Deep Learning Projects
CSE Projects, AI Projects, Deep Learning Projects
Description
D Deep Learning Projects focus on neural networks and advanced AI techniques for learning complex patterns from large datasets. We offer final year projects on image recognition, natural language processing, computer vision, speech processing, classification, prediction, and intelligent automation.Quality Factor
- Advanced AI Project Solutions
- Expert Deep Learning Guidance
- Modern Neural Network Techniques
- Affordable Project Packages
- Complete Implementation Support
1Enhancing Brain Source Reconstruction by Initializing 3-D Neural Networks With Physical Inverse Solutions
This project proposes a hybrid approach for improving brain source reconstruction using physical inverse solutions and three-dimensional neural networks. Conventional source reconstruction methods may suffer from inaccurate initialization and complex optimization requirements. The proposed system uses physically derived inverse solutions to initialize a 3-D neural network before learning from brain imaging data. This combination enables the model to learn meaningful spatial representations while reducing reconstruction errors. The approach can support more accurate analysis of brain activity and neurological research applications.
2Sub-Connection Learning for fMRI-Based Brain Functional Network
This project presents a machine learning approach for identifying meaningful sub-connections within functional brain networks using fMRI data. Brain imaging signals are processed to construct functional connectivity representations between different brain regions. The proposed sub-connection learning method identifies important connectivity patterns that may be overlooked by conventional approaches. These learned representations can be used for classification and neurological pattern analysis. The system supports improved understanding of brain functional organization and related disorders.
3Variational GAN-Enhanced Causal Effect Inference for Interpretable Learning in Heterogeneous IoT Systems
This project proposes a Variational GAN-based framework for causal effect inference in heterogeneous IoT environments. IoT systems generate diverse, incomplete, and complex data from multiple devices and sources. The proposed model generates useful data representations while identifying causal relationships between important variables. Explainable learning techniques are incorporated to make the resulting predictions easier to interpret. The system supports reliable and interpretable decision-making in complex IoT applications.
4ClustRecNet: A Novel End-to-End Deep Learning Framework for Clustering Algorithm Recommendation
ClustRecNet is an end-to-end deep learning framework designed to recommend suitable clustering algorithms for different datasets. The system analyzes dataset characteristics and learns relationships between data properties and clustering performance. A neural network automatically identifies the most appropriate clustering strategy based on the input dataset. This reduces the need for manual algorithm selection and extensive experimentation. The proposed framework can improve clustering efficiency and support automated machine learning workflows.
5OPDoctorNet: Deep Learning Revolutionizes Opportunistic Screening of Osteoporosis Based on Clinical Data
OPDoctorNet is a deep learning-based system designed for opportunistic osteoporosis screening using routinely available clinical information. Patient demographic, clinical, and health-related attributes are analyzed to identify individuals who may be at higher risk of osteoporosis. The deep learning model learns complex relationships between clinical features and osteoporosis risk. The system provides an automated risk assessment that can support early screening and clinical decision-making. This approach may improve early identification of osteoporosis without requiring dedicated screening procedures.
6Benchmark Suite for Resilience Assessment of Deep Learning Models
This project develops a benchmark suite for evaluating the resilience of deep learning models under different challenging conditions. Models are tested against factors such as noisy inputs, distribution changes, adversarial disturbances, and data corruption. Standardized evaluation metrics are used to compare model stability and performance degradation. The benchmark provides a common framework for analyzing the reliability of different deep learning architectures. It supports the development of more robust and dependable AI systems.
7ML-Enabled Dynamic Duplexing for Diverse Traffic Scenarios: A Deep Reinforcement Learning Approach
This project proposes a deep reinforcement learning approach for dynamically managing duplexing modes in wireless communication networks. The system monitors traffic conditions, channel characteristics, and communication requirements in real time. A reinforcement learning agent learns optimal duplexing decisions based on changing network conditions. The approach aims to improve spectrum utilization, throughput, and communication efficiency. It provides an adaptive solution for wireless networks with diverse and dynamic traffic patterns.
