Search

The Internet of Things is ushering in the era of connected environments where the number and diversity of data sources (devices and sensors) are inevitably increasing the size of the data that need to be stored locally (at the edge device level) and transmitted to base storages (at the sink level) of the network. This huge amount of data highlights several challenges including network bandwidth, consumption of network energy, cloud storage, and I/O throughput. These call for data pre-processing and filtering solutions to reduce the amount of data being handled and transmitted over the network. In this study, we investigate data deduplication as a prominent pre-processing method that can be used and adapted to address such challenges. Data deduplication techniques have been traditionally developed for data storage and data warehousing applications and aim at identifying and eliminating redundant data items. Few recent approaches have been designed for connected environments, yet they share various limitations, including: (i) detecting duplicates at one level only of the network (either edge or sink exclusively), (ii) overlooking the context and dynamicity of the network (disregarding device mobility and overlooking boundary separations and sensor coverage areas), (iii) relying on crisp thresholds and providing minimum-to-no expert control over the deduplication process (disregarding the domain expert's needs in defining redundancy). In this study, we propose FREDD, a new approach for Fuzzy Redundancy Elimination for Data Deduplication in a connected environment. FREDD uses simple natural language rules to represent domain knowledge and expert preferences regarding data duplication boundaries. It then applies pattern codes and fuzzy reasoning to detect duplicates at both the edge level and the sink level of the network. This reduces the time required to hard-code the deduplication process, while adapting to the domain expert's needs for different data sources and applications. Moreover, FREDD is adapted for multiple scenarios, considering both static and mobile devices, with different configurations of hard-separated and soft-separated zones, and different sensor coverage areas in the connected environment. Experiments on a real-world dataset highlight FREDD's potential and improvement compared with existing solutions. [ABSTRACT FROM AUTHOR]

Establishing a healthy lifestyle has become a very important aspect in people's lives. The latter requires maintaining a healthy nutrition by considering the nature and quantity of foods being consumed, allowing to regulate one's intake and consumption of calories and nutrients. As a result, people reach out for nutrition experts which services are costly, time-consuming, and not readily available. While various e-solutions have been developed to perform meal planning, yet most of them lack a completely automated process and require domain expert intervention at different stages of the recommendation process (e.g., identifying macronutrient distribution, providing pre-defined meal plans, or combining recommended foods into meal structures). In addition, most solutions focus on fulfilling the patients' nutrition requirements (in terms of caloric intake and macronutrients) while disregarding other relevant factors such as patient food preferences, food variety, food-meal compatibility, and inter-food compatibility. Hence, there is a need for an automated solution to produce a full-fledged meal plan from scratch, based on a recommended caloric intake and considering multiple factors. In this study, we introduce a novel solution titled MPG for automated Meal Plan Generation recommendations, designed based on an adaptation of the transportation optimization problem to simulate the "human thought process" involved in generating daily meal plans. MPG allows to: (i) generate plans which fulfill a recommended caloric intake, given a set of available foods, while (ii) personalizing the plans following patient chosen factors (e.g., food preferences, variety, and compatibility), and (iii) evaluating the relevance of the produced plans following patient preferences. We have conducted various experiments involving 9 human testers and 124 meal plans to test the performance of MPG. Results highlight MPG's effectiveness in producing "healthy" and personalized meal plans while complying with the testers' preferences. [ABSTRACT FROM AUTHOR]

Text classification is a central task in Natural Language Processing (NLP) that aims at categorizing text documents into predefined classes or categories. It requires appropriate features to describe the contents and meaning of text documents, and map them with their target categories. Existing text feature representations rely on a weighted representation of the document terms. Hence, choosing a suitable method for term weighting is of major importance and can help increase the effectiveness of the classification task. In this study, we provide a novel text classification framework for Category-based Feature Engineering titled CFE. It consists of a supervised weighting scheme defined based on a variant of the TF-ICF (Term Frequency-Inverse Category Frequency) model, embedded into three new lean classification approaches: (i) IterativeAdditive (flat), (ii) GradientDescentANN (1-layered), and (iii) FeedForwardANN (2-layered). The IterativeAdditive approach augments each document representation with a set of synthetic features inferred from TF-ICF category representations. It builds a term-category TF-ICF matrix using an iterative and additive algorithm that produces category vector representations and updates until reaching convergence. GradientDescentANN replaces the iterative additive process mentioned previously by computing the term-category matrix using a gradient descent ANN model. Training the ANN using the gradient descent algorithm allows updating the term-category matrix until reaching convergence. FeedForwardANN uses a feed-forward ANN model to transform document representations into the category vector space. The transformed document vectors are then compared with the target category vectors, and are associated with the most similar categories. We have implemented CFE including its three classification approaches, and we have conducted a large battery of tests to evaluate their performance. Experimental results on five benchmark datasets show that our lean approaches mostly improve text classification accuracy while requiring significantly less computation time compared with their deep model alternatives. • Provides a supervised weighting scheme based on Term Frequency-Inverse Category Frequency model. • Introduces three lean-architecture classifiers compared with more complex deep learning models. • Iterative-additive classifier builds term-category matrix producing category vector features. • Gradient descent classifier computes term-category matrix using gradient descent ANN model. • ANN classifier trains feed-forward ANN to transform documents to the category vector space. [ABSTRACT FROM AUTHOR]

