Volume 9, Special Issue - IJIRTM
(2025)
Impact Factor: 5.86 | Volume 9 | Special Issue
SAFE CITY VISION
👥 Sanjana, Shreya, Khushi, Ms.Aarti Chahal
📙 Abstract : This paper presents Safe City Vision, a machine-learning-based crime detection and analysis project intended to examine historical crime patterns, support prediction, and provide decision-oriented visual analytics. The project uses crime records containing geographic, temporal, and categorical information and applies preprocessing, feature extraction, classification, visualization, and interactive analysis. The source report describes K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest, time-series forecasting, and real-time analysis as components or proposed extensions. The documented project evaluation reports an overall accuracy of 82%, while the conclusion reports 85%; these values are retained as reported results rather than reconciled by assumption. The study also considers privacy, secure storage, fairness, and the responsible use of predictive crime analytics.
🔖 Keywords :️ crime detection, crime analysis, machine learning, KNN, SVM, predictive policing, data visualization, safe city.
Sentimentix: Mental Health Companion
👥 Mohan Jha, Devanshu Munjal, Vanshika Jain, Roshni
📙 Abstract : Mental-health support systems increasingly use natural language processing (NLP) and conversational artificial intelligence to provide accessible, low-barrier assistance. This paper presents Sentimentix, a software-based mental health companion designed to analyze user messages, infer sentiment or emotional state, maintain lightweight conversational context, and return supportive responses. The proposed framework combines text preprocessing, sentiment/emotion classification, contextual state management, response generation, and a safety-oriented risk and intent layer. Unlike systems that treat every message identically, Sentimentix adapts its response strategy according to the detected emotional polarity and conversational context. The design emphasizes privacy-aware local processing where feasible, transparent response rules, and escalation guidance for high-risk expressions. A prototype-oriented evaluation is described using functional scenarios rather than claiming clinical efficacy. The paper discusses the strengths and limitations of sentiment-aware conversational support and identifies future directions including multilingual modeling, transformer-based emotion recognition, longitudinal mood analytics, human-in-the-loop escalation, and clinically validated evaluation. Sentimentix is intended as a supportive technology and not as a replacement for qualified mental-health professionals.
🔖 Keywords :️ Mental health, conversational AI, sentiment analysis, natural language processing, emotion detection, chatbot, digital mental health, safety-aware AI.
Similarity-based Identity Verification System
👥 Harsh, Tejasva, Santosh Kailash Galve, Naman Narang, Hemlata
📙 Abstract : Identity verification is a fundamental task in security-sensitive applications such as authentication, access control, digital forensics, and identity management. This paper presents a similarity-based identity verification system that verifies a claimed identity by comparing extracted biometric or profile features with stored reference representations. The proposed workflow performs data acquisition, preprocessing, feature extraction, similarity computation, threshold-based decision making, and verification reporting. Distance and similarity measures such as Euclidean distance and cosine similarity are considered for measuring the closeness between an input representation and a stored identity profile. The system is designed to remain computationally practical while supporting variations in illumination, pose, expression, image quality, and other acquisition conditions. The paper describes the system architecture, feature-processing pipeline, verification algorithm, evaluation methodology, security considerations, and practical deployment issues. Rather than relying only on a raw similarity score, the framework separates feature extraction from decision policy so that thresholds can be calibrated for different operating environments. The proposed approach provides a transparent foundation for identity verification and can be extended with modern embedding models, liveness detection, multimodal authentication, and privacy-preserving storage.
🔖 Keywords :️ Identity verification, similarity measures, feature extraction, biometric authentication, Euclidean distance, cosine similarity, machine learning, security.
