LEE- Uydu Haberleşme ve Uzaktan Algılama-Doktora

Bu koleksiyon için kalıcı URI

http://hdl.handle.net/11527/19488

Gözat

Çıkarma tarihi ile LEE- Uydu Haberleşme ve Uzaktan Algılama-Doktora'a göz atma

Dizinde bir noktaya atla:
  • Öge
    Single-frame and multi-frame super-resolution on remote sensing images via deep learning approaches
    (Graduate School, 2022-07-29) Wang, Peijuan ; Sertel, Elif ; 705172003 ; Satellite Communication and Remote Sensing
    As a quite significant computer vision task, image super-resolution (SR) has been widely applied in remote sensing (RS), medical imaging, video surveillance, and biometrics. Image SR aims to restore high-resolution (HR) images by enhancing the spatial, spectral, or temporal resolution of the low-resolution (LR) inputs. In recent years, great efforts have been made for improving the SR approaches. One of the approaches is to design deeper networks. Nevertheless, this greatly increases computation and memory consumption. As a result, some mechanisms (such as cascading networks, attention mechanisms, and back projection) are proposed to improve the performance and the training process of the complex networks. Satellite imagery can be seen in various fields, namely Land cover/Land use classification, road and building extraction, observation of climate, and earthquake prediction. However, in some cases, the resolution of satellite images can not meet the application requirements due to the technology and cost limitations during the satellite design; therefore, the improvement of image resolution might be necessary. Since it is not possible to upgrade the equipment onboard the launched satellite, the software-based SR algorithms are deserved to be explored in RS fields. This thesis aims to strengthen the spatial resolution of optical satellite imageries by using deep learning (DL) methods. Generally, image SR algorithms can be categorized as single-frame image SR (SFSR) and multi-frame image SR (MFSR). The inputs of the SFSR can be a single LR image. While multi-frame image SR aims to restore HR image by using multiple LR images, which can be obtained under different conditions and at different angles. Recently, great contributions have been made to improve the SR methods including two aspects: (1) increasing the value of PSNR (Peak-Signal-Noise-Ratio); (2) improving the image quality perceptually. Nevertheless, some algorithms obtain a high PSNR but with a low perceptual quality which is more important to human perception. Therefore, this thesis has the following objectives: (1) Explore a perceptual-driven approach to enhance the SR image quality visually on single-frame and multi-frame RS imageries; (2) Explore Generative Adversarial Network (GAN)-based models for single-frame and multi-frame RS imagery super-resolution task to fulfill the multi-scale problem and blind to the degradation model; (3) Explore an image fusion method that can generate an arbitrary size of the super-resolved image rather than a small patch. This thesis firstly gives an overview of single-frame and multi-frame RS image SR methods. The single-frame RS image SR methods are briefly classified into supervised and unsupervised methods. The former mainly includes Convolutional Neural Networks (CNN)-based, GAN-based, attention-based, and Back-projection based methods. In addition, the commonly used attention mechanisms including self-attention, channel attention, spatial attention, mixed high-order attention (MHOA), non-local attention (NLA), and non-local sparse attention (NLSA) are also introduced. Moreover, loss functions including pixel-wise loss, perceptual loss, adversarial loss, and cycle consistency loss are presented. For the single-frame RSISR, firstly, an attention CNN-based SR method is proposed. Although CNN-based algorithms have made outstanding achievements in computer vision tasks, the traditional CNNs methods treat the abundant low-frequency information included in the LR inputs equally across channels. Attention-guided algorithms play a vital role in the informative features extraction in various tasks including image SR. With the application of the attention mechanism, the proposed CNN-based method can further learn the deeper relationships among the different channels. Instead of simply integrating the attention module with the residual blocks, a Layer Attention Module (LAM) and Spatial Attention Module (SAM) are proposed to further learn the relationships among the Residual Groups (RG). Moreover, the perceptual loss function is adopted in the training process to enhance the generated image quality perceptually, and Random down-sampling is applied to strengthen the model's generalization ability. Secondly, an attention GAN-based super-resolution method is explored for the single-frame RS images. CNN-based methods have made great contributions to increasing the value of PSNR/SSIM. Nevertheless, the generated outputs tend to be overly smooth and blurry. GAN can generate more realistic images than normal CNN-based methods and has been introduced to single image super-resolution (SRGAN, ESRGAN, EEGAN). Standard GANs only function on spatially local points in LR feature maps. The attention mechanism can directly learn the long-range dependencies in the feature maps both in the generator and discriminator in a powerful way. By applying the attention mechanism, the network allocates attention based on the similarity of color and texture. Therefore, based on ESRGAN, an attention GAN-based method is for the single-frame RS image SR. The ESRGAN was mainly improved from two aspects: (1) we further improved the architecture of the residual blocks by adding more skip connections; (2) we add attention modules to the residual blocks for further feature extraction. Moreover, instead of working on aerial photographs or low-resolution and medium-resolution satellite