Signals, Volume 5, Issue 4 (December 2024) – 13 articles

This study proposes a comprehensive framework for the segmentation and identification of vertebrae in CT scans using a combination of deep learning and traditional machine learning techniques. The Res U-Net architecture is employed to achieve a high model accuracy of 93.62% on the VerSe’20 dataset demonstrating effective performance in segmenting lumbar and thoracic vertebrae. Feature extraction is enhanced through the application of Otsu’s method which effectively distinguishes the vertebrae from the surrounding tissue. The proposed method achieves a Dice Similarity Coefficient (DSC) of 87.10% ± 3.72%, showcasing its competitive performance against other segmentation techniques. By accurately extracting vertebral features this framework assists medical professionals in precise preoperative planning, allowing for the identification and marking of critical anatomical features required during spinal fusion procedures. This integrated approach not only addresses the challenges of vertebrae segmentation but also offers a scalable and efficient solution for analyzing large-scale medical imaging datasets with the potential to significantly improve clinical workflows and patient outcomes. Full article

Many natural signals exhibit quasi-periodic behaviors and are conveniently modeled as combinations of several harmonic sinusoids whose relative frequencies, magnitudes, and phases vary with time. The waveform shapes of those signals reflect important physical phenomena underlying their generation, requiring those parameters to be accurately estimated and modeled. In the literature, accurate phase estimation and modeling have received significantly less attention than frequency or magnitude estimation. This paper first addresses accurate DFT-based phase estimation of individual sinusoids across six scenarios involving two DFT-based filter banks and three different windows. It has been shown that bias in phase estimation is less than 0.001 radians when the SNR is equal to or larger than 2.5 dB. Using the Cramér–Rao lower bound as a reference, it has been demonstrated that one particular window offers performance of practical interest by better approximating the CRLB under favorable signal conditions and minimizing performance deviation under adverse conditions. This paper describes the development of a shift-invariant phase-related feature that characterizes the harmonic phase structure. This feature motivates a new signal processing paradigm that greatly simplifies the parametric modeling, transformation, and synthesis of harmonic signals. It also aids in understanding and reverse engineering the phasegram. The theory and results are discussed from a reproducible perspective, with dedicated experiments supported by code, allowing for the replication of figures and results presented in this paper and facilitating further research. Full article

Accurate knowledge of the flow rate is essential for hydraulic systems, enabling the calculation of hydraulic power when combined with pressure measurements. These data are crucial for applications such as predictive maintenance. However, most flow rate sensors in fluid power systems operate invasively, disrupting the flow and producing inaccurate results, especially under transient conditions. Utilizing pressure transducers represents a non-invasive soft sensor approach since no physical flow rate sensor is used to determine the flow rate. Usually, this approach relies on the Hagen–Poiseuille (HP) law, which is limited to steady and incompressible flow. This paper introduces a novel soft sensor with an analytical model for transient, compressible pipe flow based on two pressure signals. The model is derived by solving fundamental fluid equations in the Laplace domain and converting them back to the time domain. Using the four-pole theorem, this model contains a relationship between the pressure difference and the flow rate. Several unsteady test cases are investigated and compared to a steady soft sensor based on the HP law, highlighting our soft sensor’s promising capability. It exhibits an overall error of less than $0.15$% for the investigated test cases in a distributed-parameter simulation, whereas the HP-based sensor shows errors in the double-digit range. Full article

Practical applications of compressed sensing often restrict the choice of its two main ingredients. They may (i) prescribe the use of particular redundant dictionaries for certain classes of signals to become sparsely represented or (ii) dictate specific measurement mechanisms which exploit certain physical principles. On the problem of RIP measurement matrix design in compressed sensing with redundant dictionaries, we give a simple construction to derive sensing matrices whose compositions with a prescribed dictionary have with high probability the RIP in the $klog(n/k)$ regime. Our construction thus provides recovery guarantees usually only attainable for sensing matrices from random ensembles with sparsifying orthonormal bases. Moreover, we use the dictionary factorization idea that our construction rests on in the application of magnetic resonance imaging, in which also the sensing matrix is prescribed by quantum mechanical principles. We propose a recovery algorithm based on transforming the acquired measurements such that the compressed sensing theory for RIP embeddings can be utilized to recover wavelet coefficients of the target image, and show its performance on examples from the fastMRI dataset. Full article