SMART PARKING SYSTEM USING YOLO
👥 Ujjwal Sharma, Gaurav, Nitin, Himanshu, Mr.Abhishek Gulia
📙 Abstract : The increasing number of vehicles in urban areas has intensified the problem of locating available parking spaces. This paper presents a vision-based smart parking system using the YOLO object detection algorithm to identify parking-space occupancy from camera imagery. The proposed workflow consists of image acquisition, preprocessing, YOLO-based object detection, post-processing and visualization of slot status. The project evaluates the system using precision, recall, F1 score, IoU and mean Average Precision (mAP). Reported experiments achieved precision of 0.926 and recall of 0.92, with average parking-slot accuracy of 92.6% when motion detection was enabled and 92% when motion detection was disabled. The results indicate that YOLO provides a fast and practical approach for real-time parking occupancy detection.
🔖 Keywords :️ Smart parking, YOLO, object detection, parking occupancy, computer vision, deep learning, real-time detection.
AI Based Disease Prediction and Health Consultancy
👥 Vijul, Aryan, Parveen, Ms.Shreya
📙 Abstract : This paper presents an AI-based disease prediction and health consultancy system designed to provide first-level, data-driven healthcare support. The application accepts user-selected symptoms and can optionally analyze uploaded medical reports in PDF format. A Random Forest Classifier forms the core disease-prediction engine, while PyMuPDF and regular-expression-based natural language processing extract structured patient information such as age, gender, habits, symptoms, and chronic-risk indicators. The extracted information is used to personalize recommendations including doctor selection, dietary guidance, lifestyle suggestions, risk information, and downloadable summaries. The system is implemented as a web application using Python, Streamlit, Pandas, and scikit-learn. The project report records approximately 96.2% training accuracy, 93.8% test accuracy, 92.4% average precision, and 91.7% average recall for the evaluated model. A sample symptom case returned a 92% confidence score. The system is intended as a preliminary decision-support and awareness tool rather than a replacement for professional diagnosis. The work also identifies privacy, latency, report-format variability, deployment, and evaluation challenges and proposes future extensions such as multilingual support, chatbot assistance, wearable integration, and scalable cloud deployment.
🔖 Keywords :️ Artificial intelligence, disease prediction, Random Forest, medical report parsing, NLP, Streamlit, health consultancy, personalized healthcare.
MedicoBot: An AI-Powered Multimodal Virtual Assistant for Cross-Species Health Diagnosis and Mental Wellness Support
👥 Ishu Kadian, Ridham, Muskan, Srishti, Dr.Shally
📙 Abstract : Access to timely and affordable preliminary diagnosis remains a major challenge, particularly in remote and under-served regions, for humans, livestock, and crops alike. This paper presents MedicoBot, a multimodal, AI-driven virtual healthcare assistant that interprets image and voice inputs to identify skin-related and visually observable ailments across humans, animals, and plants. The system combines automatic speech recognition (Whisper Large V3 via Groq), a vision-enabled large language model (LLaMA 4 Scout / meta-llama vision) for joint image–text reasoning, and neural text-to-speech synthesis (Google TTS and ElevenLabs) within a lightweight Gradio interface to deliver preliminary diagnoses, probable causes, and care recommendations in both text and natural speech. In addition to physical-health diagnosis, MedicoBot offers an empathetic conversational mode for basic mental-health support. Experimental deployment across sample skin, plant-leaf, and mental-health interaction scenarios shows that the pipeline reliably fuses speech and visual cues to produce coherent, human-understandable guidance while explicitly directing users toward professional consultation for serious cases. The results demonstrate the feasibility of a low-cost, multilingual-ready, cross-domain diagnostic assistant suitable for deployment in low-bandwidth and low-literacy environments.
🔖 Keywords :️ Artificial intelligence, multimodal diagnosis, speech recognition, large language models, text-to-speech, telehealth, mental health chatbot, cross-domain diagnostics, Gradio, Whisper.
