Vision Transformers (ViTs) have demonstrated strong performance in hyperspectral image (HSI) classification; however, their robustness is highly sensitive to patch size. This study investigates the impact of spatial patch size on clean accuracy and adversarial robustness using a standard ViT and a Channel Attention Fusion variant (ViT-CAF). Patch sizes from 1 × 1 to 19 × 19 are evaluated across four benchmark datasets under FGSM, BIM, CW, PGD, and RFGSM attacks. Descriptive results show that smaller patches, particularly 1 × 1 and 3 × 3, generally yield higher adversarial accuracy, while larger patches amplify localized perturbations and degrade robustness. Parameter analysis indicates that patch-size-dependent variations arise mainly from the embedding layer, with the Transformer backbone remaining fixed, confirming that robustness differences are driven primarily by spatial context rather than model capacity. These findings reveal a trade-off between spatial granularity and adversarial resilience and provide guidance for patch size selection in ViT-based HSI applications.

We present MMLSv2, a dataset for landslide segmentation on Martian surfaces. MMLSv2 consists of multimodal imagery with seven bands: RGB, digital elevation model, slope, thermal inertia, and grayscale channels. MMLSv2 comprises 664 images distributed across training, validation, and test splits. In addition, an isolated test set of 276 images from a geographically disjoint region from the base dataset is released to evaluate spatial generalization. Experiments conducted with multiple segmentation models show that the dataset supports stable training and achieves competitive performance, while still posing challenges in fragmented, elongated, and small-scale landslide regions. Evaluation on the isolated test set leads to a noticeable performance drop, indicating increased difficulty and highlighting its value for assessing model robustness and generalization beyond standard in-distribution settings. Dataset will be available at: https://github.com/MAIN-Lab/MMLS_v2

The reliability of object detection models fundamentally depends on training data quality, yet many benchmarks suffer from missing, confused, and inaccurate bounding boxes. This issue is particularly severe in the underwater domain, where visual degradations amplify annotation errors. The Real-world Underwater Object Detection (RUOD) dataset is a widely used benchmark that provides diverse real-world underwater scenes, but its annotations reflect the inherent difficulty of comprehensive annotation in such environments. To address this limitation, we present RUOD-R, a high-quality re-annotated version of RUOD developed using a rigorous protocol that combines image enhancement and quality-assured professional annotation. RUOD-R retains the original images, ensuring that performance differences stem solely from annotation quality. RUOD-R increases the total annotated instances by 3.5 & times;, with small object annotations increasing by over 140 & times;. Our comparative analysis reveals that many valid object instances, particularly small and visually degraded targets, were not annotated in the original dataset, and that matched boxes exhibit localization inaccuracy (mean IoU of 0.83 with RUOD-R). We evaluate detection performance using Faster R-CNN, RetinaNet, YOLOv11, and DynYOLO, spanning general-purpose and underwater-domain-specific architectures, on both datasets. Results show that RUOD-R is a substantially more challenging benchmark due to the increased object density and the predominance of small, occluded, and visually degraded targets. A controlled evaluation on paired images where both annotation sets contain the same objects suggests that annotation density, rather than coordinate differences, is the primary factor behind the performance gap. A flip-rate analysis reveals that up to 38% of detections counted as false positives on the original dataset are correct detections of valid objects. RUOD-R thus provides a more reliable benchmark for advancing underwater object detection. The new annotations and related metadata can be found at https://github.com/RSSL-MTU/RUOD-R

Underwater image enhancement has led to mixed performance when used in conjunction with object detectors. Some works report an improvement in object detection performance when used in conjunction with image enhancement, while other works show that enhancement degrades detection performance. In this work, we identify and attempt to quantify a confounding factor to reconcile these contradictory results: we show that poor image quality leads to low quality labels, which confounds performance assessments done with respect to these noisy labels. Application of image enhancement during the human labeling procedure recovers previously missed labels for less-biased performance assessment. We find that on the test set of the Rethinking general Underwater Object Detection (RUOD) dataset, re-annotation of enhanced images shows a mean increase of 9 labels per image, which leads to a 5 percentage point increase in precision during object detection with a YOLO-NAS model.