In this paper, we demonstrate the realistic test of a 2.4 GHz multi-hop wireless network for mountainous forest and watercourse environments. A multi-hop network using IEEE 802.15.4 XBee3 micro-modules and a communication protocol among nodes were developed. A wireless node deployment solution was introduced for practical testing. The proposed system’s communication reliability was tested in two different scenarios: a mountainous forest with sloping areas and trees and a watercourse, which referred to environmental and flooding monitoring applications. Wireless network performances were evaluated through the received signal strength indicator (RSSI) level of each wireless link, a packet delivery ratio (PDR), as the successful rate of packet transmission, and the end-to-end delay (ETED) of all data packets from the transmitter to the receiver. The experimental results demonstrate the success of the multi-hop WSN deployment and communication in both scenarios, where the RSSI of each link was kept at the accepted level and the PDR achieved the highest result. Furthermore, as a real-time response, the data from the source could be sent to the sink with a small ETED. Full article

Our long journey on the road of telecommunications is continuously evolving. We have experienced several technological changes, modernizations, optimizations, and various mergers in the past decades. Virtualization and ‘cloudification’ of legacy telecommunication equipment has made communication networks not only more flexible, but also opened new doors. Brand new types of services have become available thanks to the ongoing fusion of the two domains of telecommunications and IT (Information Technology). This overview paper first discusses the evolution of services with an enhanced focus on mobile networks. Then, the possibilities offered by IT are shown. Finally, some examples are given of how Communication Service Providers and end users can benefit from these recent changes. Full article

Change detection in biomedical signals is crucial for understanding physiological processes and diagnosing medical conditions. This study evaluates various change detection methods, focusing on synthetic signals that mimic real-world scenarios. We examine the following three methods: classical statistical techniques (thresholding based on mean and standard deviation), Support Vector Machine (SVM) classification, and time–frequency analysis using Continuous Wavelet Transform (CWT). Each method’s performance is assessed using synthetic signals, including nonlinear signals and those with simulated anomalies. We calculated the F1-score to quantify performance, providing a balanced measure of precision and recall. Results showed that SVM classification outperformed both classical techniques and CWT analysis, achieving a higher F1-score in detecting changes. While all methods struggled with synthetic nonlinear signals, classical techniques and SVM successfully detected changes in signals with simulated anomalies, whereas CWT had difficulty with both types of signals. These findings underscore the importance of selecting appropriate change detection methods based on signal characteristics. Future research should explore advanced machine learning and signal processing techniques to improve detection accuracy in biomedical applications. Full article

Neural networks have inductive biases owing to the assumptions associated with the selected learning algorithm, datasets, and network structure. Specifically, convolutional neural networks (CNNs) are known for their tendency to exhibit textural biases. This bias is closely related to image classification accuracy. Aligning the model’s bias with the dataset’s bias can significantly enhance performance in transfer learning, leading to more efficient learning. This study aims to quantitatively demonstrate that increasing shape bias within the network by varying kernel sizes and dilation rates improves accuracy on shape-dominant data and enables efficient learning with less data. Furthermore, we propose a novel method for quantitatively evaluating the balance between texture bias and shape bias. This method enables efficient learning by aligning the biases of the transfer learning dataset with those of the model. Systematically adjusting these biases allows CNNs to better fit data with specific biases. Compared to the original model, an accuracy improvement of up to 9.9% was observed. Our findings underscore the critical role of bias adjustment in CNN design, contributing to developing more efficient and effective image classification models. Full article

This paper presents a methodology for rectifying curved text lines, a crucial process in optical character recognition (OCR) and computer vision. Utilizing generalized additive models (GAMs), the proposed method accurately estimates text curvature and rectifies it into a straight format for improved text recognition. The process includes image binarization techniques like Otsu’s thresholding, morphological operations, curve estimation, and the Bresenham line drawing algorithm. The results show significant improvements in OCR accuracy among different challenging distortion scenarios. The implementation, written in Python, demonstrates the potential for enhancing text alignment and rectification in scanned text line images utilizing a flexible, robust, and customizable framework. Full article