AI-Enabled Crop Disease Prediction and Management: A Deep Learning Framework for Sustainable Agriculture
👥 Minal, Harshita, Ms.Mitu Sehgal
📙 Abstract : Crop diseases can substantially reduce agricultural productivity and create economic losses, particularly when diagnosis is delayed or dependent on manual inspection. This paper presents an AIenabled crop disease prediction and management framework based on Convolutional Neural Networks (CNNs) for image-based disease classification. The reported system uses a Kaggle-sourced dataset of 70,295 JPG images representing 38 crop-disease/healthy classes. The processing pipeline includes data cleaning, resizing, normalization, augmentation, feature learning, model training, validation, and application deployment. Transfer-learning architectures such as VGG16 and ResNet are considered to improve generalization when labeled data are limited. The framework is extended with a weather-aware risk module using temperature and humidity data, a bilingual interface, disease-specific recommendations, pesticide information, and downloadable reports. Experimental results reported in the project show 91% overall accuracy, with macro and weighted precision, recall, and F1-score also reported as 0.91. The system therefore combines visual diagnosis with contextual decision support to provide a practical precision-agriculture tool. The paper also identifies field-data diversity, model generalization, and multimodal sensing as important directions for future work.
🔖 Keywords :️ Crop disease prediction, convolutional neural network, deep learning, transfer learning, VGG16, ResNet, precision agriculture, weather-based risk assessment, sustainable agriculture.
Auto Caption AI: A Transformer-Based Vision-Language System for Automated Image Captioning, Translation and Speech Synthesis
👥 Deepika, Kirti, Pushap, Kunal Tyagi, Ms.Anisha
📙 Abstract : Auto Caption AI is an intelligent image captioning system that generates meaningful, context-aware descriptions for images using advanced vision-language transformer models. As visual content continues to grow across digital platforms, there's an increasing demand for systems that can interpret images in natural language. Manual captioning is often time-consuming and inconsistent, making automation essential. At the core of this project is the Salesforce BLIP model, a transformer-based architecture that combines a visual encoder with a text decoder to extract semantic information from images and produce fluent, human-like captions. The model is implemented using the Hugging Face Transformers library and runs on PyTorch, ensuring high performance across various image types. The user interface is built with Streamlit, allowing users to upload images and instantly receive captions through a web app. The backend runs on Google Colab, leveraging GPU acceleration for real-time inference, and public access is provided via ngrok. By integrating powerful transformer-based vision-language models with intuitive deployment tools, Auto Caption AI provides a seamless and practical solution for automatic image captioning, proving especially valuable for individuals with visual impairments who rely on textual cues for image interpretation.
🔖 Keywords :️ Image Captioning; BLIP; Vision-Language Transformer; Streamlit; MarianMT; Text-to-Speech; Accessibility; Deep Learning.
Cognitive Alert System: Real-Time Detection of Driver Drowsiness and Fatigue Using Convolutional Neural Networks and Facial Landmark Analysis
👥 Aman, Nancy, Vishal, Mr.Abhishek Gulia
📙 Abstract : The Cognitive Alert System is an intelligent real-time framework designed to monitor, analyze, and mitigate driver fatigue and drowsiness, addressing a primary contributing factor to catastrophic traffic accidents worldwide. By continuously tracking critical behavioral and physiological indicators— such as eye blink rate, eye closure duration (Eye Aspect Ratio - EAR), yawning frequency (Mouth Aspect Ratio - MAR), and head pose orientation—the proposed system issues proactive, multi-modal alerts before cognitive lapse results in operational failure. Utilizing open-source computer vision libraries (OpenCV, Dlib) for precise 68-point facial landmark localization and a TensorFlow-backed Convolutional Neural Network (CNN) for state classification, the system achieves a robust balance between latency and accuracy. Experimental evaluations across 50 diverse test subjects demonstrate an overall detection accuracy of 93.0%, a precision of 91.0%, and a real-time response latency of under 1 second. Furthermore, a two-layered classification architecture successfully resolves ambiguities between transitional states like mild fatigue and alertness. This work offers a scalable, non-invasive, and costeffective solution deployable across personal vehicles, public transport, and heavy commercial fleets.
🔖 Keywords :️ Driver Drowsiness Detection, Convolutional Neural Networks (CNN), Eye Aspect Ratio (EAR), Facial Landmarks, Real-Time Monitoring, Computer Vision, Intelligent Transportation Systems.
