Reshaping Bioacoustics Event Detection: Leveraging Few-Shot Learning (FSL) with Transductive Inference and Data Augmentation.

Bioacoustic event detection is a demanding endeavor involving recognizing and classifying the sounds animals make in their natural habitats. Traditional supervised learning requires a large amount of labeled data, which are hard to come by in bioacoustics. This paper presents a few-shot learning (FSL) method incorporating transductive inference and data augmentation to address the issues of too few labeled events and small volumes of recordings. Here, transductive inference iteratively alters class prototypes and feature extractors to seize essential patterns, whereas data augmentation applies SpecAugment on Mel spectrogram features to augment training data. The proposed approach is evaluated by using the Detecting and Classifying Acoustic Scenes and Events (DCASE) 2022 and 2021 datasets. Extensive experimental results demonstrate that all components of the proposed method achieve significant F-score improvements of 27% and 10%, for the DCASE-2022 and DCASE-2021 datasets, respectively, compared to recent advanced approaches. Moreover, our method is helpful in FSL tasks because it effectively adapts to sounds from various animal species, recordings, and durations.

Auscultation-Based Pulmonary Disease Detection through Parallel Transformation and Deep Learning.

Respiratory diseases are among the leading causes of death, with many individuals in a population frequently affected by various types of pulmonary disorders. Early diagnosis and patient monitoring (traditionally involving lung auscultation) are essential for the effective management of respiratory diseases. However, the interpretation of lung sounds is a subjective and labor-intensive process that demands considerable medical expertise, and there is a good chance of misclassification. To address this problem, we propose a hybrid deep learning technique that incorporates signal processing techniques. Parallel transformation is applied to adventitious respiratory sounds, transforming lung sound signals into two distinct time-frequency scalograms: the continuous wavelet transform and the mel spectrogram. Furthermore, parallel convolutional autoencoders are employed to extract features from scalograms, and the resulting latent space features are fused into a hybrid feature pool. Finally, leveraging a long short-term memory model, a feature from the latent space is used as input for classifying various types of respiratory diseases. Our work is evaluated using the ICBHI-2017 lung sound dataset. The experimental findings indicate that our proposed method achieves promising predictive performance, with average values for accuracy, sensitivity, specificity, and F1-score of 94.16%, 89.56%, 99.10%, and 89.56%, respectively, for eight-class respiratory diseases; 79.61%, 78.55%, 92.49%, and 78.67%, respectively, for four-class diseases; and 85.61%, 83.44%, 83.44%, and 84.21%, respectively, for binary-class (normal vs. abnormal) lung sounds.

Robust Epileptic Seizure Detection Using Long Short-Term Memory and Feature Fusion of Compressed Time-Frequency EEG Images.

Epilepsy is a prevalent neurological disorder with considerable risks, including physical impairment and irreversible brain damage from seizures. Given these challenges, the urgency for prompt and accurate seizure detection cannot be overstated. Traditionally, experts have relied on manual EEG signal analyses for seizure detection, which is labor-intensive and prone to human error. Recognizing this limitation, the rise in deep learning methods has been heralded as a promising avenue, offering more refined diagnostic precision. On the other hand, the prevailing challenge in many models is their constrained emphasis on specific domains, potentially diminishing their robustness and precision in complex real-world environments. This paper presents a novel model that seamlessly integrates the salient features from the time-frequency domain along with pivotal statistical attributes derived from EEG signals. This fusion process involves the integration of essential statistics, including the mean, median, and variance, combined with the rich data from compressed time-frequency (CWT) images processed using autoencoders. This multidimensional feature set provides a robust foundation for subsequent analytic steps. A long short-term memory (LSTM) network, meticulously optimized for the renowned Bonn Epilepsy dataset, was used to enhance the capability of the proposed model. Preliminary evaluations underscore the prowess of the proposed model: a remarkable 100% accuracy in most of the binary classifications, exceeding 95% accuracy in three-class and four-class challenges, and a commendable rate, exceeding 93.5% for the five-class classification.

Mixed-Input Deep Learning Approach to Sleep/Wake State Classification by Using EEG Signals.