8Characterization and Classification of Tactical Movements Using Wearable Motion Sensors and Deep Learning Models
This project develops a wearable sensor-based system for identifying and classifying tactical human movements. Motion sensors capture acceleration, orientation, velocity, and other movement characteristics during different activities. Deep learning models automatically extract temporal and spatial features from the collected sensor signals. The system classifies different tactical movement patterns with minimal manual feature engineering. The approach can support activity monitoring, training assessment, and intelligent movement recognition applications.
9CSI-4CAST: A Hybrid Deep Learning Model for CSI Prediction With Comprehensive Robustness and Generalization Testing
CSI-4CAST is a hybrid deep learning framework designed to predict Channel State Information in wireless communication systems. The model combines different neural learning techniques to capture both spatial and temporal characteristics of CSI data. Extensive robustness and generalization testing is performed under different channel and environmental conditions. The predicted CSI can assist beamforming, resource allocation, and wireless optimization tasks. The system aims to provide accurate and reliable channel prediction for advanced wireless networks.
10The Evolution and Future Perspectives of Artificial Intelligence-Generated Content
This project examines the development and future direction of artificial intelligence-generated content technologies. It analyzes how generative AI models have evolved in producing text, images, audio, video, and other digital content. The study explores major technological advances, applications, opportunities, and challenges associated with AI-generated content. Ethical concerns, reliability, copyright, and responsible AI usage are also considered. The project provides an overview of the future impact of generative AI across different domains.
11Enhanced Identification of Chronic Ankle Instability Under Different Conditions: A New Evaluation Framework Based on Feature Fusion and Machine Learning
This project proposes a machine learning framework for identifying chronic ankle instability under different physical and experimental conditions. Sensor-based movement data is collected and processed to extract relevant biomechanical characteristics. Feature fusion combines complementary information from multiple measurements to improve classification performance. Machine learning models analyze the fused features to distinguish stable and unstable ankle conditions. The system supports objective assessment and rehabilitation planning for individuals with chronic ankle instability.
12Adversarially Optimized Multi-Space Prototypical Network for Intrusion Detection in 5G-Enabled IoT Systems
This project proposes an adversarially optimized prototypical network for detecting cyberattacks in 5G-enabled IoT systems. IoT network traffic is processed to identify normal and malicious communication patterns. Prototypical learning enables the model to classify network activities effectively even with limited labeled attack samples. Adversarial optimization improves the model's ability to handle complex and evolving intrusion patterns. The system provides an intelligent and adaptable cybersecurity solution for 5G-IoT environments.
13Charting 15 Years of Progress in Deep Learning for Speech Emotion Recognition: A Replication Study
This project reviews and replicates research progress in deep learning-based speech emotion recognition over a fifteen-year period. Different datasets, deep learning architectures, feature extraction techniques, and evaluation methods are analyzed. Replication experiments are used to examine the consistency and reproducibility of previously reported results. Performance comparisons identify trends and limitations in existing approaches. The study provides useful insights for developing reliable speech emotion recognition systems.
14SAFE-ADVENT: Privacy-Preserving and Resilient Distributed Federated Learning for DDoS Attacks Detection in 5G-V2X Networks
SAFE-ADVENT is a federated learning framework designed to detect Distributed Denial-of-Service attacks in 5G-enabled vehicle-to-everything networks. Network data remains locally stored while participating devices collaboratively train an intrusion detection model. Privacy-preserving mechanisms reduce the need to share sensitive vehicle and network information. The framework also focuses on resilience against changing attack behaviors and unreliable network conditions. The system provides secure and scalable cybersecurity monitoring for intelligent transportation environments.
15Lower Back Muscle Fatigue Recognition Based on the Fusion-Information of Multi-Channel sEMG and NIRS Simultaneous Recordings
This project detects lower back muscle fatigue by combining surface electromyography and near-infrared spectroscopy signals. Multiple sEMG channels capture muscle electrical activity while NIRS provides information about muscle oxygenation and blood-related changes. The two signal modalities are synchronized and fused to obtain complementary fatigue indicators. Machine learning models analyze the combined features to classify different levels of muscle fatigue. The system can support ergonomic assessment, occupational safety, and physical workload monitoring.