images, we are focusing on the Very High-Resolution (VHR) satellite imageries, such as the Pleiades, and Worldview-3. The spatial resolutions of the multispectral images for the Pleiades, and Worldview-3 are 2m, and 1.24m, respectively. Furthermore, for the attention CNN-based method, we evaluated the method on the Pleiades and Worldview-3 datasets with scaling factors of 2, 4, and 8. For the attention GAN-based method, we evaluated the method on the Pleiades and Worldview-3 datasets with a scaling factor of 4. The experimental results show the attention-based method can provide better perceptual results both quantitatively and qualitatively. At last, we proposed an attention GAN-based method for the multi-frame RS image SR. Firstly, we introduced an attention mechanism to the Generator and proposed a space-based network that worked on every single frame for better temporal information extraction. Secondly, we proposed a novel attention module for better spatial and spectral information extraction. Thirdly, we applied an attention-based discriminator to enhance the discriminator's discriminative ability. Finally, the experimental results on the SpaceNet7 dataset and Jilin-1 dataset exhibit the superior of the proposed model both quantitatively and qualitatively.
  • Öge
    District-based urban sprawl monitoring and modelling using CA-Markov model: application in two mega cities
    (Graduate School, 2022-12-12) Azabdaftari, Anali ; Sunar, Filiz ; 705152008 ; Satellite Communication and Remote Sensing
    There is a well-known fact that the population all around the world is growing so fast and current estimation and trends forecast a continuous rise in world population in the future. Urbanization occurs due to the high rate of immigration from rural areas to urban areas resulting in urban growth and forcing urban development toward the border of the city. The unstructured growth of urban areas can lead to urban sprawl. There are various methods and data to observe and quantify the changes induced through urban expansion. Remote sensing and Geographical Information Systems (GIS) can be named as one of the most helpful tools in the detection of the spatio-temporal dynamics of patterns of urban growth and Land Use/Land Cover (LULC) changes at local, regional, and global scales. The main aims and objectives of this study are to evaluate the spatio-temporal pattern of urban growth over time, to predict future urban expansion, and draw attention to district- based urban expansion. Thus, in this research, the most urbanized districts of Istanbul and Sydney, were selected to study. Identification of potential areas for future urban development is important in the selection of study areas. Arnavutköy district of Istanbul has a high potential of urban development due to the construction of the Istanbul airport. On the other hand, Hills Shire Local Government Area (LGA) was selected due to the improvement in the transportation system which can affect future urban development. This study consists of two main parts: urban expansion analysis and urban expansion modelling. As a first step in urban expansion analysis multitemporal Landsat satellite images of Arnavutköy district (18-July-1997, 31-August-2007, and 25-July-2017) and Hills Shire LGA (08-December-1996, 05-January-2007, and 19-January-2018) were acquired. After pre-processing of obtained satellite images of both study areas, images were classified into five LULC classes using Maximum Likelihood supervised classification algorithm. Accuracy assessment was also done to evaluate the accuracy of LULC maps of each study area. The overall accuracy values of Arnavutköy district in 1997, 2007, and 2017 were found as 91%, 86%, and 88%, respectively. Likewise, in Hills Shire LGA, the overall accuracy values were acquired as 92%, 95%, and 91% in 1996, 2007, and 2018, respectively. The post-classification change detection technique was employed to identify conversions from one LULC class to another LULC class. The results of change detection analysis showed that the built-up area in Arnavutköy district grew up to 197% while Hills Shire LGA expanded by 78% through the study period. In both study areas, in opposition to the built-up area, forest, and agricultural area experienced a downward trend and lost a lot of hectares during the period of study. To analyze the impact of urban expansion on the Land Surface Temperature (LST) of the surrounding environment, the correlation analysis between satellite retrieved LST and the built-up area in the selected buffer zones were carried out. The results revealed that, with the increase in built-up areas, the LST has also increased over the study period in Arnavutköy district and Hills Shire LGA. This indicates that urban expansion can cause to increase in the temperature of the urban areas which may also have negative effects on the climate changes. The degree of urban sprawl and the spatial pattern of urban expansion were measured using Shannon's entropy approach for each study area. The results indicated that, in both cases, urban areas had undergone a considerable urban sprawl. Furthermore, it was observed that built-up areas mostly expanded toward the North direction in both study areas, which shows the role of human-induced activities in urban expansion. As a next step, selected landscape metrics at the class level were used to better understand the urban structure. Accordingly, six metrics namely, Class Area (CA), Number of Patches (NP), Edge Density (ED), Largest Patch Index (LPI), Patch Density (PD), and Area Weighted Mean Patch Fractal Dimension (AWMPFD/FRAC-AM) were selected to describe the pattern of urban growth. The CA metric, which describes what portion of the landscape is composed of a specific patch type, was used to calculate the absolute size of the built-up area. The NP metric is a simple