Abdulbasit K. Al-Talabani

Morphing attacks on facial images pose a serious threat to face recognition systems during identity verification. In this paper, we introduce a parameter-efficient family of convolutional neural networks, named EfficientNet-Lite (EN_Lite), for morph attack detection (MAD). The experimental framework uses eight datasets, four morph generation methods, and ten homogeneous and heterogeneous evaluation scenarios under a leave-one-datasetout protocol. Performance was assessed using dataset-size-weighted average equal error rate (EERWA ), computed across the held-out datasets within each scenario. EN_Lite was evaluated using ImageNet1K and VGGFace2-HQ pre-training, with the latter comprising 1,107,817 successfully processed facial images. The ImageNetpretrained IEN_Lite-1 achieved a mean EERWA of 2.99%, compared with 2.65% for EfficientNet-B3 while using approximately 76% fewer parameters. It also achieved a lower mean EERWA than GhostNet-1.0x (4.01%), MobileNetV3-Large (5.03%), MobileNetV3-Small (6.03%), and ShuffleNet V2 (5.31%). The VGGFace2-HQ-pretrained variants produced higher error rates than their corresponding ImageNet-pretrained variants under the evaluated protocol. Performance remained strong in homogeneous scenarios but degraded under heterogeneous cross-morph evaluation. Overall, EN_Lite provides a favorable efficiency–performance trade-off for MAD, while high-security deployment should be assessed at application-specific operating points.

Cepstral features, such as Gammatone Cepstral Coefficients (GTCC), have recently been applied in fault detection and diagnosis. However, GTCC was originally designed for speech feature extraction rather than fault detection, which limits its ability to effectively capture relevant features for fault identification. In this research, three key parameters of GTCC namely, the number of coefficients, maximum frequency, and minimum frequency are optimized using two metaheuristic algorithms: Fitness Dependent Optimizer (FDO) and Grey Wolf Optimization (GWO). These parameters are vital for enhancing the effectiveness of GTCC in extracting relevant features from the vibration signal for gear defect identification. Specifically, the maximum and minimum frequency values are critical for capturing the Gear Mesh Frequency (GMF), a common indicator of gear faults. Furthermore, optimizing the number of GTCC coefficients helps reduce model complexity. Experimental results demonstrate that optimizing GTCC parameters with GWO improves fault detection performance using an SVM classifier, achieving over 1% and 3% accuracy improvements on the PHM09 and DDS datasets, respectively.

Detecting fake speech in voice-based authentication systems is crucial for reliability. Traditional methods often struggle because they can't handle the complex patterns over time. Our study introduces an advanced approach using deep learning, specifically Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) models, tailored for identifying fake speech based on its temporal characteristics. We use speech signals with cepstral features like Mel-frequency cepstral coefficients (MFCC), Constant Q cepstral coefficients (CQCC), and open-source Speech and Music Interpretation by Large-space Extraction (OpenSMILE) to directly learn these patterns. Testing on the ASVspoof 2019 Logical Access dataset, we focus on metrics such as min-tDCF, Equal Error Rate (EER), Recall, Precision, and F1-score. Our results show that LSTM and BiLSTM models significantly enhance the reliability of spoof speech detection systems.

The field of multimodal emotion recognition is increasingly gaining popularity as a research area. It involves analyzing human emotions across multiple modalities, such as acoustic, visual, and language. Emotion recognition is more effective as a multimodal learning task than relying on a single modality. In this paper, we present an unimodal and multimodal long short-term memory model with a class weight parameter technique for emotion recognition on the CMU-Multimodal Opinion Sentiment and Emotion Intensity dataset. In addition, a critical challenge lies in selecting the most effective fusion method for integrating multiple modalities. To address this, we applied four different fusion techniques: Early fusion, late fusion, deep fusion, and tensor fusion. These fusion methods improved the performance of multimodal emotion recognition compared to unimodal approaches. With the highly imbalanced number of samples per emotion class in the MOSEI dataset, adding a class weight parameter technique leads our model to outperform the state of the art on all three modalities — acoustic, visual, and language — as well as on all the fusion models. The challenges of class imbalance, which can lead to biased model performance, and using an effective fusion method for integrating multiple modalities often result in decreased accuracy in recognizing less frequent emotion classes. Our proposed model shows 2–3% performance improvement in the unimodal and 2% in the multimodal over the state-of-the-art achieved results.

Electrocardiogram (ECG) analysis is widely used by cardiologists and medical practitioners for monitoring cardiac health. A high-performance automatic ECG classification system is challenging because there is difficulty in detecting and categorizing different waveforms in the signal, especially in manual analysis of ECG signals, which means, a better classification system is needed in terms of performance and accuracy. Hence, in this paper, the authors propose an accurate ECG classification and monitoring system called convolutional neural network-k nearest neighbor (CNN-kNN). The proposed method utilizes 1D-CNN and kNN. Unlike the existing techniques, the examined technique does not need training during classifying the ECG signals. The CNN-kNN is evaluated against the PhysioNet’s MIT-BIH and PTB diagnostics datasets. The CNN is fed using the ECG beat raw signal directly. In addition, the learned features are extracted from the 1D-CNN model and its dimensions are reduced using two fully connected layers and then fed to the k-NN classifier. The CNN-kNN model achieved average accuracies of 98% and 97.4% on arrhythmia and myocardial infarction classifications, respectively. These results are evidence of the great ability of the proposed model compared to the mentioned models in this article.

Feature extraction and representation has significant impact on the performance of any machine learning method. Mel Frequency Cepstrum Coefficient (MFCC) is designed to model features of audio signal and is widely used in various fields. This paper aims to review the applications that the MFCC is used for in addition to some issues that facing the MFCC computation and its impact on the model performance. These issues include the use of MFCC for non-acoustic signals, adopting the MFCC alone or combining it with other features, the use of time series versus global representation of the MFCC, following the standard form of the MFCC computation versus modifying its parameters, and supplying the traditional machine learning methods versus the deep learning methods.