Deep Learning Based Classification of Fruits and Vegetables
👥 Harshit Dwivedi, Pranav Thakral, Khushi Panwar, Ms.Shreya
📙 Abstract : Automated classification of fruits and vegetables is an important computer-vision problem with applications in intelligent retail, automated sorting, inventory management, post-harvest handling, dietary assistance, and agricultural robotics. This paper presents a deep learning based framework for multi-class fruit and vegetable classification using Convolutional Neural Networks (CNNs) and transfer learning. The framework covers dataset preparation, image resizing and normalization, augmentation, CNN feature extraction, pretrained backbones, classification, evaluation, explainability, and deployment. Fruits-360 is used as the principal benchmark because its different releases provide large numbers of fruit, vegetable, nut, and seed categories; the current original-size repository reports 102,551 images across 145 classes, while earlier studies used approximately 90,000 images across 131 or 141 classes [1], [2]. Recent studies report strong benchmark performance, including 98.1% for an attention-based CNN on a 141-class setting and 97.15% for a multi-fused CNN on a 131-class setting [6], [7]. Rather than presenting these published values as new experimental results, this paper analyzes them comparatively and proposes a reproducible workflow for building and evaluating a fruit-and-vegetable classifier. The discussion emphasizes the distinction between controlled benchmark performance and real-world robustness, where illumination, occlusion, background variation, similar varieties, and domain shift remain important challenges.
🔖 Keywords :️ Fruit classification, vegetable classification, deep learning, convolutional neural network, CNN, transfer learning, Fruits-360, ResNet-50, VGG16, MobileNetV2, image recognition.
Deep Learning-Driven eKYC: An AI-Based Framework for Automated Identity Verification
👥 Isha Khurana, Anurag Saika, Nikhil Bhola, Shivam, Ms.Roshni Jha
📙 Abstract : Electronic Know Your Customer (eKYC) systems are increasingly used by banks, fintech platforms, and government services to verify user identity remotely. Traditional manual KYC verification is time-consuming, error-prone, and difficult to scale, while early rule-based digital KYC systems struggle with variations in document quality, lighting conditions, and spoofing attempts. This paper proposes a deep learning-driven eKYC framework that automates identity verification through document capture, optical character recognition (OCR)-based data extraction, convolutional neural network (CNN)-based face embedding, and liveness detection. The system compares a live facial capture against the photograph extracted from a government-issued identity document and computes a similarity score using deep facial embeddings, which is then evaluated against a decision threshold to accept or reject the verification request. The proposed architecture integrates Python, OpenCV, deep learning frameworks such as PyTorch and TensorFlow, OCR engines, and web technologies including FastAPI and React to deliver an end-to-end verification pipeline. The framework was evaluated conceptually across multiple identity-document types and face-matching scenarios, and its modular design allows individual components such as document extraction or liveness detection to be replaced or upgraded independently. The paper also discusses practical challenges including spoofing attacks, poor document image quality, demographic bias in face recognition, data privacy, and regulatory compliance. The proposed framework demonstrates how deep learning can enable a faster, more accurate, and more accessible identity verification process compared to manual and rule-based approaches.
🔖 Keywords :️ KYC, deep learning, face recognition, convolutional neural network, optical character recognition, liveness detection, identity verification, computer vision.
Driving Test Result Prediction System
👥 Rahul Hans, Deepanshi, Dr.Anju Saini
📙 Abstract : Driving schools need practical ways to assess whether learners are ready for a formal driving test. Manual assessment can be subjective and may not consistently combine demographic, behavioral, and performance-related evidence. This paper presents a Driving Test Result Prediction System that uses machine learning to estimate a learner’s likelihood of passing a driving test. The proposed system considers age, prior driving experience, first vehicle used for practice, practice hours, formal driving lessons, written test score, and self-reported confidence. The implementation integrates MySQL for data storage, preprocessing for feature preparation, a machine-learning model for prediction, FastAPI for backend services, Streamlit for the interactive interface, and Power BI for visualization. The workflow is intended to support early identification of learners who may need additional practice or targeted instruction. Because the supplied project synopsis does not report a completed dataset, a finalized learning algorithm, or measured test accuracy, this paper describes the proposed system and methodology without inventing experimental performance values.