16HDF-Net: A Hybrid Deep Learning Model for Retrieving Ozone Profiles in the Arctic
HDF-Net is a hybrid deep learning model designed to estimate ozone profiles in the Arctic atmosphere. Atmospheric and remote sensing measurements are processed to obtain relevant environmental features. The hybrid neural architecture learns complex relationships between observed atmospheric variables and ozone concentration at different altitudes. The predicted profiles are compared with reference measurements to evaluate model accuracy. The system supports atmospheric monitoring, climate research, and improved understanding of Arctic ozone behavior.
17Deep Learning-Driven Decision Fusion: Spatio-Spectrogram Features for Inner Speech Recognition From Electroencephalogram Signals
This project develops a deep learning system for recognizing inner speech from EEG signals. EEG recordings are transformed into spatial and spectrogram-based representations to capture complementary temporal and frequency information. Deep learning models independently analyze these representations before combining their decisions through a fusion mechanism. The resulting system improves recognition of internally generated speech patterns without relying on spoken audio. The approach can contribute to brain-computer interfaces and assistive communication technologies.
18Deep Feature Learning From Electromyographic Signals for Gesture Recognition Systems
This project develops a deep learning-based gesture recognition system using electromyographic signals. EMG sensors capture muscle activity generated during different hand and arm gestures. Signal preprocessing is performed to reduce noise and prepare the data for feature learning. Deep neural networks automatically extract discriminative patterns and classify the corresponding gestures. The system can be applied to prosthetic control, wearable interfaces, robotics, and human-computer interaction.
19Enhancing Brain Source Reconstruction by Initializing 3-D Neural Networks With Physical Inverse Solutions
This project combines physical inverse modeling with 3-D neural networks to improve the reconstruction of neural sources from brain measurements. Physical inverse solutions provide an informed starting point for the neural network instead of relying only on random initialization. The neural model then learns spatial patterns from brain data and refines the initial reconstruction. This hybrid strategy can improve convergence and reconstruction accuracy. The system supports advanced brain activity localization and neurological analysis.
20Deep Reinforcement Learning-Based Scheduling for Wi-Fi Multi-Access Point Coordination
This project uses deep reinforcement learning to coordinate scheduling decisions across multiple Wi-Fi access points. The system continuously observes traffic demand, channel conditions, interference, and user requirements. A reinforcement learning agent learns optimal scheduling policies through interaction with the wireless environment. The proposed approach aims to improve throughput, reduce interference, and balance network resources. It provides an intelligent solution for efficient management of dense Wi-Fi environments.
21Benchmark Suite for Resilience Assessment of Deep Learning Models
This project establishes a standardized benchmark for measuring the resilience of deep learning systems. Different neural network architectures are evaluated under noisy, corrupted, adversarial, and distribution-shifted inputs. Performance degradation and recovery behavior are measured using common evaluation metrics. The benchmark allows researchers to compare model robustness under consistent testing conditions. It supports the design and selection of reliable deep learning models for real-world applications.
22ML-Enabled Dynamic Duplexing for Diverse Traffic Scenarios: A Deep Reinforcement Learning Approach
This project applies machine learning and deep reinforcement learning to dynamically select suitable duplexing configurations in wireless networks. The system observes variations in traffic demand, channel quality, and network conditions. A learning agent determines appropriate duplexing decisions to maximize communication efficiency. The approach adapts automatically to different traffic scenarios without requiring fixed configuration rules. It can improve network throughput, spectrum utilization, and overall wireless performance.
