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.

CycleGAN, a deep learning model for image conversion tasks is an extension of the Pix2Pix architecture [1] offering cycle consistency loss that offers image-to-image translation where you have two sets of images from different domains. Through deep learning in this way, it can generate high-quality and diverse image transformation results, which are widely used in the fields of style transfer, and domain adaptation. With generative AI taking center stage, the underlying principles of CycleGAN are used extensively across several domains including geospatial analysis. In this study we explore the potential of additional geospatial derivatives that when used can have the potential to enhance CycleGAN's domain adaptation. The analysis focuses on integrating digital surface model and NDVI to enhance image translation accuracy, maintain content integrity, and optimize CycleGAN for practical applications. Our results show that combining these two with multispectral bands yields better outcomes than isolating them from the training process, highlighting the effectiveness of this synergistic approach. © 2024 IEEE.

Polyethylene glycol (PEG) is a flexible, nontoxic polymer commonly used in biological and medical research, and it is generally regarded as biologically inert. PEG molecules of variable sizes are also used as crowding agents to mimic intracellular environments. A recent study with PEG crowders revealed decreased catalytic activity of Escherichia coli prolyl-tRNA synthetase (Ec ProRS), where the smaller molecular weight PEGs had the maximum impact. The molecular mechanism of the crowding effects of PEGs is not clearly understood. PEG may impact protein conformation and dynamics, thus its function. In the present study, the effects of PEG molecules of various molecular weights and concentrations on the conformation and dynamics of Ec ProRS were investigated using a combined experimental and computational approach including intrinsic tryptophan fluorescence spectroscopy, atomic force microscopy, and atomistic molecular dynamic simulations. Results of the present study suggest that lower molecular weight PEGs in the dilute regime have modest effects on the conformational dynamics of Ec ProRS but impact the catalytic function primarily via the excluded volume effect; they form large clusters blocking the active site pocket. In contrast, the larger molecular weight PEGs in dilute to semidilute regimes have a significant impact on the protein’s conformational dynamics; they wrap on the protein surface through noncovalent interactions. Thus, lower-molecular-weight PEG molecules impact protein dynamics and function via crowding effects, whereas larger PEGs induce confinement effects. These results have implications for the development of inhibitors for protein targets in a crowded cellular environment. © 2024 The Authors. Published by American Chemical Society.

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. © 2023 IEEE.