Project Reports

Permanent URI for this community

http://172.16.1.132:4000/handle/123456789/14042

Browse

Now showing 1 - 20 of 43

A Novel based approach of home automation system for constrained edge IOT devices
(NHCE, 2024) Priyanshu Roy 1NH20CE033; Vinay Krishnan 1NH20AI138; Tarun S 1NH20AI108; Kavya Sivakumar 1NH20AI047
In the context of contemporary living, existing smart home systems exhibit limitations in offering a comprehensive, tailored experience. Our forthcoming home automation project, leveraging BluFi Mesh and Matter OS integration, seeks to address these shortcomings. Focused on rectifying the deficiencies in current systems, our initiative aims to create an intelligent, adaptable ecosystem that emphasizes user preferences, robust security, seamless entertainment integration, energy efficiency, and real-time responsiveness. Our novel approach merges BluFi Mesh networking with the advanced Matter OS platform, establishing a robust infrastructure for a home server that operates discrectly, resourcefully, and securely, granting occupants control while safeguarding their privacy. Distinguishing our system integrates Al-powered features recognizing occupants, their preferences, and prioritizing safety. When entering the house, the occupant is instantly identified, customizing ambiance based on preferences, and ensuring security through face recognition and radar-based classifications. Beyond this, our system also aims to integrate popular entertainment services offering effortless access and control via simple voice commands or device taps. Additionally, our project emphasizes on power usage visualization that empowers users to monitor and optimize energy consumption, contributing to environmental sustainability while reducing utility costs. Our project aims for low-latency operations, guaranteeing instant, seamless interactions.
Advanced Traffic Monitoring & Speed Regulation using YOLOVII
(New Horizon College of Engineering, 2025) Alwin Jose-1NH22CE002, Prathyek K- 1NH22CE018
This work proposes an end-to-end real-time traffic monitoring and over-speed detection system with deep learning and computer vision to improve road safety and automate the process of traffic enforcement. Traditional speed detection methods—such as radar guns, fixed speed cameras, and manual surveillance—suffer from scalability, high costs, and limited coverage. In response, the proposed system will present a fully automated, vision based framework that integrates high-accuracy vehicle detection, speed estimation, license plate recognition, and automated alert delivery using modern AI techniques. At its core, the system employs the YOLOv11 object detection architecture, Paddle OCR for optical character recognition, and edge-compatible processing workflows for real-time performance. The system's architecture design surrounds four major modules: real-time vehicle detection, speed estimation through pixel displacement, license plate recognition, and alert notification. Initial steps incorporate an Ultralytics version of the YOLOv11 model-an effective, deep-learning-based convolutional neural network-to identify the presence of vehicles in video streams, either live or recorded. With high confidence, the model distinguishes between objects of interest such as cars, trucks, motorcycles, and buses in each frame. It detects at a rate of 30 frames per second, ensuring applicability for deployment in real-time applications.
AI Based disease detection system
(New Horizon College of Engineering, 2026) Ashvin Sam Suraj M 1NH22CE065 K Aman Kumar 1NH22CE016 Sharanya Kattishetti 1NH22EC149 Sanjana Shaji Nair 1NH22EC146
Timely and accurate identification of plant diseases is critical for improving crop yields and ensuring food security. Manual inspection methods are often laborious, time-consuming, and may require specialized agronomic expertise that is not always accessible, particularly to small-scale farmers. This project, titled "AI-BEDS (AI Based Early Diagnosis System)," addresses these limitations by developing an innovative software-based system. The system utilizes deep learning and computer vision techniques to automatically detect and identify various plant leaf diseases directly from digital images. The core of AI-BEDS is a robust Convolutional Neural Network (CNN) model, trained on a comprehensive dataset of plant leaf images. This model is integrated into a user-friendly mobile or web application. This application allows farmers and agricultural workers to easily capture new images of plant leaves using their mobile devices or upload existing images. Upon submission, the system processes the image through the AI model and provides prompt diagnostic feedback. This feedback includes the identified plant species, the name of the detected disease (if any), and a confidence score indicating the reliability of the prediction. A significant aspect of the AI-BEDS project is its design for versatility and accessibility. While the primary interface is through mobile and desktop platforms to ensure broad reach, the system architecture also supports optional deployment on a Raspberry Pi. This feature enables the system