Large language models (LLMs) have shown excellent performance in zero-shot learning using natural language prompts. However, in the domain of computer vision (CV), the paradigm of pretraining followed by finetuning remains dominant. The aim of this study is to reduce this gap by utilizing the capability of Self-Supervised Learning (SSL) in semantic understanding for zero-shot segmentation, without relying on human-provided labels or vision-language supervision. We introduce a novel evaluation framework that employs visual prompts, including a threshold and a query patch. This framework evaluates the ability of SSL models to derive concepts from observational data. Through this evaluation, we identify the strengths and limitations of SSL models in understanding semantics. Building on the insights from various SSL methods, we further propose the MMC approach to enhance the representations for objects, which integrates Masked image modeling, Momentum-based self-distillation, and global Contrastive learning. MMC achieves a better balance between the inter-object discriminability and the intra-object compactness of learned features. Our experiments on COCO, DAVIS-2017, PASCAL VOC, and ADE20K demonstrate outstanding performance of MMC’s representations.

Land cover classification using satellite imagery plays a crucial role in monitoring changes on the earth’s surface. This paper presents an analysis of the EuroSAT dataset using state-of-the-art deep learning models to benchmark the impact of additional bands on classification accuracy. The dataset consists of 27,000 images across 10 classes captured by the Sentinel-2 satellite, including RGB and multispectral bands. Performance evaluation was conducted using popular convolutional neural network models based on Resnet variants and Vision Transformer (ViT). The results show that the combination of all bands achieved the highest accuracy, with ResNet-152 achieving a validation accuracy of 96.63% on the multispectral dataset. Precision, recall, and F1 scores were also utilized to assess the models’ performance. The findings highlight the significance of incorporating additional bands for improved classification accuracy in satellite image analysis.

Climate data is a valuable resource for understanding past weather patterns, assessing long-term climate trends, and conducting climate-related research. However, most existing knowledge graphs for climate data rely heavily on the standardized (per W3C recommendations) SOSA/SSN ontology, which can help improve general data accessibility, but typically overlooks the analytical applications of multisource climate data. To further enhance the accessibility of heterogeneous data for climate data analytics, this paper extends the CA ontology and implements a virtual knowledge graph for analytical applications. We emphasize the importance of incorporating observation metadata and geospatial representation into analytical applications. Through our study, we demonstrate the applicability of the proposed ontological model in deriving the ETCCDI indices. An example of the formation of the annual maximum daily temperature is given. Furthermore, we showcase the potential of LinkedGeoData in providing a more comprehensive geographical context for accessing climate data within the knowledge graph, leveraging the proposed ontological modeling and linked data principles.

Reducing industrial greenhouse gas (GHG) emissions has become imperative for mitigating the adverse effects of climate change. Accurate measurement and monitoring of industrial smoke plumes, which are a significant source of GHG emissions, are crucial for effective emission control strategies. This paper addresses the prospect of utilizing satellite images to measure industrial smoke plumes and explores the effectiveness of various computer vision (CV) technologies in this context. The study focuses on examining both modern deep learning and traditional machine learning models for detecting and segmenting industrial smoke plumes in satellite images. While deep learning models have shown remarkable performance in various CV tasks, their ability to accurately segment smoke plumes in satellite images remains limited, with an average intersection over union (IOU) of no more than 60%. However, certain deep learning models, such as U-Net and AttU-Net, exhibit promising capabilities in identifying challenging types of noise, including clouds, white building surfaces, and snow, which traditional machine learning models struggle with. Employing deep learning models for industrial smoke plume detection proves advantageous, as all models achieve an approximate detection accuracy and F1-Score of 90%. The findings from this research serve as a valuable foundation for further advancements in developing advanced deep learning models specifically tailored to handle the identified types of noise.

Today's atmospheric data is generated swiftly as a result of the growth of IoT and sensor technologies and is available via data suppliers' RESTful APIs. However, sensor data mostly consists of live data streams including sensor observations, which are produced in a dispersed manner by several heterogeneous infrastructures, with little or no interoperability. RDF streams incorporating semantic data interoperability have arisen in last years and can be the foundation of intelligent semantic applications (e.g. semantic complex event processing). To enable semantic analysis of live atmospheric data streams, this article proposes a methodology for converting live data streams into Linked Data. The process leverages the most recent technologies for RML semantic mapping, ontology modeling, and Linked Data to extend the semantic usefulness of live atmospheric data, for example, by allowing for easy integration of atmospheric data streams with other live RDF streams.

