Saeid Mahmoodizadeh; Ali Esmaeily
Abstract
Extended Abstract Introduction Information obtained from change detection processes in urban regions has a remarkable effect on urban planning and management. Due to the variety of land coversin urban regions, they are considered as a complex region extracting information from which is quite challengeable. ...
Read More
Extended Abstract Introduction Information obtained from change detection processes in urban regions has a remarkable effect on urban planning and management. Due to the variety of land coversin urban regions, they are considered as a complex region extracting information from which is quite challengeable. Hence, independent application ofoptical and radar data in changedetection may result in improper recognition of some altered regions and falsification ofobtained results. These two sensors record different kinds of information from different phenomenonat the earth’s surface, and thus can be considered as complementing each other. So, the fusion of these two data sources (radar and optical) can improve the detection of altered area. Radar data do not depend on the sun and atmospheric conditions and has thus gained much attention. In fact, radar data provide information on the spatial and geometrical characteristics of the geographical features, while optical sensors are sensitive to the reflectance of different surfaces at visible and infrared wavelengths.Therefore, the surface reaction is different in optical and radar data. Application of radar data in urban regions is limited merely due to the dependence of the intensity data (i) on the incidence angle and the speckle noise.On the other hand, independent application of optical data cannot produce accurate results in urban regions due to the spectral similarity of materials. And since the nature of these two types of images is different, it seems that their fusion improves and increases the accuracy of the information collectedfrom urban areas. Materials and Methodology Considering thebenefits of optical and radar data integrationas well as the application of unsupervised techniques in change detection studies, the present research has developed an unsupervised method for the integration of optical and radar data in order to detect changes. The area under study is a region located in the northwestof Mashhad city in northeastern Iran which has experienced considerable changes in its land cover from 2016 to 2018. Optical and radar dataare used toevaluate the proposed method. Optical data consists of a pair of multispectral imagesacquired from Sentinel-2 in 9/2016 and 9/2018. Radar data consists of a pair of SAR imagesacquired from Sentinel-1 in 9/2016 and 9/2018. The proposed method was used to integrate radar and optical data with the aim of obtaining a single band image with a higher information content. This method is an effective solution used to integrate data and reduce data dimensions from n to one dimension. In this method, necessary preprocessing was first performed on the radar and optical data, and then the characteristics extracted from optical and radar images were integratedpixel-to-pixel. technique was used to integrate these characteristics and detect changes. Generally in this method, input is divided into two categories of radar and optical data. The optical characteristics include spectral indices calculated from different bands at t1 and t2. These indices include NDVI, ARVI, SAVI, NDWI, NDBI, which are efficient for studying and identifying three types of land cover: vegetation, water and residential areas. In fact, to reduce the effects of topography and image brightness and to increase the possibility of detecting and segregating geographical features, the spectral indices were used as the input of optical part. Normalized ratio images obtained from the VV and VH polarizations of the radar images at t1 and t2 were considered as the input of radar data part. Then, a weight was estimated for each feature entering the segment using the PSO algorithm. Since the present study seeks to estimate the optimal weight of characteristics extracted from optical and radar images and ultimately to combine these features and obtain a single-band image, each particle in this algorithm contains the n weight of the extracted features from the images. OTSU thresholding techniquewhich is the relation used for inter-class variance maximization is also used as thecost function to assess the particles. In this function, the weight of each characteristic should be selected in a way that the inter-class (two classes of altered and unaltered regions)variancereaches its maximum value and the most optimal threshold limit can be estimated. The output of the proposed method will be a single-band image with higher information content. After applying the OTSU threshold limit, two classesof altered and unaltered regions are formed. The proposed method was also compared with other unsupervised change detection methods. Results Findings of the present study indicate high efficiency and accuracy of the method developed for changedetection. In this method, the ratio of pixels wronglydetected to the total number of evaluated pixels was 9.21% which is the lowest value. The overall accuracy and Kappa coefficients of the classification were respectively 90.79 and 0.819, which were the highest values compared to the other methods used in the present study. Conclusion Considering the benefits of optical and radar data integration, as well as unsupervised techniques application in change detection study, the present research has developed an unsupervised method for integration of optical and radar data andchangedetection. This unsupervised method for data integration is usedto achieve a single band image with higher information content. The technique makes it possible to integrate the optical and radar data and reduce data dimensions from n to one. For all input characteristics entering section, a weight was estimated using PSO algorithm. Since the proposed method is unsupervised, OTSU thresholding technique which is the relation used for inter-class variance maximization, is also used to assess the particles. The results have revealed high capability of the proposed method todetectchanges witha higher accuracy.