to function as an offline, field-deployable diagnostic tool, extending its utility to rural or remote areas where internet connectivity may be unreliable or unavailable. The methodology for this project encompasses several key stages: the collection and meticulous preprocessing of a large and diverse dataset of plant leaf images (e.g., from public repositories like PlantVillage, potentially augmented with custom data); the training and optimization of a suitable deep learning classification model (such as MobileNetV2 or EfficientNet, chosen for their balance of accuracy and efficiency) using frameworks like TensorFlow/Keras; the conversion of the trained model to a lightweight format (e.g., TensorFlow Lite) for efficient deployment on edge devices, including mobile phones and the Raspberry Pi; and the development of an intuitive and responsive user interface for both the mobile/web application and the optional Raspberry Pi setup. Furthermore, the system is designed to include functionality for logging diagnostic results, which can help track disease prevalence over time, and for suggesting general, evidence-based treatment guidelines to assist farmers in making informed decisions. The anticipated outcomes of the AI-BEDS project are a highly accurate, fast, and reliable mobile or web application capable of identifying common plant diseases from leaf images. The optional offline capability provided by Raspberry Pi deployment will further II enhance its practical value in diverse agricultural settings. From a societal perspective, AI-BEDS aims to empower farmers with an accessible, low-cost tool
AI-Powered Real-Time Safety Monitoring and Hazard Detection System for Industrial Workspace
(New Horizon College of Engineering, 2026) Madhuri Chillarge 1NH22CE025 Syeda Suhasana 1NH22CE059 K J Mohit 1NH22CE017 D Navyashri 1NH22CE009
An innovative web-based platform designed to increase safety in industrial and construction settings is the GuardianEye Safety Al Monitoring System. It operates directly within a standard web browser and employs real-time Al recognition, monitoring, and immediate waming to spot risks and make sure employees abide by safety regulations, it becomes simple to use in both small and large businesses because it doesn't require complicated installation or hefty gear The three fundamental tenets of the system are modularity, accessibility, and efficiency The major control point of its hub-and-spoke architecture is its main dashboard (Index.html). Three distinct Al-powered modules are available to users via this dashboard: Vehicle Speed Monitoring, Fall Detection, and PPE Compliance Monitoring. Since each module is built as a stand-alone web application, changes or enhancements can be made without impacting the system as a whole. GuardianEye is scalable, adaptable, and appropriate for long-term deployment thanks to its modular design. The Göögle Gemini Vision API is used by the PPE Compliance Module to verify if employees are donning safety vests and helmets. When infractions happen, the technology automatically monitors live footage and generates alerts TensorFlow.js MoveNet is used by the Fall Detection Module to monitor human body motions and apot unusual postures or unexpected falls In order to minimise accidents caused by overspeed, the Speed Detection Module uses TensorFlow.js COCO-SSD to identify vehicles and determine their speed. Every module uses the same pipeline, which includes utilising the MediaDevices API to record live video, running Al models locally or via cloud services, analysing the findings, and using the Canvas and Web auditory APIs to create real-time warnings with visual and auditory signals. For convenience, localStorage is used to save user settings such as theme mode and sensitivity levels. All things considered, GuardianiEye is a strong and clever safety platform. It enhances situational awareness, lowers human error, and fortifies industry-wide workplace safety regulations by fusing cutting-edge Al technologies
BinancePy: Python - Powered Crypto Trading
(NHCE, 2024) Aravind Aripaka 1NH20CE004; Tundiya Krish J 1NH20CE055; Harshchand Aravind Aripaka 1NH20CE004
Fingerprint images in crime scene are important clues to solve serial cases. Crime scene fingerprint identification system using deep machine learning with Convolutional Neural Network (CNN). Images are acquired from crime scene using methods ranging from precision photography to complex physical and chemical processing techniques and saved as the database. The images collected from the crime scene are usually incomplete and hence difficult to categorize. The fingerprint recognition system is divided into three stages that are fingerprint image pre-processing, feature extraction and matching. Suitable enhancement methods are required for pre-processing the fingerprint images. The output of this stage will be passed to feature extraction stage which is extract the minutiae point (ridge ending, Bifurcation) from thinning fingerprint image, then the false minutiae removal is applied to extract real minutiae. The features of pre- processed data are fed into the CNN as input to train and test the network.
