<p dir="ltr">Land cover classification has always been a crucial topic in the remote sensing domain. Utilizing data collected by unmanned aerial vehicles and satellites, researchers can detect land degradation, monitor environmental changes, and provide insights for urban planning. Recent advancements in large multi-modal models have enabled open-vocabulary classification, which is particularly beneficial in this field. Becuase of the pre-training method, these models can perform zero-shot inference on unseen data, significantly reducing the costs associated with data collection and model training. This open-vocabulary feature of large-scale vision-language pre-training aligns well with the requirements of land cover classification, where benchmark datasets in the remote sensing domain comprise various categories, and transferring results from one dataset to another through supervised learning methods is challenging.</p><p dir="ltr">In this thesis, the author explored the performance of zero-shot CLIP and linear probe CLIP to assess the feasibility of using the CLIP model for land cover classification tasks. Further, the author fine-tuned CLIP by creating hierarchical label sets for the datasets, leading to better zero-shot classification results and improving overall accuracy by 2.5%. Regarding data engineering, the author examined the performance of zero-shot CLIP and linear probe CLIP across different categories and proposed a categorization method for land cover datasets. In summary, this work evaluated CLIP's overall performance on land cover datasets of varying spatial resolutions and proposed a hierarchical classification method to enhance its zero-shot performance. The thesis also offers a practical approach for modifying current dataset categorizations to better align with the model.</p>

No description available.

No description available.

Recently, there is a tremendous amount of high resolution imagery that wasn’t available years ago, mainly because of the advancement of the technology in capturing such images. Most of the very high resolution (VHR) imagery comes in three bands only the red, green and blue (RGB), whereas, the importance of using such imagery in remote sensing studies has been only considered lately, despite that, there are no enough studies examining the usefulness of these imagery in urban applications. This research proposes a method to investigate high resolution imagery to analyse an urban area using UAV imagery for land use and land cover classification. Remote sensing imagery comes in various characteristics and format from different sources, most commonly from satellite and airborne platforms. Recently, unmanned aerial vehicles (UAVs) have become a very good potential source to collect geographic data with new unique properties, most important asset is the VHR of spatiotemporal data structure. UAV systems are as a promising technology that will advance not only remote sensing but GIScience as well. UAVs imagery has been gaining popularity in the last decade for various remote sensing and GIS applications in general, and particularly in image analysis and classification. One of the concerns of UAV imagery is finding an optimal approach to classify UAV imagery which is usually hard to define, because many variables are involved in the process such as the properties of the image source and purpose of the classification. The main objective of this research is evaluating land use / land cover (LULC) classification for urban areas, whereas the data of the study area consists of VHR imagery of RGB bands collected by a basic, off-shelf and simple UAV. LULC classification was conducted by pixel and object-based approaches, where supervised algorithms were used for both approaches to classify the image. In pixel-based image analysis, three different algorithms were used to create a final classified map, where one algorithm was used in the object-based image analysis. The study also tested the effectiveness of object-based approach instead of pixel-based in order to minimize the difficulty in classifying mixed pixels in VHR imagery, while identifying all possible classes in the scene and maintain the high accuracy. Both approaches were applied to a UAV image with three spectral bands (red, green and blue), in addition to a DEM layer that was added later to the image as ancillary data. Previous studies of comparing pixel-based and object-based classification approaches claims that object-based had produced better results of classes for VHR imagery. Meanwhile several trade-offs are being made when selecting a classification approach that varies from different perspectives and factors such as time cost, trial and error, and subjectivity.       Classification based on pixels was approached in this study through supervised learning algorithms, where the classification process included all necessary steps such as selecting representative training samples and creating a spectral signature file. The process in object-based classification included segmenting the UAV’s imagery and creating class rules by using feature extraction. In addition, the incorporation of hue, saturation and intensity (IHS) colour domain and Principle Component Analysis (PCA) layers were tested to evaluate the ability of such method to produce better results of classes for simple UAVs imagery. These UAVs are usually equipped with only RGB colour sensors, where combining more derived colour bands such as IHS has been proven useful in prior studies for object-based image analysis (OBIA) of UAV’s imagery, however, incorporating the IHS domain and PCA layers in this research did not provide much better classes. For the pixel-based classification approach, it was found that Maximum Likelihood algorithm performs better for VHR of UAV imagery than the other two algorithms, the Minimum Distance and Mahalanobis Distance. The difference in the overall accuracy for all algorithms in the pixel-based approach was obvious, where the values for Maximum Likelihood, Minimum Distance and Mahalanobis Distance were respectively as 86%, 80% and 76%. The Average Precision (AP) measure was calculated to compare between the pixel and object-based approaches, the result was higher in the object-based approach when applied for the buildings class, the AP measure for object-based classification was 0.9621 and 0.9152 for pixel-based classification. The results revealed that pixel-based classification is still effective and can be applicable for UAV imagery, however, the object-based classification that was done by the Nearest Neighbour algorithm has produced more appealing classes with higher accuracy. Also, it was concluded that OBIA has more power for extracting geographic information and easier integration within the GIS, whereas the result of this research is estimated to be applicable for classifying UAV’s imagery used for LULC applications.