In recent years, Single-Image Super-Resolution (SISR) has gained significant attention in the geoscience and remote sensing community for its potential to improve the resolution of low-quality underwater imagery. This paper introduces MIMAR-Net (Multiscale Inception-based Manhattan Attention Residual Network), a new deep learning architecture designed to increase the spatial resolution of input color images. MIMAR-Net integrates a multiscale inception module, cascaded residue learning, and advanced attention mechanisms, such as the MaSA layer, to capture both local and global contextual information effectively. By utilizing multiscale processing and advanced attention strategies, MIMAR-Net allows us to handle the complexities of underwater environments with precision and robustness. We evaluate the model on three popular underwater image datasets, namely UFO-120, USR-248, and EUVP, and perform extensive comparisons against state-of-the-art methods. Experimental results demonstrate that MIMAR-Net consistently outperforms existing approaches, achieving superior qualitative and quantitative improvements in image quality, making it a reliable solution for underwater image enhancement in various challenging scenarios.

The advent of deep learning (DL) has significantly advanced artificial intelligence, driving notable progress in fields such as language translation, object recognition, and recommendation systems. Despite these successes, the computational complexity of advanced DL models continues to impede their practical deployment, particularly in clinical settings. Addressing this challenge, we introduce NExNet Seg, the Neuron Expansion Network for Medical Image Segmentation. Inspired by Progressively Expanded Neuron (PEN) structures and Manhattan Self-Attention (MaSA) mechanisms, NExNet Seg achieves exceptional accuracy with high parameter efficiency. It substantially reduces computational overhead, making it especially suitable for segmentation tasks in skin lesions and colorectal cancer using dermoscopic and endoscopic imagery. Empirical evaluations conducted on publicly available datasets, including PH2, ISIC (2016-2018), CVC Clinic, and Kvasir, demonstrate that NExNet Seg consistently outperforms current state-of-the-art methods in terms of accuracy, computational efficiency, and generalization capability. These results highlight its potential as an effective, scalable solution for clinical deployment in medical image segmentation. Code available at: https://github.com/MAINLab/NExNet_Seg

Underwater images often suffer from severe color distortion, low contrast, and reduced visibility, motivating the widespread use of image enhancement as a preprocessing step for downstream computer vision tasks. However, recent studies have questioned whether enhancement actually improves object detection performance. In this work, we conduct a comprehensive and rigorous evaluation of nine state-of-the-art enhancement methods and their interactions with modern object detectors. We propose a unified evaluation framework that integrates (1) a distribution-level quality assessment using a composite quality index (Q-index), (2) a fine-grained per-image detection protocol based on COCO-style mAP, and (3) a mixed-set upper-bound analysis that quantifies the theoretical performance achievable through ideal selective enhancement. Our findings reveal that traditional image quality metrics do not reliably predict detection performance, and that dataset-level conclusions often overlook substantial image-level variability. Through per-image evaluation, we identify numerous cases in which enhancement significantly improves detection accuracy—primarily for low-quality inputs—while also demonstrating conditions under which enhancement degrades performance. The mixed-set analysis shows that selective enhancement can yield substantial gains over both original and fully enhanced datasets, establishing a new direction for designing enhancement models optimized for downstream vision tasks. This study provides the most comprehensive evidence to date that underwater image enhancement can be beneficial for object detection when evaluated at the appropriate granularity and guided by informed selection strategies. The data generated and code developed are publicly available.

Underwater image enhancement is often perceived as a disadvantageous process to object detection. We propose a novel analysis of the interactions between enhancement and detection, elaborating on the potential of enhancement to improve detection. In particular, we evaluate object detection performance for each individual image rather than across the entire set to allow a direct performance comparison of each image before and after enhancement. This approach enables the generation of unique queries to identify the outperforming and underperforming enhanced images compared to the original images. To accomplish this, we first produce enhanced image sets of the original images using recent image enhancement models. Each enhanced set is then divided into two groups: (1) images that outperform or match the performance of the original images and (2) images that underperform. Subsequently, we create mixed original-enhanced sets by replacing underperforming enhanced images with their corresponding original images. Next, we conduct a detailed analysis by evaluating all generated groups for quality and detection performance attributes. Finally, we perform an overlap analysis between the generated enhanced sets to identify cases where the enhanced images of different enhancement algorithms unanimously outperform, equally perform, or underperform the original images. Our analysis reveals that, when evaluated individually, most enhanced images achieve equal or superior performance compared to their original counterparts. The proposed method uncovers variations in detection performance that are not apparent in a whole set as opposed to a per-image evaluation because the latter reveals that only a small percentage of enhanced images cause an overall negative impact on detection. We also find that over-enhancement may lead to deteriorated object detection performance. Lastly, we note that enhanced images reveal hidden objects that were not annotated due to the low visibility of the original images.

