Photoplethysmography (PPG) is a key component in a myriad of continuous and non-invasive health monitoring solutions, increasingly widespread in wearable devices, such as smartwatches and smart rings. Its high susceptibility to noise, such as motion artifacts and ambient light interference, however, can significantly hinder the learning process, as well as the resulting performance, of the deployed models. Given that, a Signal Quality Assessment (SQA) auxiliary module is usually employed in such applications, for upfront selection of the PPG segments that should be used for reliable extraction of physiological information from the user. Most SQA strategies adopted in these devices rely either on (1) Deep Learning (DL) models, capable of obtaining high performance metrics despite being of increased complexity and energy consumption, or (2) Decision Rule-based strategies that can assess signal quality in an increasedly energy-efficient way, albeit with reduced robustness. In this work, we introduce Hexa-SymmLTP-CC, a novel signal quality classifier composed of an ensemble of Local Pattern-based feature extractors followed by a downstream linear binary classifier, which outperforms state-of-the-art solutions, achieving accuracies of 93.93%, 96.06% and 96.55% across three clinical expert-annotated smartwatch and smart ring PPG datasets, while respecting the lightweight restrictions for wearable-based real-time monitoring applications.

Poor dietary habits are frequently associated with diverse health conditions. Although there are apps designed to monitor eating habits, they require manual annotation, usually leading to user disengagement. This paper presents a data-driven solution using smartwatch sensors to automate meal session detection, addressing limitations of wearables such as a laterality unified model, capturing meal sessions with different durations, model compression to enable embedding, and dynamic inference times to reduce energy consumption. We tested two pipeline variations on a dataset of 91 subjects in a free-living scenario for 12 days, resulting in 93 % Precision and 82 % Recall, indicating the feasibility of free-living meal session detection. We also developed an application running on Samsung smartwatches to assess the solution feasibility with limited resources. This solution can be used as the building block to track eating-related phenomena such as autonomous dietary guidance and real-time non-invasive glucose monitoring, which remain unsolved problems in literature.

Getting prompt insights about health and well-being in a non-invasive way is one of the most popular features available on wearable devices. Among all vital signs available, heart rate (HR) monitoring is one of the most important since other measurements are based on it. Real-time HR estimation in wearables mostly relies on photoplethysmography (PPG), which is a fair technique to handle such a task. However, PPG is vulnerable to motion artifacts (MA). As a consequence, the HR estimated from PPG signals is strongly affected during physical exercises. Different approaches have been proposed to deal with this problem, however, they struggle to handle exercises with strong movements, such as a running session. In this paper, we present a new method for HR estimation in wearables that uses an accelerometer signal and user demographics to support the HR prediction when the PPG signal is affected by motion artifacts. This algorithm requires a tiny memory allocation and allows on-device personalization since the model parameters are finetuned in real time during workout executions. Also, the model may predict HR for a few minutes without using a PPG, which represents a useful contribution to an HR estimation pipeline. We evaluate our model on five different exercise datasets - performed on treadmills and in outdoor environments - and the results show that our method can improve the coverage of a PPG-based HR estimator while keeping a similar error performance, which is particularly useful to improve user experience.

Over the past few years, wearable devices have become quite popular, in particular, smartwatches. One reason for this popularity is the possibility to monitor health and well-being in a non-invasive way. Heart Rate (HR) monitoring is one of the most important health features available in wearables. Normally, HR estimation is achieved using photoplethysmography (PPG), a common low-cost optical technique that achieves fair HR estimation in wearables. However, this technique is energy-consuming and significantly affects the device’s battery life for long-term monitoring – such as during physical exercises. In this work, we proposed a model based on linear regression and a Proportional–Integral–Derivative (PID) controller that uses an accelerometer and user’s demographics to estimate HR. The main goal of this model is to reduce power consumption since the accelerometer is a low-power sensor. We perform experiments to evaluate the performance of our method using three datasets containing more than 180 hours of data composed of a large number of different subjects. The results show that our method is competitive with a PPG-based approach and for some occasions, it is plausible to use such a model in order to save battery.