Large amounts of carbon are stored in soils of the northern circumpolar permafrost region. High-resolution mapping of this soil organic carbon (SOC) is important to better understand and predict local to global scale carbon dynamics. In this thesis, studies from five different areas across the permafrost region indicate a pattern of generally higher SOC storage in Arctic tundra soils compared to forested sub-Arctic or Boreal taiga soils. However, much of the SOC stored in the top meter of tundra soils is permanently frozen, while the annually thawing active layer is deeper in taiga soils and more SOC may be available for turnover to ecosystem processes. The results show that significantly more carbon is stored in soils compared to vegetation, even in fully forested taiga ecosystems. This indicates that over longer timescales, the SOC potentially released from thawing permafrost cannot be offset by a greening of the Arctic. For all study areas, the SOC distribution is strongly influenced by the geomorphology, i.e. periglacial landforms and processes, at different spatial scales. These span from the cryoturbation of soil horizons, to the formation of palsas, peat plateaus and different generations of ice-wedges, to thermokarst creating kilometer scale macro environments. In study areas that have not been affected by Pleistocene glaciation, SOC distribution is highly influenced by the occurrence of ice-rich and relief-forming Yedoma deposits. This thesis investigates the use of thematic maps from highly resolved satellite imagery (&lt;6.5 m resolution). These maps reveal important information on the local distribution and variability of SOC, but their creation requires advanced classification methods including an object-based approach, modern classifiers and data-fusion. The results of statistical analyses show a clear link of land cover and geomorphology with SOC storage. Peat-formation and cryoturbation are identified as two major mechanisms to accumulate SOC. As an alternative to thematic maps, this thesis demonstrates the advantages of digital soil mapping of SOC in permafrost areas using machine-learning methods, such as support vector machines, artificial neural networks and random forests. Overall, high-resolution satellite imagery and robust spatial prediction methods allow detailed maps of SOC. This thesis significantly increases the amount of soil pedons available for the individual study areas. Yet, this information is still the limiting factor to better understand the SOC distribution in permafrost environments at local and circumpolar scale. Soil pedon information for SOC quantification should at least distinguish the surface organic layer, the mineral subsoil in the active layer compared to the permafrost and further into organic rich cryoturbated and buried soil horizons. / <p>At the time of the doctoral defense, the following papers were unpublished and had a status as follows: Paper 3: Manuscript. Paper 4: Manuscript.</p>

