جامعة الانبار

This work presents the design and development of “Al-Rehab Glove,” a wearable smart-glove system that monitors and analyzes hand tremor in Parkinson’s patients using inertial sensing and artificial intelligence. The glove embeds an MPU6050 accelerometer-gyroscope module and an Arduino Nano microcontroller to record three-axis motion signals, which are stored on an SD card and later processed in MATLAB. After noise reduction and feature extraction, three supervised machine-learning algorithms—K-Nearest Neighbor (KNN), Decision Tree, and Naïve Bayes—are applied to classify tremor severity as mild, moderate, or severe. The classification result then drives a pneumatic actuation mechanism, built from flexible air tubes and finger holders, that assists finger movement at a treatment level matched to the detected severity. Experimental evaluation on motion data collected from Parkinson’s patients showed that the Decision Tree algorithm achieved the highest classification accuracy among the three models, confirming the feasibility of the proposed system as an objective, continuous tool for tremor monitoring and rehabilitation support.

Attend AI is an intelligent university attendance management system that deploys three concurrent neural network models executing entirely within the web browser, achieving real-time student identity verification and uniform compliance detection without server-side GPU dependency. The system implements a TinyFaceDetector (SSD + MobileNet v1 backbone) for sub-100ms face localization, a 68-point facial landmark network for geometric alignment normalization, and a ResNet-based face recognition network generating 128-dimensional embedding vectors with Euclidean distance matching. A custom RGB color-space uniform compliance engine operates in parallel using face bounding box geometry — requiring zero additional neural network inference. Key innovations include: (1) client-side triple neural network inference eliminating cloud GPU costs, (2) dual-output single-pass detection combining identity and dress-code verification, (3) enrollment-aware attendance validation preventing cross-section fraud, and (4) a five-level RBAC architecture with 15 granular permissions. Evaluation results: face recognition accuracy 96.2% (F1=0.974), uniform compliance accuracy 91.8%, end-to-end latency under 200ms on standard consumer hardware. Built with Next.js 15, React 19, PostgreSQL (Neon), Prisma 6, and TensorFlow.js.

HerbalDr AI is a cross-platform intelligent medical consultation system supporting 13 Arabic dialects and 5 English variants. The system uses a dual-AI engine that dynamically routes requests between Google Gemini 2.5 Flash and Meta Llama 3.3 70B, delivering culturally-sensitive herbal medicine guidance to 422 million Arabic speakers. Key innovations include: (1) a Firestore-backed dynamic prompt management engine enabling zero-downtime behavioral updates, (2) a quantitative Risk Scoring Algorithm classifying patient severity on a 0–100 scale, (3) a 3-tier automated ban escalation system, and (4) a multi-layer security architecture with WAF, server-side API isolation, and role-based access control. The system achieves 99.9% uptime through API key rotation and dual-model failover, with response times under 5 seconds. Built with React Native (Expo SDK 52) and Next.js 15, supporting SDG 3, SDG 9, and SDG 10.

Plant diseases represent a major threat to agricultural productivity and food security, especially in developing regions where access to agricultural experts is limited. This project presents LeafGuard AI, an intelligent deep learning-based system for real-time plant disease detection and bilingual treatment recommendation. The proposed system combines image classification using a custom-trained ResNet50 Convolutional Neural Network (CNN), cloud-based inference using FastAPI, and dynamic treatment generation using the Google Gemini Large Language Model (LLM). The model was trained and evaluated using the CCMT Dataset-Augmented, which includes images from four crop species: cashew, cassava, maize, and tomato. After collecting the four crop folders and merging duplicated class labels, the final dataset contained 105,252 images distributed across 18 disease, pest, and healthy classes. Images were resized to 224 × 224 pixels and augmented using random flipping, rotation, zooming, and translation to improve generalization. The ResNet50 model was used through transfer learning with a custom classification head consisting of GlobalAveragePooling2D, Dropout, and a Softmax output layer. After disease detection, the system invokes the Google Gemini API to generate structured treatment recommendations in both Arabic and English. These recommendations are cached locally using AsyncStorage to reduce repeated API calls and improve response time. The proposed system provides a practical end-to-end solution for precision agriculture by integrating plant disease classification, cloud deployment, multilingual treatment guidance, and cross-platform accessibility.

This project proposed an intelligent neuroprosthetic control system to liberate patients from sensory isolation and delayed reflex response associated with conventional artificial limbs. The proposed system consists of a multisensory electronic skin, an acquisition unit based on Arduino Mega, AI Cortex and Human Digital Twin for the real-time monitoring, prediction of danger, and reflex action to protect the person. In this case, the prototype developed can be a set of twenty distributed sensors covering both key sensory functions such as human skin. It typically has nine temperature sensors, five force sensors, and six moisture sensors. Sensor data also gets processed to produce pain index, injury risk, detected state and recommended motor response. Human Digital Twin serves to visualize and analyze the digital surrogate of a prosthetic limb; based on all physiological parameters, Predictive Reflex Intelligence layer can raise premonitory alerts and even execute automated decisions before we reach an injury risk. The integrated hardware sense, AI and digital twin visualization in a unified platform enhances safety, interpretability and user awareness contrary to the conventional microcontroller-only prosthetic control. It has high potential for smart prosthetics, rehabilitation robotics, medical assistive devices to enhance human-machine interaction systems.

يركز هذا المشروع على مشكلة أساسية تواجه التقنيات المساعدة، وهي صعوبة التواصل بين الصم والبكم نتيجةً لعدم توفر لغة الإشارة. لذا، تقدم هذه الورقة حلاً عملياً يتمثل في قفازات ذكية مزودة بمستشعرات مرنة تستخدم الذكاء الاصطناعي لتحويل إيماءات اليد إلى كلمات منطوقة. ويعتمد النظام على قاموس لغة الإشارة العربية (قاموس عربي معتمد) لضمان الدقة الدلالية. يتتبع النظام حركة الأصابع تلقائياً ويترجمها إلى إشارات رقمية تُنقل إلى وحدة التحكم. يستقبل القفاز الأول الصوت، بينما ينقل القفاز الآخر البيانات لاسلكياً. تُخزن بيانات المستشعرات على بطاقة ذاكرة SD مدمجة، مما يُتيح تدريب نموذج تصنيف باستخدام خوارزمية أقرب الجيران (KNN). من المتوقع أن تُسفر هذه الجهود عن تحسين مهارات التواصل وزيادة استقلالية المستخدمين.

Smart Lecture Follower is an intelligent real-time lecture synchronization system that automatically aligns spoken lecture content with written educational materials. The system integrates OpenAI Whisper ASR for speech transcription, intelligent document processing (PDF/DOCX/TXT), text normalization, and fuzzy matching algorithms to continuously identify and highlight the exact paragraph being discussed by the lecturer — with no manual intervention required. Key innovations include: (1) real-time speech-to-paragraph matching operating in 5-second audio intervals, (2) fuzzy matching engine handling pronunciation differences and speech variability without requiring exact text match, (3) offline deployment with no cloud infrastructure dependency, (4) multilingual recognition support including Arabic and English, and (5) a five-layer architecture covering Input, Speech Processing, Document Analysis, Matching, and Visualization. Performance evaluation shows 98.2% matching accuracy in quiet environments at normal speaking speeds, with sub-2-second response latency across all tested conditions. The system directly addresses SDG 4 (Quality Education) and SDG 10 (Reduced Inequalities) by providing critical accessibility support for non-native speakers, students with attention difficulties, and learners in multilingual environments. Built with Python, OpenAI Whisper, and a modular desktop architecture.