With the active development of ICT and Internet technologies in climate research, individuals often need to gather different and disparate datasets and preprocess them in preparation for downstream data analysis in order to have a more full understanding of the challenges. This preparatory procedure is often lengthy due to the primary issue that data providers can-not ensure a homogeneous data format for data integration purposes. To overcome this problem, this study proposes enhancing existing relational climate data by layering a virtual knowledge graph on top of the original databases provided by various data vendors. The primary benefit of doing this is that data consumers are able to simply integrate climate data with other data sources using Linked Data principles, and climate data producers do not have to modify their data to conform to standard knowledge graph protocols.

The latest analytical models are becoming frequently used in meteorological science research. For instance, machine learning and deep learning models are being trained for weather forecasting. A solid machine learning model can give trustworthy findings that aid individuals in making weather-related decisions. However, the performance of analytical models is largely determined not only by the design of the model body but also by the input features. We address common issues of modern meteorological studies that take sensor data as the input for various analytical models. In contrast to the traditional practice of combining and preprocessing many fixed sensor data bulks to create an augmented dataset, we tunnel into remote knowledge graphs to fetch and augment the sensor data in a scalable way. As a consequence, we reduce the amount of time and storage space to preprocess diverse data in preparation for analytical models leveraging the high interoperability between knowledge graphs.

Recently ontologies have been exploited in a wide range of research areas for data modeling and data management. They greatly assists in defining the semantic model of the underlying data combined with domain knowledge. In this paper, we propose the Climate Analysis (CA) Ontology to model climate datasets used by remote sensing analysts. We use the data published by National Oceanic and Atmospheric Administration (NOAA) to further explore how ontology modeling can be used to facilitate the field of climatic data processing. The idea of this work is to convert relational climate data to the Resource Description Framework (RDF) data model, so that it can be stored in a graph database and easily accessed through the Web as Linked Data. Typically, this provides climate researchers, who are interested in datasets such as NOAA, with the potential of enriching and interlinking with other databases. As a result, our approach facilitates data integration and analysis of diverse climatic data sources and allows researchers to interrogate these sources directly on the Web using the standard SPARQL query language.

The cosine-based softmax loss functions greatly enhance intra-class compactness and perform well on the tasks of face recognition and object classification. Outperformance, however, depends on the careful hyperparameter selection. Adaptively Scaling Cosine Logits (AdaCos) tries to propose a parameter-free version by leveraging an adaptive scaling parameter. Nevertheless, the application of AdaCos is limited in specific domains because of improper approximation.In this paper, to promote intra-class compactness and interclass separability, we propose an Angular Gradient Margin Loss (ArcGrad) that generates a gradient margin by maximizing the angular gradient. Our work suggests that the margin parameter on cosine-based methods is not necessary, and the scaling parameter is inversely proportional to the margin. Furthermore, a stable and large gradient promotes better feature representation. In experiments, we test our method, as well as other methods enhancing discriminative information, on CIFAR and 15 datasets from UCI. Experimental results show ArcGrad consistently outperforms both on large and small scale problems and has the superiority in discriminative information and time-consumption.

BACKGROUND: DNA capture technologies combined with high-throughput sequencing now enable cost-effective, deep-coverage, targeted sequencing of complete exomes. This is well suited for SNP discovery and genotyping. However there has been little attention devoted to Copy Number Variation (CNV) detection from exome capture datasets despite the potentially high impact of CNVs in exonic regions on protein function. RESULTS: As members of the 1000 Genomes Project analysis effort, we investigated 697 samples in which 931 genes were targeted and sampled with 454 or Illumina paired-end sequencing. We developed a rigorous Bayesian method to detect CNVs in the genes, based on read depth within target regions. Despite substantial variability in read coverage across samples and targeted exons, we were able to identify 107 heterozygous deletions in the dataset. The experimentally determined false discovery rate (FDR) of the cleanest dataset from the Wellcome Trust Sanger Institute is 12.5%. We were able to substantially improve the FDR in a subset of gene deletion candidates that were adjacent to another gene deletion call (17 calls). The estimated sensitivity of our call-set was 45%. CONCLUSIONS: This study demonstrates that exonic sequencing datasets, collected both in population based and medical sequencing projects, will be a useful substrate for detecting genic CNV events, particularly deletions. Based on the number of events we found and the sensitivity of the methods in the present dataset, we estimate on average 16 genic heterozygous deletions per individual genome. Our power analysis informs ongoing and future projects about sequencing depth and uniformity of read coverage required for efficient detection.