BinancePy: Python - Powered Crypto Trading
(NHCE, 2024) Aravind Aripaka 1NH20CE004; Tundiya Krish j 1NH20CE055; Harshchand V 1NH20CE058; Vaibhav C 1NH20CE059
Cryptocurrency mining has emerged as a popular method for generating digital currencies, such as Bitcoin and Ethereum, through the validation of transactions and addition of new blocks to the blockchain. This paper presents the development and implementation of an automated cryptocurrency mining bot using Python, focusing on optimizing mining efficiency and ensuring operational security. The proposed bot leverages the computational power of high-performance hardware to solve complex cryptographic puzzles, thereby validating transactions and earning cryptocurrency rewards. The mining bot is designed to operate autonomously, managing the mining process with minimal human intervention. It integrates various functionalities, including real-time monitoring of mining metrics, dynamic adjustment of mining strategies based on network difficulty and computational load, and secure handling of mined assets. Cryptocurrency mining has emerged as a popular method for generating digital currencies, such as Bitcoin and Ethereum, through the validation of transactions and addition of new blocks to the blockchain. This paper presents the development and implementation of an automated cryptocurrency mining bot using Python, focusing on optimizing mining efficiency and ensuring operational security. The proposed bot leverages the computational power of high-performance hardware to solve complex cryptographic puzzles, thereby validating transactions and earning cryptocurrency rewards. The mining bot is designed to operate autonomously, managing the mining process with minimal human intervention. It integrates various functionalities, including real-time monitoring of mining metrics, dynamic adjustment of mining strategies based on network difficulty and computational load, and secure handling of mined assets.
Blockchain based voting system
(New Horizon College of Engineering, 2025) Goutham KrishnaA -1NH22CE012, Rohith S G-1NH22CE042, Sanjan Manjunath-1NH22CE045, Baladitya Godavarthi-1NH22CE005
This project presents a secure and transparent Blockchain-Based Voting System integrated with Geo-Fencing, designed to address the critical challenges of modern electoral processes. Traditional paper-based and electronic voting systems suffer from problems such as lack of transparency, centralized control, vote tampering, limited auditability, and vulnerabilities in both physical and digital infrastructures. Internet-based voting systems, while convenient, face additional threats including cyberattacks, DDoS disruptions, weak authentication, and absence of geographical verification. These constraints highlight the need for a technologically advanced, tamper-proof, and jurisdiction-compliant voting mechanism. The proposed system leverages blockchain technology to ensure immutability, decentralization, transparency, and end-to-end verifiability of votes. Every vote is recorded as a cryptographically secured transaction on a decentralized ledger, eliminating the risk of unauthorized modification. The use of smart contracts ensures automatic enforcement of election rules, prevents duplicate voting, validates voter eligibility, and supports real-time tallying without manual intervention. By eliminating the need for a central authority, the system enhances trust and provides a fully auditable election trail. To strengthen jurisdictional compliance, a Geo-Fencing module is incorporated that validates whether the voter is physically within the authorized constituency or polling boundary before vote submission. The system captures real-time GPS coordinates with ±10-meter accuracy, verifies location using the point-in-polygon algorithm, and generates a signed location token as proof. This ensures that all votes originate from approved locations, preventing cross-border or unauthorized remote voting, and aligning with legal election requirements.
Breast Cancer Classification aided by Homomorphic Encryption
(New Horizon College of Engineering, 2026) M. Preeti 1NH22CE023 ShakthiPriya 1NH22CE050 Gauri B Nair 1NH22CE011 Ishmita Menon 1NH22CE014
wealthcare has sundergime a major change in technology awww the last ten years, with the mam factors being rapid aduancemem Armficial Intelligence (Al), Machine Learning (MU, data analytics, and secure computational methods. Theur new technologies have changed the face of medicine by introduung automated analysis, predictive modelling, and des support systems that nela comcians in thagnoung donirses with improved apuracy and efficiency M impact related issues were numimoun, the one on breast cancer detection was very entical because of the disease's orevalence, mortality rate, and importance of the early stage diagnosis, Breast cancer is a complicated medical condition that results from the abromal development of cells in the breast area, farly detection & a great advantage in treatment success, but accurate diagnosis a must which can only be achieved bo analysing varied types of clinical data such as manunography images, biopsy reports, Tumor measurements, hormonal recesitor status, and patient medical histories The datasets aned in the case are extremely sensitive since they contain personal denshable health informator that a protested under very stric globul privacy regulations. The digitization of medical records along with the demand for collaborative diagnonte soort has brought to light the issues related to the secure handling and processing of sach data Medical datasets used for first canem lemon typically came of Tumur characteristics jeg margin terture Tenging fratures (ng, density masi shape) Demapphic dietalli leg age, family history) Pathology.