Sleep stage classification plays a pivotal role in predicting and diagnosing numerous health issues from human sleep data. Manual sleep staging requires human expertise, which is occasionally prone to error and variation. In recent times, availability of polysomnography data has aided progress in automatic sleep-stage classification. In this paper, a hybrid deep learning model is proposed for classifying sleep and wake states based on a single-channel electroencephalogram (EEG) signal. The model combines an artificial neural network (ANN) and a convolutional neural network (CNN) trained using mixed-input features. The ANN makes use of statistical features calculated from EEG epochs, and the CNN operates on Hilbert spectrum images generated during each epoch. The proposed method is assessed using single-channel Pz-Oz EEG signals from the Sleep-EDF database Expanded. The classification performance on four randomly selected individuals shows that the proposed model can achieve accuracy of around 96% in classifying between sleep and wake states from EEG recordings.

Optimizing Urban Air Pollution Detection Systems.

Air pollution has become a serious problem in all megacities. It is necessary to continuously monitor the state of the atmosphere, but pollution data received using fixed stations are not sufficient for an accurate assessment of the aerosol pollution level of the air. Mobility in measuring devices can significantly increase the spatiotemporal resolution of the received data. Unfortunately, the quality of readings from mobile, low-cost sensors is significantly inferior to stationary sensors. This makes it necessary to evaluate the various characteristics of monitoring systems depending on the properties of the mobile sensors used. This paper presents an approach in which the time of pollution detection is considered a random variable. To the best of our knowledge, we are the first to deduce the cumulative distribution function of the pollution detection time depending on the features of the monitoring system. The obtained distribution function makes it possible to optimize some characteristics of air pollution detection systems in a smart city.

Sensor Fault Diagnosis Using a Machine Fuzzy Lyapunov-Based Computed Ratio Algorithm.

Anomaly identification for internal combustion engine (ICE) sensors has become an important research area in recent years. In this work, a proposed indirect fuzzy Lyapunov-based computed ratio observer integrated with a support vector machine (SVM) was designed for sensor fault classification. The proposed fuzzy Lyapunov-based computed ratio observer integrated with SVM has three main layers. In the preprocessing (first) layer, the resampled root mean square (RMS) signals are extracted from the original signals to the designed indirect observer. The second (observation) layer is the principal part with the proposed indirect fuzzy sensor-fault-classification technique. This layer has two sub-layers: signal modeling and estimation. The Gaussian autoregressive-Laguerre approach integrated with the fuzzy approach is designed for resampled RMS fuel-to-air-ratio normal signal modeling, while the subsequent sub-layer is used for resampled RMS fuel-to-air-ratio signal estimation using the proposed fuzzy Lyapunov-based computed ratio observer. The third layer, for residual signal generation and classification, is used to identify ICE sensor anomalies, where residual signals are generated by the difference between the original and estimated resampled RMS fuel-to-air-ratio signals. Moreover, SVM is suggested for residual signal classification. To test the effectiveness of the proposed method, the results are compared with two approaches: a Lyapunov-based computed ratio observer and a computed ratio observer. The results show that the accuracy of sensor anomaly classification by the proposed fuzzy Lyapunov-based computed ratio observer is 98.17%. Furthermore, the proposed scheme improves the accuracy of sensor fault classification by 8.37%, 2.17%, 6.17%, 4.57%, and 5.37% compared to other existing methods such as the computed ratio observer, the Lyapunov-based computed ratio observer, fuzzy feedback linearization observation, self-tuning fuzzy robust multi-integral observer, and Kalman filter technique, respectively.

CAFD: Context-Aware Fault Diagnostic Scheme towards Sensor Faults Utilizing Machine Learning.

Sensors' existence as a key component of Cyber-Physical Systems makes it susceptible to failures due to complex environments, low-quality production, and aging. When defective, sensors either stop communicating or convey incorrect information. These unsteady situations threaten the safety, economy, and reliability of a system. The objective of this study is to construct a lightweight machine learning-based fault detection and diagnostic system within the limited energy resources, memory, and computation of a Wireless Sensor Network (WSN). In this paper, a Context-Aware Fault Diagnostic (CAFD) scheme is proposed based on an ensemble learning algorithm called Extra-Trees. To evaluate the performance of the proposed scheme, a realistic WSN scenario composed of humidity and temperature sensor observations is replicated with extreme low-intensity faults. Six commonly occurring types of sensor fault are considered: drift, hard-over/bias, spike, erratic/precision degradation, stuck, and data-loss. The proposed CAFD scheme reveals the ability to accurately detect and diagnose low-intensity sensor faults in a timely manner. Moreover, the efficiency of the Extra-Trees algorithm in terms of diagnostic accuracy, F1-score, ROC-AUC, and training time is demonstrated by comparison with cutting-edge machine learning algorithms: a Support Vector Machine and a Neural Network.