Real-time health monitoring is revolutionizing healthcare delivery nowadays. Using everyday settings, especially due to the recent wearable health devices, it is possible to monitor individuals at any place and moment, allowing the detection and prevention of many diseases. Among the various technologies present in wearable devices that allow continuous health monitoring, Photoplethysmography (PPG) is one of the most important techniques. PPG is non-invasive, low-cost, easy-to-implement, and, therefore, convenient to track physiological signals, such as oxygen saturation in the bloodstream, heart rate variability, respiration rate, etc. Due to these advantages, PPG is widely used in diverse health applications, notably in commercial wearable apparatuses. However, despite its advantages, PPG presents a main drawback of being highly susceptible to motion artifacts and environmental noises, which impair PPG-based applications, especially when PPG signals are recorded via wearable devices. Therefore, to enable reliable measurements, signal quality must be assessed, and unreliable signals should be rejected. With such signal reliability needs, the most important thing is Signal Quality Assessment (SQA). In this paper, we introduce a novel SQA method that projects the 1D PPG signals into 2D images and then uses a Vision Transformer (ViT) to classify their quality. Results show that the proposed method presents a competitive quality prediction accuracy when compared with the state-of-the-art.

In the last few years, several works have tackled the problem of novel view synthesis from stereo images or even from a single picture. However, previous methods are computationally expensive, specially for high-resolution images. In this paper, we address the problem of generating a multiplane image (MPI) from a single high-resolution picture. We present the adaptive-MPI representation, which allows rendering novel views with low computational requirements. To this end, we propose an adaptive slicing algorithm that produces an MPI with a variable number of image planes. We present a new lightweight CNN for depth estimation, which is learned by knowledge distillation from a larger network. Occluded regions in the adaptive-MPI are inpainted also by a lightweight CNN. We show that our method is capable of producing high-quality predictions with one order of magnitude less parameters compared to previous approaches. The robustness of our method is evidenced on challenging pictures from the Internet.

Flooding is the world's most costly type of natural disaster in terms of both economic losses and human causalities. A first and essential procedure toward flood monitoring is based on identifying the area most vulnerable to flooding, which gives authorities relevant regions to focus. In this letter, we propose several methods to perform flooding identification in high-resolution remote sensing images using deep learning. Specifically, some proposed techniques are based upon unique networks, such as dilated and deconvolutional ones, whereas others were conceived to exploit diversity of distinct networks in order to extract the maximum performance of each classifier. The evaluation of the proposed methods was conducted in a high-resolution remote sensing data set. Results show that the proposed algorithms outperformed the state-of-the-art baselines, providing improvements ranging from 1% to 4% in terms of the Jaccard Index.

We develop an approach for the detection of ruins of livestock enclosures (LEs) in alpine areas captured by high-resolution remotely sensed images. These structures are usually of approximately rectangular shape and appear in images as faint fragmented contours in complex background. We address this problem by introducing a rectangularity feature that quantifies the degree of alignment of an optimal subset of extracted linear segments with a contour of rectangular shape. The rectangularity feature has high values not only for perfectly regular enclosures but also for ruined ones with distorted angles, fragmented walls, or even a completely missing wall. Furthermore, it has a zero value for spurious structures with less than three sides of a perceivable rectangle. We show how the detection performance can be improved by learning a linear combination of the rectangularity and size features from just a few available representative examples and a large number of negatives. Our approach allowed detection of enclosures in the Silvretta Alps that were previously unknown. A comparative performance analysis is provided. Among other features, our comparison includes the state-of-the-art features that were generated by pretrained deep convolutional neural networks (CNNs). The deep CNN features, although learned from a very different type of images, provided the basic ability to capture the visual concept of the LEs. However, our handcrafted rectangularity-size features showed considerably higher performance.