23Beyond Subject-Specific Models in Dynamical Human–Machine Interaction: Benchmarking and Optimization Strategies
This project investigates machine learning models for human-machine interaction that can generalize across different users. Traditional systems often require individual-specific training, which limits scalability and practical deployment. The proposed approach benchmarks different learning and optimization strategies for improving cross-subject performance. Data from multiple users is analyzed to identify generalizable behavioral patterns. The system aims to develop adaptable human-machine interaction models requiring less personalized calibration.
24Characterization and Classification of Tactical Movements Using Wearable Motion Sensors and Deep Learning Models
This project uses wearable motion sensors to capture and analyze tactical human movement patterns. Sensor signals are processed to identify temporal and spatial characteristics associated with different tactical activities. Deep learning models learn meaningful representations directly from the sensor data. The trained system classifies movements automatically and evaluates recognition performance. The solution can support tactical training, activity recognition, performance analysis, and wearable monitoring.
25Fast Real-World Classification of ECH-Enabled Applications
This project develops a machine learning system for rapidly classifying ECH-enabled applications in real-world environments. Application-related features are collected and processed to identify distinguishing characteristics. A lightweight classification model is trained to provide fast predictions while maintaining acceptable accuracy. Real-world testing is performed to evaluate the system under changing application and network conditions. The solution can support intelligent network management and application identification.
26HypBench: Hyperbolic Benchmark for Graph Neural Network Performance
HypBench is a benchmark framework for evaluating graph neural networks operating in hyperbolic spaces. Graph datasets with hierarchical and complex relationships are used to compare different graph learning approaches. The benchmark evaluates model accuracy, computational performance, scalability, and generalization. Results help identify situations where hyperbolic representations provide advantages over conventional Euclidean methods. The framework supports systematic evaluation of graph neural network architectures.
27Digital Twin-Driven Continual Deep Reinforcement Learning for Coexistence of Multiple Radio Access Technology IoT Links With Nonlinear Receivers
This project proposes a digital twin-driven reinforcement learning framework for managing multiple radio access technologies in IoT environments. A digital twin represents the communication environment and continuously reflects changing network conditions. Continual deep reinforcement learning enables the system to adapt its decisions as traffic, interference, and receiver characteristics change. The framework focuses on efficient coexistence of multiple wireless links. It supports adaptive resource management and reliable IoT communication.
28CSI-4CAST: A Hybrid Deep Learning Model for CSI Prediction With Comprehensive Robustness and Generalization Testing
CSI-4CAST develops a hybrid deep learning architecture for accurate prediction of wireless Channel State Information. Historical channel measurements are processed to learn temporal and spatial signal dependencies. The model is evaluated under different environments, channel conditions, and data variations to measure robustness and generalization. Accurate CSI prediction can improve beamforming and wireless resource allocation. The framework provides a practical approach for reliable channel prediction in modern wireless systems.
29Assessing the Robustness of Deep Learning Based Brain Age Prediction Models Across Multiple EEG Datasets
This project evaluates how reliably deep learning models can predict brain age using EEG signals from different datasets. EEG recordings are preprocessed and used to train and test neural network-based age prediction models. Cross-dataset evaluation measures whether learned patterns remain effective when data sources and populations change. Robustness metrics are used to identify factors affecting prediction consistency. The study supports the development of generalizable brain age estimation systems.
30The Evolution and Future Perspectives of Artificial Intelligence-Generated Content
This project studies the evolution of AI-generated content from early generative methods to modern foundation and multimodal models. It examines applications in text generation, image synthesis, audio creation, video generation, and interactive content production. The study also analyzes limitations related to accuracy, bias, copyright, privacy, and ethical use. Future developments in generative AI are explored from both technological and societal perspectives. The project provides a comprehensive view of the opportunities and challenges surrounding AI-generated content.
31Network Slice Placement in Large Scale Infrastructures Using Efficient and Highly Scalable Deep Reinforcement Learning Agent
This project proposes a scalable deep reinforcement learning solution for network slice placement in large-scale communication infrastructures. The system considers computing resources, network capacity, latency, and service requirements when placing network slices. A reinforcement learning agent learns efficient placement strategies through continuous interaction with the network environment. The approach aims to reduce resource wastage and improve service quality. It provides an intelligent solution for scalable network slicing management.