measure of the discontinuous built-up patch in the landscape and used as a indication on the fragmentation degree of the urban area. The PD metric is a fragmentation metric that reveals the number of patches per unit area. The absolute size of urban patches was determined using the LPI metric, which measures the percentage of entire landscape area covered by the largest patch. The ED and FRAC_AM metrics which measure the patch shape complexity were used to analyze the the shape irregularity of urban areas and spatial heterogeneity of the landscape. Metrics results which provide useful information about the aggregation and dispersion of urban also confirmed an expansion in Arnavutköy district and Hills Shire LGA. The results also showed that Arnavutköy district is more fragmented and Hills Shire LGA has a more compact urban expansion process. These results also confirm the findings of the change detecion and Shannon's entropy analysis. In the final step of this research, the second part of this study which was urban expansion modelling was carried out. Several modelling methods such as; Cellular Automata (CA), Markov Chain (MC), Logistic Regression and Artificial Neural Network have been used by many resarchers to evaluate the current and possible future state of the urban areas. The Cellular Automata model is a computational method that can simulate the process of growth by using a set of simple rules in analyzig complex systems. This model is comprised of a set of rules that define the interaction of cells and collection of variables that lead to the investigation of various urban forms. The acpability of CA to simulate urban expansion, land use change, and population growth has become appropriate for simulating complex spatial systems. The Markov Chain model is a stochastic process that experiences transitions from one state to another considering certain probabilistic rules and predicts future changes based on the past. Logistic Regression creates a model that can best describe the relationship between dependent and independent variables using the least number of variables. The Artificial Neural Networks model is a powerful tool that models patterns using a machine learning approach. ANN output values vary from 0 to 1. In urban modelling, a value of 1 represents the greatest potential for future urban growth, while a value of 0 represents the least potential for future urban growth. Each of these methods has its own advantages and disadvantages, however, the integrated models have proved to be more accurate. The integrated CA–Markov model does multiple principle evaluation functions which combines Cellular Automata and Markov Chain models together. The MC is a widely known model, which predicts future change by considering the past, however it does not take into account the spatial distribution of the categories, while CA is a dynamic model which detects the spatial location of changes. To this end, the integrated CA-Markov model was used for predicting future urban expansion. In the modelling, first, the suitability map of each LULC class were created using Multi Criteria Evaluation (MCE) method. Second, transition area and transition probability matrices were determined by using the Markov Chain model. After that, the LULC map of Arnavutköy district in 2017 and LULC map of Hills Shire LGA in 2018 were simulated using CA_Markov model. Subsequently, the model was validated by comparing the actual and simulated LULC maps of Arnavutköy district and Hills Shire LGA. After obtaining the reasonable results of validation which were above 80% for Arnavutköy district and Hills Shire LGA, the suitability maps, transition area and transition probability matrices were created to predict the future LULC maps of 2050 for both study areas. To better evaluate the urban expansion of both study areas in the future, the area statistics of the LULC map of 2050 were calculated and the results revealed that urban growth will continue in the future by increasing 45% and 51% in Arnavutköy district and Hills Shire LGA, respectively. On the other hand, the forest area will decrease in Arnavutköy district by 26%, while in Hills Shire LGA 30% of the forest will be lost throughout the 32 years. This study presents a significant contribution to land use modelling, which logically integrates remote sensing data and ancillary data to be used as an input in CA- Markov model to successfully simulate and predict the temporal and spatial changes of LULC classes. Moreover, it has been determined that using different spatial pattern analysis such as Shannon's entropy and landscape metrics can help to better evaluate the urban expansion and model the future urban growth. The findings of this study can help to better evaluate the dynamic nature of built-up areas in both Arnavutköy district and Hills Shire LGA. The simulated future LULC maps can be used as a system of early warning to recognize the consequences of land use changes. The simulation results can also be investigated as a strategic guide for urban planners to better evaluate a complex system and create optimized land use managemnet that can better control urban expansion and environmental protection. In conclusion, this study highlights the importance of district-based urban expansion analysis. However, there are many research studies on urban expansion in big cities all over the world, but there are relatively few studies that consider district-based urban expansion in megacities. Most researchers mainly focus on the physical and socio-economic issues of districts neglecting the role of urbanization in the district. In this regard, considering this issue is very important, as district-based analysis of the causes and effects of urban sprawl can provide urban planners with more detailed insights to analyze and understand the nature of this phenomenon on a city basis.