🔖 Keywords :️ Driving test prediction, machine learning, driving school, FastAPI, MySQL, Streamlit, Power BI, predictive analytics
EDUDRAW: Problem Solver Using AI
👥 Ayush Thakur, Param, Rajat Pahal, Revant Rana, Harkesh Kumar
📙 Abstract : EDUDRAW is an AI-powered educational problem solver designed to provide an interactive and natural method for solving mathematical problems. The system combines computer vision, handgesture recognition, virtual drawing, and generative artificial intelligence to allow users to enter mathematical expressions through hand gestures instead of conventional keyboards, mice, or touchscreens. A webcam captures the user's hand movements, while MediaPipe detects hand landmarks and identifies gesture-based operations. The fingertip movement is converted into virtual ink on a digital canvas using computer vision and numerical processing techniques. Once a mathematical expression is drawn, the processed image is provided to a generative AI model for interpretation, calculation, and stepby- step explanation. The project integrates technologies including Python, OpenCV, MediaPipe, Google Gemini AI, NumPy, FastAPI, LangChain, Torch, React, TypeScript, and Vite. The system was evaluated using arithmetic expressions, algebraic and calculus expressions, and diagram-based geometry problems. The reported tests successfully recognized mathematical components and generated solutions for the tested cases. The project also identifies practical challenges involving gesture variability, lighting, camera quality, complex mathematical symbols, network dependency, privacy, user fatigue, and device scalability. EDUDRAW demonstrates how artificial intelligence and gesture-based human-computer interaction can be combined to create an accessible and interactive educational assistant.
🔖 Keywords :️ Artificial intelligence, computer vision, gesture recognition, mathematical problem solving, MediaPipe, Gemini AI, human-computer interaction, educational technology.
Enhancing Healthcare with Medicine Recommendations Using Machine Learning and Personalized Clinical Data
👥 Ravi Kumar, Akshit, Teesha, Yash, Ms.Arpana Dureja
📙 Abstract : The rapid growth of electronic health records and clinical data has created unprecedented opportunities for machine learning (ML) to support personalized medicine recommendation. However, medication selection remains a complex clinical challenge requiring consideration of patient symptoms, diagnosis, medical history, demographics, allergies, drug-drug interactions, and safety factors. This paper presents a comprehensive ML-based medicine recommendation framework that integrates patient clinical characteristics with drug safety constraints. The proposed system employs multiple ML algorithms including logistic regression, random forest, gradient boosting, and deep neural networks to predict appropriate medications while incorporating a drug-drug interaction (DDI) filtering layer. Using a dataset of 50,000 patient records from synthetic EHR systems, the framework achieves 91.3% recommendation accuracy, 87.8% precision, and reduces potentially unsafe medication combinations by 94.2%. The system incorporates interpretability mechanisms to provide clinicians with justifiable recommendations. Results demonstrate that ML-based approaches can effectively support clinical decision-making while prioritizing patient safety. This work contributes to clinical decision support systems by providing a systematic framework for personalized medicine recommendation with explicit safety considerations.
🔖 Keywords :️ Machine Learning, Medicine Recommendation, Drug Recommendation, Electronic Health Records, Personalized Healthcare, Clinical Decision Support, Drug-Drug Interaction, Deep Learning, Healthcare Analytics, Patient Safety.
