Iranian Journal of Electrical and Electronic Engineering

---

The incidence of heart-related illnesses is on the rise worldwide. Heart diseases are primarily caused by a multitude of parameters, including high blood pressure, diabetes, and excessive cholesterol, which are controlled by poor dietary and lifestyle choices. The growth in cardiovascular diseases (CVD) is mostly due to several other behaviors, such as smoking, drinking, and sleeplessness. In the research, machine learning-based prediction methods work on the audio recordings of heartbeats known as phonocardiograms (PCG) to develop an algorithm that differentiates a normal healthy heart from an abnormal heart based on the heart sounds. The data set consists of 831 normal and 260 abnormal data, and the duration of each sample is 5 seconds. Features extracted from the data are up-sampled and applied to the logistic regression and random forest classification models. The developed models record a classification accuracy of 71% for logistic regression and 94% for the random forest model. Further, artificial neural networks (ANN) and Deep learning networks have been trained to improve performance and demonstrated an accuracy of 94.5%.

---

This document presents the design of a virtual robotic system for the supervision of physical training exercises, to be carried out in a closed environment, which only requires a computer equipment with a web camera. To do this, deep learning algorithms such as convolutional networks and short- and long-term memory networks are used to recognize voice commands and the user's video actions. A predefined dialogue template is used to guide a user's training cycle based on the execution of the exercises: push-ups, abdominal, jump or squat. The contribution of the work focuses on the integration of deep learning techniques to design and personalize virtual robotic assistants for everyday task. The results show a high level of accuracy by the virtual robot both in understanding the audio and in predicting the exercise to be performed, with a final accuracy value of 97.75% and 100%, respectively.

---

Abnormal activity detection is crucial for video surveillance and security systems, aiming to identify behaviors that deviate from normal patterns and may indicate threats or incidents such as theft, vandalism, accidents, and aggression. Timely recognition of these activities enhances public safety across various environments, including transportation hubs, public spaces, workplaces, and homes. In this study, we focus on detecting violent and non-violent activities of humans using a YOLOv9-based deep learning model considering the above issues. A diverse dataset has been built of 9,341 images from various platforms, and then the dataset has been pre-processed, i.e., augmentation, resizing, and annotating. After pre-processing, the proposed model has been trained which demonstrated strong performance, achieving an F1 score of 95% during training for 150 epochs. It was also trained for 200 epochs, but early stopping was applied at 148 epochs as there was no significant improvement in the results. Finally, the results of the YOLOv9-based model have been analyzed with other baseline models (YOLOv5, YOLOv7, YOLOv8, and YOLOv10) and it performed better compared with others.

---

High peak-to-average power ratio (PAPR) has been a major drawback of Filter bank Multicarrier (FBMC) in the 5G system. This research aims to calculate the PAPR reduction associated with the FBMC system. This research uses four techniques to reduce PAPR. They are classical tone reservation (TR). It combines tone reservation with sliding window (SW-TR). It also combines them with active constellation extension (TRACE) and with deep learning (TR-Net). TR-net decreases the greatest PAPR reduction by around 8.6 dB compared to the original value. This work significantly advances PAPR reduction in FBMC systems by proposing three hybrid methods, emphasizing the deep learning-based TRNet technique as a groundbreaking solution for efficient, distortion-free signal processing.

---

This study proposes a descriptor-based approach combined with deep learning, which recognizes facial emotions for safe driving.  Paying attention to the driver's facial expressions is crucial to address the increasing road accidents. This project aims to develop a Facial Emotion Recognition (FER) system that monitors the driver's facial expressions to identify emotions and provide instant assistance for safety control. In the initial stage, Viola-Jones face detection was employed to detect the facial region, followed by Butterworth high-pass filtering to enhance the identified region for locating the eye, nose, and mouth regions, using Viola-Jones face detection. Secondly, the Local Binary Patterns (LBP) feature descriptor is utilized to extract features from the identified eye, nose, and mouth regions. Using 3 RGB channels, the extracted features from these three regions are fed into RessNet-50 and EfficientNet deep networks. The outputs of the two deep learning models' classifiers are combined and integrated using two ensemble methods: ensemble maximum voting and ensemble mean. Based on these combining classifier rules, the performance was evaluated on the JAFFE and KMU-FED databases. The experimental results demonstrate that the proposed method can effectively and with higher accuracy than other competitors recognize emotions in the JAFFE and KMU-FED datasets. The novelty and originality of this paper lie in its significant application in the automotive industry. Implementing our proposed method in a system capable of high accuracy and precision can help mitigate numerous driving hazards. Our approach has achieved 99% and 98% accuracy on the JAFFE and KMU-FED databases, respectively. This high level of accuracy, coupled with its practical relevance, underscores the innovative nature of our work.

---

This paper presents an application of deep learning in computer graphics, utilizing learn-based networks for 3D shape matching. We propose an efficient method for shape matching between 3D models with non-isometric deformation. Our method organizes intrinsic and directional attributes in a structured manner. For this purpose, we use a hybrid feature derived from Diffusion-Net and spectral features. In fact, we combine learned-based intrinsic properties with orientation-preserving features and demonstrate the effectiveness of our method. We achieve this by first extracting features from Diffusion-Net. Then, we compute two maps based on the functional map networks to obtain intrinsic and directional features. Finally, we combine them to achieve a desired map that can resolve symmetry ambiguities on models with high deformation. Quantitative results on the TOSCA dataset indicate that the proposed method achieves lowest average geodetic error of 0.0023, outperforming state-of-the-art methods and reducing the error by 70.66%. We demonstrate that our method outperforms similar approaches by leveraging an accurate feature extractor and effective geometric regularizers, allowing for better handling of non-isometric shapes and resulting in reduced matching errors.