Corona and Hexagon satellite photographs with a spatial resolution of 1.8 to 9 meters are a good source for monitoring and evaluating changes in surface phenomena. The purpose of this study is to monitor the changes in Zaribar Lake in the period 1969-2019 using the data of Corona, Hexagon and Aster satellites and their ability to monitor and extract the coastline, lake boundary and water surface. In this study, to geometric correction the corona and hexagon data from Google Earth images, linear extraction algorithms, binary mask and mean shift segmentation were used to extract the coastline and lake boundary, detect lake changes and extract and monitor the water area of Zaribar Lake. The results showed that in the first step: geometric correction of corona and hexagon images was obtained using Google Earth images with RMSE of 0.3 and 0.4 pixels. In the second stage, the linear extraction algorithm for extracting the lake boundary and coastline using corona and hexagon photographs has high accuracy and has a high correlation with the topographic map of 1.50000. In the third step, the unsupervised classification of binary mask method, in order to detect lake changes using corona and hexagon photographs, acceptablely identified the altered and unchanged pixels, so that 11 hectares of lake surface had the most changes. Finally, in the fourth step, it was found that the mean shift segmentation algorithm and threshold worked better by applying the corona, hexagon, and Aster events to extract the water surface, and in the meantime, the corona image performed better due to its higher resolution. The above results showed that Zaribar Lake decreased by 6.5% from 1969 to 2019 and the findings have a high correlation with the product of the ester aquifer. In general, the findings of this study show the potential of using digital image processing methods for corona and hexagon data to monitor and detect changes in lakes.

The aim of this study was to analyze the temporal and spatial nature of dust storms during the period 2016 to 2018 in Kermanshah Using the HYSPLIT routing model and the MCD19A2 product, the Modis sensor was performed in the Google Earth web engine.In order to route the origin of dust particles, the Lagrangian method of HYSPLIT model was used in 48 hours before the occurrence of dust phenomenon in Kermanshah at three altitude levels of 200, 1000 and 1500 meters.Findings from HYSPLIT model tracking maps indicate that the general route for dust transfer to the study area is the north-west-southeast route with the origin of the deserts of Iraq and Syria at three altitudes of 200, 1000 and 1500 meters. On June 17, 2016 and October 27, 2018, as well as the southwest-west route originating in Kuwait, Northern Saudi Arabia and part of Iraq on November 2, 2017.The results of the maps obtained from the MCD19A2 product of the Modis sensor, especially the maps of periodicity, cumulative concentration, spatial variation and the highest AOD map, show a high correlation with the routed maps extracted from the HYSPLIT model. In general, based on the findings of the maps extracted from the product MCD19A2, Modis sensor during the period 2016 to 2018 in Kermanshah, the central and eastern regions have always been more affected by dust storms than in other parts of the city. On average, they were more exposed to dust pollution than other parts of the city. In this regard, the final results show a high correlation between the actual PM10 data and the AOD values derived from the MODIS sensor.

Therefore, the aim of this dissertation is to Accuracy Assesment the of Road extraction methods based on deep neural networks using radar and optical satellite images of Sentinel series with equal conditions, ie uniformity in terms of study area, spatial resolution and sample Are educational. Research method: In this study, radar images of Sentinel 1 and optical images of Sentinel 2 in Tehran as test data and from 7 cities of Mashhad, Isfahan, Shiraz, Tabriz, Kermanshah, Urmia and Baghdad (for optical images) and the same images with the difference that instead of Kermanshah The image of Beijing, China (for radar images) was used as training and validation data. In the meantime, after preparing and labeling all the pixels related to the road complication, the images are converted into 256 × 256 parts and after separating the inappropriate parts, for test, training and validation data, respectively. 135, 1500 and 100 image pieces were obtained for each of the total images of Sentinel 1 and Sentinel 2. Finally, refined deep residual convolutional neural network (RDRCNN) and fully Convolution networks (FCNs) were used to train and extract the road. Findings: The results show that in the optical image outputs, both RDRCNN and FCN models have detected and extracted the road better than radar images. The findings also show that the RDRCNN model in radar images, and the FCN model in optical images, performed better both visually and in terms of evaluation metrics; So that for FCN model (radar images) the criteria of Recall are 53.21%, Precision 46%, F1 score 49.35 and overall accuracy 91.33%, RDRCNN model (radar images) Recall 57.66%,Precision 51.29%, F1 score 54.43% and Overall accuracy 92.78%, FCN model (optical images) Recall 82.92%, Precision 77.67%, F1 score 80.20% and overall accuracy 96.30% and finally RDRCNN model (optical images) Recall 81.43%, Precision 74.37%, score F1 77.74% and overall accuracy 95.71%.was obtained. Conclusion: In general, the findings of this study show the potential of using DNN methods to extract the road from images with medium spatial resolution, especially from Sentinel 2 optical images. In the meantime, due to low spatial resolution, spectral integration and Different redistribution, and low road width, especially in radar images, there were Problems arose during network labeling and training that overshadowed the final accuracy of the output.