32Enhanced Identification of Chronic Ankle Instability Under Different Conditions: A New Evaluation Framework Based on Feature Fusion and Machine Learning
This project introduces a feature-fusion-based machine learning framework for identifying chronic ankle instability. Multiple sensor and biomechanical features are collected under different movement conditions. Feature fusion combines complementary information to improve the representation of ankle movement behavior. Machine learning models classify stable and unstable conditions based on the fused feature set. The system can assist clinicians in objective assessment, diagnosis support, and rehabilitation monitoring.
33Charting 15 Years of Progress in Deep Learning for Speech Emotion Recognition: a Replication Study
This study examines fifteen years of deep learning research in speech emotion recognition through systematic replication. Different neural architectures, speech features, datasets, and evaluation approaches are compared. Selected previous experiments are reproduced to determine whether reported performance can be consistently achieved. The analysis identifies improvements, limitations, and reproducibility challenges across the research period. The findings provide guidance for future development of robust speech emotion recognition systems.
34A Deep Reinforcement Learning Approach to Time Delay Differential Game Deception Resource Deployment
This project uses deep reinforcement learning to optimize resource deployment in time-delay differential game environments. The system considers delayed information and strategic interactions between competing agents. A reinforcement learning agent learns how to allocate deception resources under changing conditions. The model aims to maximize strategic effectiveness while considering resource limitations and delayed responses. The approach can support intelligent decision-making in complex security and defense scenarios.
35Enhancing Dynamic Security Assessment in Smart Grids Through Quantum Federated Learning
This project combines quantum machine learning concepts with federated learning for dynamic smart grid security assessment. Distributed grid components collaboratively train security models without directly sharing sensitive operational data. Quantum-enhanced learning techniques are explored to handle complex patterns and improve threat analysis. The system continuously evaluates changing grid conditions to identify potential security risks. It provides a privacy-aware and intelligent approach to smart grid cybersecurity.
36Handwritten Text Recognition: A Survey
This project provides a comprehensive survey of techniques used for handwritten text recognition. Traditional image processing, optical character recognition, machine learning, and modern deep learning approaches are reviewed. The study examines preprocessing, segmentation, feature extraction, sequence modeling, and recognition techniques. Different datasets, evaluation metrics, and application areas are also compared. The survey highlights current challenges and future research directions in automated handwriting recognition.
37Digital Twin-Assisted Explainable AI for Robust Beam Prediction in mmWave MIMO Systems
This project proposes a digital twin-assisted framework for robust beam prediction in millimeter-wave MIMO communication systems. The digital twin represents the wireless environment and provides simulated information for training and prediction. Explainable AI techniques are used to identify the factors influencing beam selection decisions. The model predicts suitable beam directions under changing environmental and channel conditions. The system aims to improve communication reliability, beamforming efficiency, and model interpretability.
38Deterministic and Probabilistic Forecasting of Wind Power Generation and Ramp Rate With Expectation-Implemented Deep Learning
This project develops deep learning models for forecasting wind power generation and sudden ramp-rate changes. Historical wind speed, weather, and power generation data are analyzed to learn temporal relationships. Both deterministic predictions and probabilistic forecasts are generated to represent expected values and uncertainty. The forecasting results can support grid operators in planning renewable energy integration. The system improves decision-making for reliable and efficient wind power management.
39Integrating Clinical Knowledge Graphs and Gradient-Based Neural Systems for Enhanced Melanoma Diagnosis via the Seven-Point Checklist
This project combines clinical knowledge graphs with neural networks to improve melanoma diagnosis using the Seven-Point Checklist. Patient and lesion characteristics are represented as structured clinical knowledge and combined with learned neural features. The system evaluates important diagnostic indicators and generates a classification of potentially suspicious lesions. Knowledge-based information improves interpretability and clinical relevance of the prediction. The approach supports computer-assisted melanoma screening and diagnostic decision-making.
