This study develops a site specific burn severity modelling using remote sensing techniques to develop severity patterns on vegetation and soil in the fire prone region of the Palo Verde National Park in Guanacaste, Costa Rica. Terrain physical features, soil cover, and scorched vegetation characteristics were examined to develop a fire risk model and to quantify probable burned areas. Spectral signatures of affected areas were captured through multi-spectral analysis; i.e., Normalized Burn Ratio (NBR), Landsat derived differenced Normalized Burn Ratio (dNBR) and relativized dNBR (RdNBR). A partial unmixing algorithm, Mixture Tuned Matched Filtering (MTMF) was used to isolate endmembers for scorched vegetation and soil. The performance of dNBR and RdNBR for predicting ground cover components was acceptable with an overall accuracy of 84.4% and Cohen’s Kappa 0.82 for dNBR and an overall accuracy of 89.4% and Cohen’s Kappa 0.82 for RdNBR. Landsat derived RdNBR showed a strong correlation with scorched vegetation (r² = 0.76) and moderate correlation with soil cover (r² = 0.53), which outperformed dNBR. The ecologically diverse and unique park area is threatened by wetland fires, which pose a potential threat to various species. Human induced fires by poachers are a common occurrence in such areas to gain access to these species. This paper aims to prioritize areas that are at a higher risk from fire and model spatial adaptations in relation to the direction of fire within the affected wetlands. This assessment will help wildlife personnel in managing disturbed wetland ecosystems.

Coronavirus disease (COVID-19) has had a significant impact on global health since the start of the pandemic in 2019. As of June 2022, over 539 million cases have been confirmed worldwide with over 6.3 million deaths as a result. Artificial Intelligence (AI) solutions such as machine learning and deep learning have played a major part in this pandemic for the diagnosis and treatment of COVID-19. In this research, we review these modern tools deployed to solve a variety of complex problems. We explore research that focused on analyzing medical images using AI models for identification, classification, and tissue segmentation of the disease. We also explore prognostic models that were developed to predict health outcomes and optimize the allocation of scarce medical resources. Longitudinal studies were conducted to better understand COVID-19 and its effects on patients over a period of time. This comprehensive review of the different AI methods and modeling efforts will shed light on the role that AI has played and what path it intends to take in the fight against COVID-19.

Image classification in remote sensing and geographic information system (GIS) data containing various land cover classes is essential for efficient and sustainable land use estimation and other tasks like object detection, localization, and segmentation. Deep learning (DL) techniques have shown tremendous potential in the GIS domain. While convolutional neural networks (CNNs) have dominated image analysis, transformers have proven to be a unifying solution for several AI-based processing pipelines. Vision transformers (ViTs) can have comparable and, in some cases, better accuracy than a CNN. However, they suffer from a significant drawback associated with the excessive use of training parameters. Using trainable parameters generously can have multiple advantages ranging from addressing model scalability to explainability. This can have a significant impact on model deployment in edge devices with limited resources, such as drones. In this research, we explore, without using pre-trained weights, how the inherent structure of vision transformers behaves with custom modifications. To verify our proposed approach, these architectures are trained on multiple land cover datasets. Experiments reveal that a combination of lightweight convolutional layers, including ShuffleNet, along with depthwise separable convolutions and average pooling can reduce the trainable parameters by 17.85% and yet achieve higher accuracy than the base mobile vision transformer (MViT). It is also observed that utilizing a combination of convolution layers along with multi-headed self-attention layers in MViT variants provides better performance for capturing local and global features, unlike the standalone ViT architecture, which utilizes almost 95% more parameters than the proposed MViT variant.

Classification of hyperspectral images is an important step of image interpretation from high spatial resolution imagery. Different studies demonstrate that spatial features can provide complementary information for increasing the accuracy of hyperspectral image classification. In this study, we evaluate different methods of spectral-spatial classification of hyperspectral images that are based on denoising methods using convolutional autoencoders. The resulting high-dimensional vectors of spectral features are classified by supervised algorithms such as support vector machine (SVM), maximum likelihood (ML), and random forest (RF). The experiments are performed on several widely known hyperspectral images that reveal a patch-based 3D convolutional autoencoder is more effective in reducing noise in the dataset and retaining spectral-spatial information. Random Forest classifier provides the highest classification accuracy across all the models.

The task of image retrieval is common in the world of data science and deep learning, but it has received less attention in the field of remote sensing. The authors seek to fill this gap in research through the presentation of a web-based landscape search engine for the US state of Wisconsin. The application allows users to select a location on the map and to find similar locations based on terrain and vegetation characteristics. It utilizes three neural network models—VGG16, ResNet-50, and NasNet—on digital elevation model data, and uses the NDVI mean and standard deviation for comparing vegetation data. The results indicate that VGG16 and ResNet50 generally return more favorable results, and the tool appears to be an important first step toward building a more robust, multi-input, high resolution landscape search engine in the future. The tool, called LSE Wisconsin, is hosted publicly on ShinyApps.io.

Image classification in Remote Sensing and Geographic Information Systems (GIS) containing various land-cover classes is essential for efficient and sustainable land-use estimation, and other tasks like object detection, localization and segmentation. Deep Learning (DL) techniques have shown a tremendous potential in the GIS domain. While Convolutional Neural Networks (CNN) have dominated most of the image analysis domain, a new architecture called transformers have proved to be a unifying solution for several AI-based processing pipelines. Vision Transformers (ViT), a variant of transformers can have comparable and in some cases better accuracy than a CNN. However, they suffer from a significant drawback associated with an excessive use of training parameters. In this research we explore several modifications in the vision transformer architectures, especially MobileViT that can be optimized while boosting accuracy. To verify our proposed approach these new architectures are trained on four land-cover datasets AID, EuroSAT, UC-Merced, and WHU-RS19. Experiments reveal that combination of lightweight convolutional layers including ShuffleNet along with depthwise separable convolutions and average pooling can reduce the trainable parameters by 17.85% and yet achieve higher accuracy than the base MobileViT. It is also observed that utilizing a combination of convolution layers along with multi-headed self attention layers in MobileViT variants provide better performance in capturing local and global features unlike the standalone ViT architecture that utilizes almost 95% more parameters than the proposed MobileViT variant.