Joint Relay Selection and Power Allocation through a Genetic Algorithm for Secure Cooperative Cognitive Radio Networks.

In cooperative cognitive radio networks (CCRNs), there has been growing demand of transmitting secondary user (SU) source information secretly to the corresponding SU destination with the aid of cooperative SU relays. Efficient power allocation (PA) among SU relays and multi-relay selection (MRS) are a critical problem for operating such networks whereas the interference to the primary user receiver is being kept below a tolerable level and the transmission power requirements of the secondary users are being satisfied. Subsequently, in the paper, we develop the problem to solve the optimal solution for PA and MRS in a collaborative amplify-and-forward-based CCRNs, in terms of maximizing the secrecy rate (SR) of the networks. It is found that the problem is a mixed integer programming problem and difficult to be solved. To cope with this difficulty, we propose a meta-heuristic genetic algorithm-based MRS and PA scheme to maximize the SR of the networks while satisfying transmission power and the interference requirements of the networks. Our simulation results reveal that the proposed scheme achieves near-optimal SR performance, compared to the exhaustive search scheme, and provides a significant SR improvement when compared with some conventional relay selection schemes with equal power allocation.

Energy-Efficient Power Allocation and Relay Selection Schemes for Relay-Assisted D2D Communications in 5G Wireless Networks.

Device-to-device (D2D) communications allows user equipment (UE) that are in close proximity to communicate with each other directly without using a base station. Relay-assisted D2D (RA-D2D) communications in 5G networks can be applied to support long-distance users and to improve energy efficiency (EE) of the networks. In this paper, we first establish a multi-relay system model where the D2D UEs can communicate with each other by reusing only one cellular uplink resource. Then, we apply an adaptive neuro-fuzzy inference system (ANFIS) architecture to select the best D2D relay to forward D2D source information to the expected D2D destination. Efficient power allocation (PA) in the D2D source and the D2D relay are critical problems for operating such networks, since the data rate of the cellular uplink and the maximum transmission power of the system need to be satisfied. As is known, 5G wireless networks also aim for low energy consumption to better implement the Internet of Things (IoT). Consequently, in this paper, we also formulate a problem to find the optimal solutions for PA of the D2D source and the D2D relay in terms of maximizing the EE of RA-D2D communications to support applications in the emerging IoT. To solve the PA problems of RA-D2D communications, a particle swarm optimization algorithm is employed to maximize the EE of the RA-D2D communications while satisfying the transmission power constraints of the D2D users, minimum data rate of cellular uplink, and minimum signal-to-interference-plus-noise-ratio requirements of the D2D users. Simulation results reveal that the proposed relay selection and PA methods significantly improve EE more than existing schemes.

Joint Full-Duplex/Half-Duplex Transmission-Switching Scheduling and Transmission-Energy Allocation in Cognitive Radio Networks with Energy Harvesting.

The full-duplex transmission protocol has been widely investigated in the literature in order to improve radio spectrum usage efficiency. Unfortunately, due to the effect of imperfect self-interference suppression, the change in transmission power and path loss of non-line-of-sight fading channels will strongly affect performance of full-duplex transmission mode. This entails that the full-duplex transmission protocol is not always a better selection compared to the traditional half-duplex transmission protocol. Considering solar energy-harvesting-powered cognitive radio networks (CRNs), we investigate a joint full-duplex/half-duplex transmission switching scheduling and transmission power allocation in which we utilize the advantages of both half-duplex and full-duplex transmission modes for maximizing the long-term throughput of cognitive radio networks. First, we formulate the transmission rate of half-duplex and full-duplex links for fading channels between cognitive user and base station in which the channel gain is assumed to follow an exponential distribution. Afterward, by considering the availability probability of the primary channel, the limitation of the energy-harvesting capacity of the cognitive user, and the transmission capacity of half-duplex and full-duplex links, we describe the problem in terms of long-term expected throughput. The problem is then solved by adopting the partially observable Markov decision process framework to find the optimal transmission policy for the transmission pair between cognitive user and base station in order to maximize the long-term expected throughput. The optimal policy consists of either the half-duplex or the full-duplex transmission protocols as well as the corresponding amount of transmission energy in each time slot. In addition, to reduce the complexity in formulation and calculation, we also apply the actor⁻critic-based learning method to solve the considered problem. Finally, the performance of the proposed scheme was evaluated by comparing it with a conventional scheme in which the context of energy harvesting and long-term throughput is not considered.