We introduce an Attention-based Spectral Profile Representation (ASPR) leveraging multi-head attention mechanisms for pixel-wise hyperspectral image classification. By incorporating attention-driven importance scores, ASPR effectively identifies and preserves essential spectral regions, discarding redundant bands. By leveraging attention scores of spectral signatures, the proposed method effectively captures key spectral features, reduces redundancy, and improves classification accuracy. Experimental findings indicate that ASPR offers superior performance compared to conventional and current deep learning approaches for classifying hyperspectral images.

Classifying mine tailings impoundments is challenging due to the complexity and variability of these landscapes. A major hurdle is the scarcity of annotated real-world images, leading to issues such as underfitting and class imbalance in machine learning models. To address these challenges, we incorporate synthetic data during training to improve classification accuracy. This research aims to boost the accuracy and generalization of the model using real-world Sentinel-1 and Sentinel-2 datasets and synthetic images generated by generative models. We evaluated several image generation techniques, such as Conditional Generative Adversarial Networks, Variational Autoencoders, PixelCNN and Diffusion Models. These methods are assessed for their ability to produce realistic, high-quality synthetic images to augment datasets and thus improve pattern recognition tasks. In addition, we can accurately capture the structure of mine tailings landscapes by integrating amplitude information from synthetic aperture radar and multispectral optical data. Through our analysis of these approaches, we aim to reduce the reliance on large labeled datasets, mitigate mislabeling, and enhance both the accuracy and generalization of classification models. This work contributes to more effective monitoring of mine tailings impoundments, ultimately supporting better environmental risk assessments and the advancement of sustainable mining practices.

Surgical image and video applications using endoscopic datasets have been actively investigated to develop advanced surgical assistant systems. These applications are particularly crucial for understanding surgical scenes during procedures. Specifically, segmentation techniques allow for identifying anatomical structures and surgical instruments, while quality control methods refine surgical techniques, and action recognition aids in discerning surgical steps. A significant improvement in performance across different downstream tasks has been achieved due to the advancements in deep neural networks and the expansive training dataset available. However, the exploration of surgical action recognition remains limited. Existing methods face challenges in real-world settings, mainly due to the lack of adaptability in a dynamic imaging environment. In this study, we present a framework for surgical action recognition in endoscopic datasets by leveraging video-masked autoencoders (VideoMAE), which has shown promise in video dataset analysis with minimal datasets. Additionally, we incorporate a temporal data augmentation technique to represent diverse imaging conditions and resolve the issue of using single-source data with low quality. For our experiments, we utilize VideoMAE v2 pre-trained on Unlabeled Hybrid datasets and fine-tune the model on the CholecT45 dataset for validation. Our proposed method shows the effectiveness of using the VideoMAE structure with focal loss, particularly for action recognition tasks in surgical scenarios.

Computer Organization and Design is a pivotal course for learning an instruction set architecture (ISA), the microarchitecture which implements it, and determining a system’s overall performance. However, many concepts in this course have been found daunting to understand. To illustrate those concepts, an interactive desktop application is being developed using the Unity game engine for the purpose of enhancing student learning. This tool covers material that is traditionally covered in the course and adds security concepts to make students aware of the security issues associated with micro-architecture.

Underwater imagery often exhibits significant degradation and poor quality as compared to outdoor imagery. To compensate for this, Single-Image Super-Resolution (SISR) and enhancement algorithms are used to lessen this degradation and produce high-resolution images. In this study, we apply state-of-the-art Simultaneous Enhancement and Super-Resolution (SESR) and SISR models to different sets of downscaled images from the comprehensive RUOD dataset. We then conduct a qualitative and quantitative analysis of the upscaled and enhanced images using standard underwater image quality metrics (IQMs). Subsequently, we evaluate the robustness of the state-of-the-art YOLO-NAS detector against image sets with varying downscaled spatial resolutions. Lastly, we examine the impact that the SISR and SESR models has on YOLO-NAS detector performance. The findings reveal a decline in the detection performance on the downscaled test images and a further decline on the upscaled and enhanced images produced by SISR and SESR models, suggesting a negative relationship between such models and detection.