40Fine-Tuning Myoelectric Control Through Reinforcement Learning in a Game Environment
This project uses reinforcement learning to improve myoelectric control through an interactive game environment. Electromyographic signals from muscle activity are converted into control commands for game actions. The reinforcement learning agent learns from user feedback and task performance to optimize control mappings. Continuous interaction allows the system to adapt to individual muscle signal patterns. The approach can support personalized prosthetic control and human-machine interaction systems.
41Flapping-Wings Drones for Pests and Diseases Detection in Horticulture
This project develops an intelligent flapping-wing drone system for detecting pests and diseases in horticultural environments. The drone captures images of crops using onboard cameras while flying through agricultural fields. Computer vision and deep learning models analyze the images to identify symptoms, pests, and disease patterns. Detection results can be used to locate affected plants and support targeted intervention. The system provides a flexible approach to automated crop monitoring and precision agriculture.
42Leveraging Swin Transformer for Enhanced Diagnosis of Alzheimer’s Disease Using Multi-Shell Diffusion MRI
This project uses a Swin Transformer to analyze multi-shell diffusion MRI data for Alzheimer's disease diagnosis. Diffusion MRI captures detailed information about brain tissue structure and connectivity. The proposed Transformer model learns spatial relationships and complex imaging patterns associated with Alzheimer's disease. The system classifies subjects based on learned neuroimaging representations and evaluates diagnostic performance. The approach supports advanced medical image analysis and computer-assisted neurological diagnosis.
43MirrorFuzz: Leveraging LLM and Shared Bugs for Deep Learning Framework APIs Fuzzing
MirrorFuzz is an intelligent fuzzing framework that uses large language models to identify potential bugs in deep learning framework APIs. The system generates diverse and meaningful API inputs based on learned programming patterns and previously identified issues. Shared bug information is used to improve the generation of test cases and increase testing effectiveness. The framework automatically executes generated cases and analyzes abnormal behavior. It supports systematic reliability and security testing of deep learning software frameworks.
44From Inconsistency to Unity: Benchmarking Deep Learning-Based Unsupervised Domain Adaptation for RUL
This project benchmarks deep learning-based unsupervised domain adaptation techniques for remaining useful life prediction. Different datasets and operating conditions may create significant differences between training and deployment environments. Domain adaptation methods are evaluated for their ability to transfer learned knowledge without requiring extensive target-domain labels. Multiple architectures and performance metrics are compared systematically. The study helps identify effective strategies for reliable RUL prediction across different operating domains.
45Explainable Normative Modeling for Brain Disorder Identification in Resting-State fMRI
This project develops an explainable normative modeling approach for identifying brain disorders using resting-state fMRI data. The system learns patterns representing normal brain functional behavior across a population. Individual brain measurements are compared against the learned normative distribution to identify unusual patterns. Explainable methods highlight brain regions and connectivity characteristics contributing to abnormality detection. The approach supports personalized neurological assessment and interpretable brain disorder identification.
46Fusion-Based Off-Grid Direction of Arrival Estimation via Deep Learning With Array Imperfection
This project proposes a deep learning approach for estimating the direction of arrival of signals when sensor arrays contain imperfections. Real-world array errors such as sensor mismatch and positioning inaccuracies can reduce conventional estimation accuracy. Multiple signal representations are fused to provide richer information for the learning model. The neural network estimates signal directions even when the actual source lies between predefined grid points. The system improves robustness of direction estimation for practical sensing and communication applications.
47Low-Light Image Enhancement via Diffusion Models With Semantic Priors of Any Region
This project develops a diffusion-model-based approach for enhancing images captured under low-light conditions. Semantic information is incorporated to preserve important objects, structures, and visual details during enhancement. The diffusion model learns to reconstruct brighter and more natural image content while reducing noise and distortion. The system can process different image regions according to their semantic characteristics. It supports improved image quality for computer vision, surveillance, and photography applications.