Market Mind Analyzer: An AI-Driven Framework for Financial Sentiment, Market Trend Analysis, and Predictive Decision Support
👥 Dikshita, Vijay Gautam, Dr.B K Verma
📙 Abstract : Financial markets are influenced by historical price behavior, technical indicators, macroeconomic conditions, corporate events, financial news, and investor sentiment. The high volume and velocity of financial information make manual analysis increasingly difficult. This paper proposes Market Mind Analyzer (MMA), an artificial intelligence-driven decision-support framework that integrates financial market data with natural language processing (NLP), sentiment analysis, technical indicators, machine learning, and explainable visualization. The proposed framework collects financial news and market information, preprocesses textual and numerical data, extracts sentiment using a domain-adapted language model such as FinBERT, calculates technical indicators, and combines these heterogeneous features for market trend classification. The system produces a composite Market Mind Score representing the prevailing positive, neutral, or negative market state and generates an interpretable dashboard containing sentiment trends, technical signals, volatility information, and predicted market direction. Unlike systems based exclusively on historical prices or textual sentiment, MMA combines behavioral and quantitative signals in a unified architecture. The paper presents the system architecture, feature-engineering methodology, prediction strategy, evaluation metrics, and limitations. Numerical experimental results are intentionally not fabricated; the proposed evaluation protocol is provided for execution on a selected dataset.
🔖 Keywords :️ Financial Market Prediction, Sentiment Analysis, FinBERT, Natural Language Processing, Machine Learning, Technical Indicators, Market Intelligence, Explainable AI, Stock Market, Decision Support System.
MULTIPLE DISEASE PREDICTION USING MACHINE LEARNING
👥 Vidhi, Arpita, Ashika, Nancy, Ms.Ritu Rani
📙 Abstract : This paper presents a unified machine-learning-based framework for preliminary prediction of heart disease, diabetes, and facial skin diseases. The system uses supervised classifiers for structured clinical data and a convolutional neural network (CNN) for image-based skin-disease classification, followed by integration through a Streamlit web interface. The project report describes preprocessing, feature engineering, model selection, hyperparameter tuning, evaluation, and report generation as core stages of the workflow. The reported CNN experiment achieved 91.40% training accuracy, 63.78% validation accuracy, and 62.70% test accuracy after 10 epochs; the gap between training and validation/test performance indicates a need for stronger generalization. Numerical accuracy values for the heart- and diabetes-prediction models are not explicitly reported in the source project report and are therefore not fabricated here. The framework is intended as a decision-support and screening aid rather than a replacement for clinical diagnosis.
🔖 Keywords :️ Multiple disease prediction, machine learning, deep learning, CNN, Streamlit, healthcare AI.
NLP-BASED CHATBOT FOR FOOD WEBSITES
👥 Krish, Harshit Sukhija, Nitish, Vinay, Ms.Aarti Chahal
📙 Abstract : Food websites provide a large number of menu items, but users often have difficulty identifying suitable dishes because conventional interfaces depend on manual browsing, filtering, and keyword search. This paper presents an NLP-based conversational chatbot designed for food websites to transform natural-language requests into personalized food recommendations. The proposed framework combines intent detection, named-entity and attribute extraction, semantic matching, dialogue-state tracking, and a recommendation layer connected to structured menu data. User requests such as “I want a spicy vegetarian meal under ?300” are converted into constraints involving cuisine, dietary preference, price, taste, and dish type. A hybrid retrieval strategy combines lexical matching with sentence-level semantic similarity, while a dialogue manager asks clarification questions when required information is missing or ambiguous. The paper also defines an evaluation methodology based on intent accuracy, slot/attribute extraction F1-score, recommendation precision, response relevance, latency, and user satisfaction. The design emphasizes explainable recommendations and safeguards against hallucinated menu information by grounding responses in the website's menu database. The proposed system can serve as a practical architecture for intelligent food ordering, discovery, and customer-support applications.
🔖 Keywords :️ Natural language processing, chatbot, food recommendation, conversational AI, intent detection, semantic search, restaurant websites.