Given that the meaning of an image is rarely self-evident using traditional keyword and/or content-based descriptions, the general goal of this study is to convert, with minimal human intervention, a stream of web vector graphics into a searchable knowledge graph structure that encodes semantically relevant image contents. To do so, we introduce an original framework titled SSG which automatically converts a stream of SVG images and objects into a semantic graph. We introduce an incremental clustering approach to semantically annotate SVG images and their constituent objects in a fast and efficient manner, using an aggregation of shape, area, color, and location similarity measures. We then produce an RDF graph representation of the input image and integrate it in a reference knowledge graph, incrementally extending its semantic expressiveness to improve future annotation tasks. This achieves semantization of vector image contents with minimum human effort and training data, while complying with native Web standards (i.e., SVG and RDF) to preserve transparency in representing and searching images using Semantic Web stack technologies. Our solution is of linear complexity in the number of images and clusters used. We have conducted a large battery of experiments to test and evaluate our approach. We have created a labelled SVG dataset consisting of 22,553 objects from 750 images based on panoramic dental X-ray images. To our knowledge, it is the first significant dataset of labelled SVG objects and images, which we make available online as a benchmark for future research in this area. Results underline our approach's effectiveness, and its applicability in a practical application domain. [ABSTRACT FROM AUTHOR]

Low-light image (LLI) enhancement is an important image processing task that aims at improving the illumination of images taken under low-light conditions. Recently, a remarkable progress has been made in utilizing deep learning (DL) approaches for LLI enhancement. This paper provides a concise and comprehensive review and comparative study of the most recent DL models used for LLI enhancement. To our knowledge, this is the first comparative study dedicated to DL-based models for LLI enhancement. We address LLI enhancement in two ways: (i) standalone, as a separate task, and (ii) end-to-end, as a pre-processing stage embedded within another high-level computer vision task, namely object detection and classification. The paper consists of six logical parts. First, we provide an overview of the background and literature in LLI enhancement. Second, we describe the test data and experimental setup of the study. Third, we present a quantitative and qualitative comparison of the visual and perceptual quality achieved by 10 of the most recent DL-based LLI enhancement models. Fourth, we present a comparative analysis for object detection and classification performance achieved by 4 different object detection models applied on LLIs and their enhanced counterparts. Fifth, we perform a feature analysis of DL feature maps extracted from normal, low-light, and enhanced images, and perform the occlusion experiment to better understand the effect of LLI enhancement on the object detection and classification task. Finally, we provide our conclusions and highlight future steps and potential directions. • Evaluates Deep Learning (DL) models for Low-light Image (LLI) enhancement. • Compares 10 LLI enhancement models and 4 object detection and classification models. • Provides a quantitative and qualitative comparison of visual and perceptual quality. • Evaluates impact of LLI enhancement on object detecting and classification quality. • Performs occlusion experiment to study LLI enhancement's effect on object detection. [ABSTRACT FROM AUTHOR]

Low-light image (LLI) enhancement techniques have recently demonstrated remarkable progress especially with the use of deep learning approaches. However, most existing techniques are developed as standalone solutions and do not take into account the impact of LLI enhancement on high-level computer vision tasks like object classification. In this paper, we propose a new LLI enhancement model titled LLHFNet (Low-light Homomorphic Filtering Network) which performs image-to-frequency filter learning and is designed for seamless integration into classification models. Through this integration, the classification model is embedded with an internal enhancement capability and is jointly trained to optimize both image enhancement and classification performance. We have conducted a large battery of experiments using SICE, Pascal VOC, and ExDark datasets, to quantitatively and qualitatively evaluate our approach's enhancement quality and classification performance. When evaluated as a standalone enhancement model, our solution consistently ranks among the best existing image enhancement techniques. When embedded with a classification model, our solution achieves an average 5.5% improvement in classification accuracy, compared with the traditional pipeline of separate enhancement followed by classification. Results produce robust classification quality on both LLIs and normal-light images (NLIs), and highlight a clear improvement to the literature. • Introduces Low-Light Image (LLI) Enhancement model for Object Classification task. • Introduces LLHFNet deep learning model for image-to-frequency filter learning. • Introduces LLI enhancer–classifier to integrate LLHFNet in object classification. • Performs joint training to optimize both enhancement and classification models. • Evaluates impact of LLI enhancement on object detecting and classification quality. • Produces 5.5% improvement in classification accuracy on Pascal VOC & ExDark LLIs. [ABSTRACT FROM AUTHOR]