BACKGROUND: Pixel-level tissue classification for ultrasound images, commonly applied to carotid images, is usually based on defining thresholds for the isolated pixel values. Ranges of pixel values are defined for the classification of each tissue. The classification of pixels is then used to determine the carotid plaque composition and, consequently, to determine the risk of diseases (e.g., strokes) and whether or not a surgery is necessary. The use of threshold-based methods dates from the early 2000s but it is still widely used for virtual histology. METHODOLOGY/PRINCIPAL FINDINGS: We propose the use of descriptors that take into account information about a neighborhood of a pixel when classifying it. We evaluated experimentally different descriptors (statistical moments, texture-based, gradient-based, local binary patterns, etc.) on a dataset of five types of tissues: blood, lipids, muscle, fibrous, and calcium. The pipeline of the proposed classification method is based on image normalization, multiscale feature extraction, including the proposal of a new descriptor, and machine learning classification. We have also analyzed the correlation between the proposed pixel classification method in the ultrasound images and the real histology with the aid of medical specialists. CONCLUSIONS/SIGNIFICANCE: The classification accuracy obtained by the proposed method with the novel descriptor in the ultrasound tissue images (around 73%) is significantly above the accuracy of the state-of-the-art threshold-based methods (around 54%). The results are validated by statistical tests. The correlation between the virtual and real histology confirms the quality of the proposed approach showing it is a robust ally for the virtual histology in ultrasound images.

This paper presents a novel unsupervised learning approach that takes into account the intrinsic dataset structure, which is represented in terms of the reciprocal neighborhood references found in different ranked lists. The proposed Reciprocal kNN Distance defines a more effective distance between two images, and is used to improve the effectiveness of image retrieval systems. Several experiments were conducted for different image retrieval tasks involving shape, color, and texture descriptors. The proposed approach is also evaluated on multimodal retrieval tasks, considering visual and textual descriptors. Experimental results demonstrate the effectiveness of proposed approach. The Reciprocal kNN Distance yields better results in terms of effectiveness than various state-of-the-art algorithms.

Many methods have been recently proposed to deal with the large amount of data provided by high-resolution remote sensing technologies. Several of these methods rely on the use of image segmentation algorithms for delineating target objects. However, a common issue in geographic object-based applications is the definition of the appropriate data representation scale, a problem that can be addressed by exploiting multiscale segmentation. The use of multiple scales, however, raises new challenges related to the definition of effective and efficient mechanisms for extracting features. In this paper, we address the problem of extracting histogram-based features from a hierarchy of regions for multiscale classification. The strategy, called H-Propagation, exploits the existing relationships among regions in a hierarchy to iteratively propagate features along multiple scales. The proposed method speeds up the feature extraction process and yields good results when compared with global low-level extraction approaches.

This work is developed in the context of the placing task of the MediaEval 2011 initiative. The objective is to geocode (or geotag) a set of videos, i.e., automatically assign geographical coordinates to them. This paper presents an architecture for multimodal geocoding that exploits both visual and textual descriptions associated with videos. This work also describes our efforts regarding the implementation of this architecture to demonstrate its applicability. Conducted experiments show how our multimodal approach enhances the results compared to relying on a single modality.

This paper presents a novel approach for video representation, called bag-of-scenes. The proposed method is based on dictionaries of scenes, which provide a high-level representation for videos. Scenes are elements with much more semantic information than local features, specially for geotagging videos using visual content. Thus, each component of the representation model has self-contained semantics and, hence, it can be directly related to a specific place of interest. Experiments were conducted in the context of the MediaEval 2011 Placing Task. The reported results show our strategy compared to those from other participants that used only visual content to accomplish this task. Despite our very simple way to generate the visual dictionary, which has taken photos at random, the results show that our approach presents high accuracy relative to the state-of-the art solutions.