Polysomnography (PSG) measures brain activity during sleep via electroencephalography (EEG) using six leads. Artifacts caused by movement or loose leads distort EEG measurements. We developed a method to automatically identify such artifacts in a PSG EEG trace. After preprocessing, we extracted power levels at frequencies of 0.5–32.5 Hz with multitaper spectral analysis using 4 s windows with 3 s overlap. For each resulting 1 s segment, we computed segment-specific correlations between power levels for all pairs of leads. We then averaged all pairwise correlation coefficients involving each lead, creating a time series of segment-specific average correlations for each lead. Our algorithm scans each averaged time series separately for “bad” segments using a local moving window. In a second pass, any segment whose averaged correlation is less than a global threshold among all remaining good segments is declared an outlier. We mark all segments between two outlier segments fewer than 300 s apart as artifact regions. This process is repeated, removing a channel with excessive outliers in each iteration. We compared artifact regions discovered by our algorithm to expert-assessed ground truth, achieving sensitivity and specificity of 80% and 91%, respectively. Our algorithm is an open-source tool, either as a Python package or a Docker. Full article

Reliable streamflow forecasting is crucial for several tasks related to water-resource management, including planning reservoir operations, power generation via Hydroelectric Power Plants (HPPs), and flood mitigation, thus resulting in relevant social implications. The present study is focused on the application of Automated Machine-Learning (AutoML) models to forecast daily streamflow in the area of the upper Teles Pires River basin, located in the region of the Amazon biomes. The latter area is characterized by extensive water-resource utilization, mostly for power generation through HPPs, and it has a limited hydrological data-monitoring network. Five different AutoML models were employed to forecast the streamflow daily, i.e., auto-sklearn, Tree-based Pipeline Optimization Tool (TPOT), H2O AutoML, AutoKeras, and MLBox. The AutoML input features were set as the time-lagged streamflow and average rainfall data sourced from four rain gauge stations and one streamflow gauge station. To overcome the lack of training data, in addition to the previous features, products estimated via remote sensing were leveraged as training data, including PERSIANN, PERSIANN-CCS, PERSIANN-CDR, and PDIR-Now. The selected AutoML models proved their effectiveness in forecasting the streamflow in the considered basin. In particular, the reliability of streamflow predictions was high both in the case when training data came from rain and streamflow gauge stations and when training data were collected by the four previously mentioned estimated remote-sensing products. Moreover, the selected AutoML models showed promising results in forecasting the streamflow up to a three-day horizon, relying on the two available kinds of input features. As a final result, the present research underscores the potential of employing AutoML models for reliable streamflow forecasting, which can significantly advance water-resource planning and management within the studied geographical area. Full article

Switch points are a specific kind of cue point that DJs carefully look for when mixing music tracks. As the name says, a switch point is the point in time where the current track in a DJ mix is replaced by the upcoming track. Being able to identify these positions is a first step toward the interpretation and the emulation of DJ mixes. With the aim of automatically detecting switch points, we evaluate one experience-driven and several statistics-driven methods. By comparing the decision process of each method, contrasted by their performance, we deduce the characteristics linked to switch points. Specifically, we identify the most impactful features for their detection, namely, the novelty in the signal energy, the timbre, the number of drum onsets, and the harmony. Furthermore, we expose multiple interactions among these features. Full article

There is a growing need to facilitate the production of recorded music as independent musicians are now key in preserving the broader cultural roles of music. A critical component of the production of music is multitrack mixing, a time-consuming task aimed at, among other things, reducing spectral masking and enhancing clarity. Traditionally, this is achieved by skilled mixing engineers relying on their judgment. In this work, we present an adaptive filtering method based on a novel masking detection scheme capable of identifying masking contributions, including temporal interchangeability between the masker and maskee. This information is then systematically used to design and apply filters. We implement our methods on multitrack music to improve the quality of the raw mix. Full article