48FedEMG: Achieving Generalization, Personalization, and Resource Efficiency in EMG-Based Upper-Limb Rehabilitation Through Federated Prototype Learning
FedEMG is a federated learning framework designed for personalized upper-limb rehabilitation using electromyographic signals. EMG data from multiple users is processed locally to preserve privacy while contributing to collaborative model training. Prototype learning helps the model generalize across users while maintaining personalized recognition capabilities. Resource-efficient training reduces communication and computational requirements for wearable devices. The system supports intelligent, privacy-preserving, and personalized rehabilitation monitoring.
49Bridging Speech Emotion Recognition and Personality: Dataset and Temporal Interaction Condition Network
This project investigates the relationship between speech emotion recognition and personality characteristics. A dedicated dataset is used to capture speech patterns, emotional states, and personality-related information. A temporal interaction condition network learns relationships between speech features, emotional expressions, and personality traits. The model analyzes how emotional characteristics change over time and influence personality prediction. The system can support advanced affective computing, behavioral analysis, and human-computer interaction applications.
50Evaluating Facial Expression Recognition Datasets for Deep Learning: a Benchmark Study with Novel Similarity Metrics
This project benchmarks facial expression recognition datasets to determine their suitability for deep learning applications. Dataset characteristics such as image quality, class distribution, facial diversity, and expression similarity are analyzed. Novel similarity metrics are introduced to compare the visual and semantic relationships between datasets. Deep learning experiments are performed to evaluate how dataset selection influences recognition performance. The study provides useful guidance for selecting and developing reliable facial expression datasets.
51The Evolution and Future Perspectives of Artificial Intelligence-Generated Content
This project examines the historical development and future direction of AI-generated content systems. It studies the progression from traditional generative models to modern large-scale multimodal AI systems. Applications across text, image, speech, music, video, and digital media are explored. The project also discusses challenges including misinformation, copyright, bias, privacy, and responsible AI usage. The study provides an overview of how generative AI may influence future digital content creation.
52Deep Feature Learning From Electromyographic Signals for Gesture Recognition Systems
This project proposes a deep feature learning approach for recognizing gestures from electromyographic signals. Muscle activity is captured using EMG sensors while users perform predefined gestures. Signal preprocessing and normalization are applied before feeding the data into deep learning models. The network automatically learns discriminative features and classifies gestures with minimal manual feature extraction. The system can support wearable interfaces, assistive devices, prosthetics, and robotic control.
53Courier Working Time Aware Vehicle Scheduling for Efficient Urban Logistics
This project develops an intelligent vehicle scheduling system that considers courier working hours and urban delivery constraints. Delivery locations, vehicle availability, courier schedules, travel times, and delivery requirements are analyzed to create efficient routes. Optimization and machine learning techniques can be used to minimize travel distance, delivery delays, and operational costs. The system dynamically adjusts schedules according to available working time and changing delivery conditions. It supports efficient and practical urban logistics management.
54Extending Multiscale Characterization of Heart Rate Variability via Deep Learning for Mortality Risk Prediction
This project uses deep learning to analyze heart rate variability at multiple temporal scales for mortality risk prediction. ECG or heart rate signals are processed to extract variations that may indicate physiological instability. Multiscale representations are combined with deep learning models to identify complex risk-related patterns. The system generates a predictive assessment of mortality risk based on learned cardiovascular characteristics. It can support clinical risk assessment and data-driven patient monitoring.
55Classification of Functional Brain Patterns Elicited by Deep Brain Stimulation of the Subthalamic Nucleus in Parkinson’s Disease
This project analyzes functional brain patterns produced by deep brain stimulation of the subthalamic nucleus in Parkinson's disease. Brain imaging or neural activity data is collected under different stimulation conditions. Machine learning techniques extract and classify patterns associated with stimulation responses and neurological changes. The system can help identify differences between functional brain states and evaluate stimulation effects. The approach supports research into personalized deep brain stimulation and improved Parkinson's disease treatment strategies.