Changes in land cover and fish assemblage structure were assessed across two spatial and temporal scales in the Piedmont and Coastal Plain regions of the Chesapeake Bay watershed in Virginia. A long-term, local study (1953 to 2014) on the Tuckahoe Creek watershed used digitized aerial photography and satellite images (Landsat 5 TM and Landsat 8 OLI/TIRS) to quantify land cover change for five nested catchments in 1953, 1990, and 2014. Instream fish collections from 1958, 1990, and 2014 were utilized to assess a variety of fish assemblage metrics for each of the five catchments, and analyses were performed to assess associations between changes in land cover and changes in fish assemblage structure across all three time periods. A short-term, regional study assessed 21 catchments in the region using 1997 Landsat 5 TM satellite images and 2014 Landsat 8 OLI/TIRS satellite images to quantify land cover change. Fish collections from 1995-1999 and 2014 were utilized to assess a variety of fish assemblage metrics from samples taken at instream sites for each of the 21 catchments. Analyses were performed to discover any associations between changes in land cover and changes in fish assemblage structure from a regional perspective. This study found that there were significant changes in land cover over all study periods in the Tuckahoe Creek watershed and that land cover changes were correlated to changes in fish assemblage structure over the long-term study. Regionally, there were significant changes in land cover, with no correlation to changes in fish assemblage structure found. The data suggests that anthropogenic alterations to the landscape have had long-term effects on fish assemblage structure in Tuckahoe Creek, but the results from the short-term assessments did not detect a relationship between land cover changes and changes in fish assemblage structure. It is possible that the fish communities were already established in moderately degraded catchments by the 1990s due to previous anthropogenic stressors.

No description available.

No description available.

Remote sensing techniques are vital for early detection of several problems such as natural disasters, ecological problems and collecting information necessary for finding optimum solutions to those problems. Remotely sensed information has also important uses in predicting the future risks, urban planning, communication.Recent developments in remote sensing instrumentation offered a challenge to the mathematical and statistical methods to process the acquired information. Classification of satellite images in the context of land cover classification is the main concern of this study. Land cover classification can be performed by statistical learning methods like additive models, decision trees, neural networks, k-means methods which are already popular in unsupervised classification and clustering of image scene inverse problems. Due to the degradation and corruption of satellite images, the classification performance is limited both by the accuracy of clustering and by the extent of the classification. In this study, we are concerned with understanding the performance of the available unsupervised methods with k-means, supervised methods with Gaussian maximum likelihood which are very popular methods in land cover classification. A broader approach to the classification problem based on finding the optimal discriminants from a larger range of functions is considered also in this work. A novel method based on threshold decomposition and Boolean discriminant functions is developed as an implementable application of this approach. All methods are applied to BILSAT and Landsat satellite images using MATLAB software.

abstract: Large datasets of sub-meter aerial imagery represented as orthophoto mosaics are widely available today, and these data sets may hold a great deal of untapped information. This imagery has a potential to locate several types of features; for example, forests, parking lots, airports, residential areas, or freeways in the imagery. However, the appearances of these things vary based on many things including the time that the image is captured, the sensor settings, processing done to rectify the image, and the geographical and cultural context of the region captured by the image. This thesis explores the use of deep convolutional neural networks to classify land use from very high spatial resolution (VHR), orthorectified, visible band multispectral imagery. Recent technological and commercial applications have driven the collection a massive amount of VHR images in the visible red, green, blue (RGB) spectral bands, this work explores the potential for deep learning algorithms to exploit this imagery for automatic land use/ land cover (LULC) classification. The benefits of automatic visible band VHR LULC classifications may include applications such as automatic change detection or mapping. Recent work has shown the potential of Deep Learning approaches for land use classification; however, this thesis improves on the state-of-the-art by applying additional dataset augmenting approaches that are well suited for geospatial data. Furthermore, the generalizability of the classifiers is tested by extensively evaluating the classifiers on unseen datasets and we present the accuracy levels of the classifier in order to show that the results actually generalize beyond the small benchmarks used in training. Deep networks have many parameters, and therefore they are often built with very large sets of labeled data. Suitably large datasets for LULC are not easy to come by, but techniques such as refinement learning allow networks trained for one task to be retrained to perform another recognition task. Contributions of this thesis include demonstrating that deep networks trained for image recognition in one task (ImageNet) can be efficiently transferred to remote sensing applications and perform as well or better than manually crafted classifiers without requiring massive training data sets. This is demonstrated on the UC Merced dataset, where 96% mean accuracy is achieved using a CNN (Convolutional Neural Network) and 5-fold cross validation. These results are further tested on unrelated VHR images at the same resolution as the training set. / Dissertation/Thesis / Masters Thesis Computer Science 2016