• Automates health assessment, recommendation, and monitoring services offered by expert • Uses fuzzy logic to simulate the "human common sense" thought process in nutrition assessment • Performs patient health and weight assessment and recommends target BFP and weight goal • Performs caloric intake and physical exercise recommendation based on target BFP and weight goal • Performs progress evaluation of patient health state and adjusts recommendations accordingly Establishing a healthy lifestyle has become a very important aspect in people's lives. The latter requires maintaining a healthy nutrition by considering the type and quantity of consumed foods. It also requires maintaining an active lifestyle including the necessary amount of physical exercise to regulate one's intake and consumption of calories and nutrients. As a result, people reach out for nutrition experts to perform health assessment, whose services are costly, time consuming, and not readily available. While various e-nutrition solutions have been developed, yet most of them perform meal planning without performing health state assessment or evaluation (traditionally provided by human experts). To our knowledge, there is no existing automated solution to perform nutrition health assessment, recommendation, and progress evaluation, which are central pre-requites to the meal planning task. In this study, we introduce a novel framework titled PIN for Personalized Intelligent Nutrition recommendations. PIN relies on the fuzzy logic paradigm to simulate human expert health assessment capabilities, including weight, caloric intake, and exercise recommendations as well as progress evaluation and recommendation adjustments. It includes three essential and complementary modules: i) Weight Assessment and Recommendation (WAR), ii) Caloric Intake and Exercise Recommendation (CIER), and iii) Progress Evaluation and Recommendation Adjustment (PERA). This underlines the first computerized solution for nutrition health assessment. We have conducted a large battery of experiments involving 50 patient profiles and 11 nutrition expert evaluators to test the performance of PIN and evaluate its health assessment quality. Results show that PIN 's assessment and recommendations are on a par with and sometimes surpass those of human nutritionists. [ABSTRACT FROM AUTHOR]

Major advances in the fields of Internet and Communication Technology (ICT), data modeling/processing, and sensing technology have rendered traditional environments (e.g., cities, buildings) more connected. Although the sensed data could be useful for various applications (e.g., event detection in cities, energy management in commercial buildings), it first requires pre-processing to clean various inconsistencies (e.g., anomalies, redundancies, missing values). In this work, we focus on managing data redundancies in connected environments. Existing approaches suffer from (i) disregarding edge data redundancies either at the edge or at the core of the network; (ii) disregarding sensor mobility and the dynamicity of the network; (iii) disregarding the limited resources of edge devices; (iv) disregarding network/infrastructure resources; and (v) disregarding data consumer needs/requirements when cleaning the data redundancies. To address these limitations, we propose here DRMF: Data Redundancy Management for leaF-edges allowing to identify and remove data redundancies in connected environments at the device level. DRMF considers both static and mobile edge devices, and provides two algorithms for temporal and spatio-temporal redundancy detection. Once redundancies are identified, DRMF performs data deduplication taking into account the dynamic requirements of data consumers and device resources (e.g., processing, battery, memory). Experimental results highlight the performance and accuracy of our solution in detecting and eliminating edge data redundancies. [ABSTRACT FROM AUTHOR]

During the past two decades, 3D simulation models have gained importance in the development of software solutions that aim to mimic real-world events and phenomena with increasing levels of accuracy and detail. In this context, knowledge representation and processing have recently shown a significant contribution to the simulation modeling domain, where knowledge graphs have been used in different fields to build knowledge representations for multiple purposes. In this paper, we introduce VOWES, a Virtual Outdoor Weather Event Simulator to replicate and measure outdoor weather events in vivid 3D visualizations. We design and implement an integrated knowledge graph (KG) representation for VOWES, by creating two constituent KGs: (i) Weather KG describing weather data and events, and (ii) Simulator KG describing 3D simulation components and properties, and connecting them with the (iii) Semantic Sensor Network (SSN) KG to form an integrated structure serving as the knowledge backbone of the VOWES simulation environment. We make use of the Unity 3D engine to build and design the simulator environment and its virtual sensors, and integrate the Mapbox SDK and the WeatherStack API for realistic real-world weather mapping. We have conducted qualitative evaluations involving 13 expert and 30 non-expert testers, to assess the quality of VOWES' KGs and its simulation environment. Results show that more than 80% of the testers gave a combined quality score ≥ 3 out of 4 on most evaluation criteria. We have also conducted performance evaluations to test VOWES loading, execution, and data search time, among other features. Results show that most operations require almost instantaneous or linear time, where search, refresh, and export operations share almost identical performance levels, with execution time increasing by approximately 179 μs for every added game object. This highlights the simulation tool's time performance in running large simulation projects, and its ability to simulate complex weather environments with large numbers of sensors and weather phenomena. [ABSTRACT FROM AUTHOR]