parts of Iran and continuously over many years, and during erosion and the transfer of high sediment volume, it destroyed roads, facilities, pastures, slopes, etc. This will require the identification of high risk areas and the development of sensitivity maps. In recent years, the processing of satellite imagery as an advanced method has been widely used by researchers to increase the accuracy and save time and money. The object-oriented analysis of images is one of the most important methods for extracting information from satellite imagery, which is based on spectral, form and spatial characteristics and using expert knowledge to identify complications.

In this research, the Lighwan watershed basin was studied as one of the most important sub-basins of Aji Chay in East Azarbaijan Province and the satellite images of Sentinel-2 (2016) with spatial resolution of 10, 20 and 60 meters for the processing and identification of gully Was used. The images were processed using the eCognition software and applied with different types of algorithms to design a semi-automatic model based on object-oriented analysis. Finally, in order to evaluate the accuracy of the model, the identified Gully were mapped out and calculated using ArcGIS software to match the ground reality map and the formation of the error matrix, manufacturer accuracy, user accuracy and kappa coefficient for each of the algorithms.

The results showed that the Density and Compactness algorithms had the highest and lowest accuracy of the manufacturer (manufacturer accuracy was 88 and 78 respectively). While based on Kappa coefficient, the asymmetry algorithm has the highest accuracy compared to other methods (kappa = 0.91). Then, the shape index and density algorithms with kappa coefficient equal to 0.89 and 0.85 provided acceptable accuracy for the classification and identification of the gully.

The use of object-oriented methods due to the increased accuracy of classifying and identifying surface effects and phenomena can be used as a suitable solution for future soil studies and natural phenomena. In the present study, semi-automatic semi-automatic model for ditch identification was presented using spectral and geometric properties of Sentinel-2 satellite images and object-oriented processing in eCognition software environment. Key words: Object Oriented Algorithms, Object-Oriented Processing, Segmentation, Classification