A Hybrid Voting-Based Machine Learning Framework for Real-Time Phishing Web Page Detection System
👥 Diksha, Chanchal, Utkarsh, Anshul, Ms.Neelam Goswami
📙 Abstract : Phishing remains one of the most persistent and damaging forms of cybercrime, exploiting deceptive emails and fraudulent websites that closely mimic legitimate platforms to steal sensitive user information. The rapid growth of online trading and internet-based services has widened the pool of potential victims, while traditional detection methods such as blacklisting and heuristic-based approaches have proven increasingly inadequate against evolving attack techniques. Given the absence of a fully reliable defense mechanism, machine learning has emerged as a promising direction for combating phishing threats. This work explores potentially using a URL-based dataset comprising over 11,000 legitimate and phishing website samples represented in vector form. After preprocessing, several algorithms were trained and compared, including Decision Tree, Logistic Regression, Random Forest, Naive Bayes, Gradient Boosting, K-Nearest Neighbors, and Support Vector Classifier, alongside a proposed hybrid “LSD” model that combines Logistic Regression, Support Vector Machine, and Decision Tree through soft and hard voting. Canopy-based feature selection, k-fold cross-validation, and Grid Search hyperparameter optimization were applied to refine the hybrid model. Performance was assessed using accuracy, precision, recall, F1-score, and specificity. The comparative results reported here are illustrative and intended to demonstrate the evaluation methodology pending completion of full experimental validation; they indicate that the proposed hybrid approach can surpass individual classifiers, offering a robust and computationally efficient solution suitable for real-time phishing URL detection.
🔖 Keywords :️ Phishing detection; machine learning; hybrid ensemble classifier; voting classifier; feature selection; URL analysis; cybersecurity.
Plant Species Identification Using Convolutional Neural Networks with Principal Component Analysis: An Interpretable Deep Learning Approach
👥 Bhavay Jain, Gourav Tyagi, Shivam Garg, Ms.Anisha
📙 Abstract : Accurate identification of plant species plays a critical role in various domains such as agriculture, botany, environmental conservation, and medicine. Traditionally, plant identification relies heavily on expert knowledge and manual analysis, which can be time-consuming and error-prone. This project presents an automated system for plant species identification using advanced image processing and machine learning techniques. The proposed system leverages a deep learning-based Convolutional Neural Network (CNN) architecture to extract meaningful features from plant images, primarily focusing on leaves, flowers, and other identifiable parts. These features are then used to classify the plant species with high accuracy. The system has been trained and tested on a diverse dataset containing images of various plant species, addressing challenges like intra-species variability, similar morphological traits among species, and varying lighting conditions. To enhance performance and usability, techniques such as data augmentation, transfer learning using pre-trained models like MobileNet and ResNet, and optimization algorithms have been employed, and an unsupervised Gaussian Process Latent Variable Model combined with Principal Component Analysis (PCA) is used to make the CNN's decisions more biologically interpretable. This project aims to provide an accessible tool for researchers, students, farmers, and nature enthusiasts, contributing to biodiversity awareness and sustainable ecological practices.
🔖 Keywords :️ Plant Species Identification; Convolutional Neural Network; Principal Component Analysis; Gaussian Process Latent Variable Model; Deep Learning; Interpretable AI.
Real Time Face Based Energy Monitoring System
👥 Deepanshi, Sakshi, Riya, Diksha, Harkesh
📙 Abstract : Unnecessary electricity consumption in classrooms, laboratories, offices, and homes can occur when appliances remain active after occupants leave. This paper presents a software-based realtime monitoring prototype that combines webcam-based face detection with simulated energy consumption to identify possible energy wastage. OpenCV’s Haar Cascade classifier detects frontal faces, NumPy generates simulated energy values between 200 W and 600 W, Pygame produces an audible warning, and Matplotlib displays the latest ten readings. An alert is generated when no face is detected and simulated energy exceeds 400 W. The system operates locally for a 40-second monitoring cycle and does not require physical energy sensors or external APIs. The project reports successful operation under suitable lighting, smooth graph updates, and reliable alert playback. Limitations include sensitivity to lighting and face orientation and the use of simulated rather than measured energy data. The prototype provides a low-cost foundation for future sensor-based and IoT-enabled energy management systems.
🔖 Keywords :️ Face Detection, Real-Time Monitoring, Energy Simulation, Sound Alert, Matplotlib, Smart Room Automation, OpenCV, Python, Presence Detection, Graph Visualization.