Climate Adaptive IoT Drip Irrigation System
(New Horizon College of Engineering, 2026) Sankalpa Kashyap 1NH22CE048 Mrinank Kumar Saini 1NH22CE030 Jaydeep Mukharjee 1NH22EC066 Kiran Ghosh 1NH22EC076
Agriculture today faces growing challenges due to erratic weather conditions, water scarcity, and inefficient irrigation methods. Most conventional systems follow fixed irrigation schedules or manual operation, which often results in over-irrigation, increased labour, and suboptimal crop performance. To overcome these limitations, our project—Climate-Adaptive IoT Drip Irrigation System—aims to design a real-time, intelligent irrigation solution that adapts based on soil conditions and weather forecasts. The system integrates soil moisture sensors to continuously monitor field conditions. A microcontroller-based control unit processes this data and makes irrigation decisions dynamically. Instead of using ESP32, an alternative microcontroller suited for stable connectivity and modular expansion has been implemented to support multiple sensor inputs and control components. To avoid unnecessary irrigation, the system fetches real-time weather forecasts from the OpenWeatherMap API, delaying irrigation when rain is expected. For data communication, we use the MQTT (Message Queuing Telemetry Transport) protocol—a lightweight and efficient IoT protocol that allows real-time transmission of sensor data and control commands. The microcontroller publishes the data to a cloud server, where it is stored and displayed through a custom web dashboard. The dashboard provides users with access to live soil conditions, system status, and irrigation history, along with manual override features for better user control. This is an interdisciplinary project, with contributions from both Electronics and Communication Engineering (ECE) and Computer Engineering (CE) domains. The ECE team handled sensor calibration, hardware interfacing, and relay-based control of solenoid valves for drip irrigation. Meanwhile, the CE team focused on backend development, API integration, cloud database setup, and dashboard design, ensuring smooth real-time monitoring and user interaction.
CNN based implementation of crowd detection using raspberry pi
(NHCE, 2024) Dhimant R S 1NH20CE011; Ganya P R 1NH20CE013; Grishma S G 1NH20CE014
There is a lot of population numbers in the world and India is one among them. So, it is very common to expect crowded events with many number of people participating in the events like cart festivals-, strikes-, protests and any other sort of event where people gathering is too common in these days but these gatherings will be confronted to multiple security issues. To get free from such issues, security cases are majorly involved to handle the crowd and provide security to the people surrounded in the gatherings. But at last the security provided to people is limited with the technology, the security forces can easily become tired and remain unsuccessful in maintaining the crowd. Due to the negligence and poor managed system this might be problematic issue to the people like people get injured and it might also lead to loss of lives accordingly crowd also increases. So that’s the reason we are here with the solution for crowd detection by using Raspberry Pi Model 4B and by implementing the concept of deep learning and machine learning on AI, by training a model based on it. And at last this model will be deployed on our Raspberry Pi board. This will help us to identify the crowd on a real time scenario. The crowd can be categorized under one person, two people, three people, four people, crowd or protest crowd. The crowd can then be detected by CCTV or by drones for betterment view of the crowd. To achieve this the process is divided into three parts, firstly we will identify pedestrian using the collected data set. Secondly, we will try the same thing on people. Thirdly, we will try this on real time crowd detection.
Deep Reinforcement Learning for Image Hashing
(NHCE, 2024) Athira Musaliyath Dinesh 1NH20CE005; Dhanush Y J 1NH20CE010; Anusha Sai 1NH20CE019
The rapid increase in digital images necessitates efficient retrieval and storage methods, where traditional hashing techniques struggle with scalability and accuracy. This project introduces Deep Reinforcement Learning (DRL) for Image Hashing (DRLIH), leveraging DRL's sequential decision-making to enhance hashing function learning and retrieval performance. Traditional hashing methods like Locality-Sensitive Hashing (LSH) and spectral hashing treat hashing functions independently, ignoring correlations between them and leading to suboptimal accuracy. Deep learning methods using convolutional neural networks (CNNs) for hash code generation also lack mechanisms to refine hashing functions sequentially based on previous errors. DRLIH models the hashing process as a Markov Decision Process (MDP), incorporating state representation from recurrent neural networks (RNNs) and CNNs, an action space of possible hashing functions, and a reward function providing feedback on hash code accuracy. The DRL agent, which features an RNN-based architecture with Long Short-Term Memory (LSTM) units, is trained using the policy gradient method to maximize cumulative rewards through iterative interactions. By iteratively correcting errors and leveraging historical context, DRLIH offers a robust, scalable solution for large-scale image retrieval, setting a new standard in the field. Future work will explore advanced RL techniques to further enhance performance.
Design and development of Multi-Purpose Quad copter
(NHCE, 2024) Sanjay Saheel 1NH20CE041; Sudar P 1NH20CE050; Ravi B 1NH21CE401
The Earth offers a supply of clean air and drinkable water, with a wealth of undiscovered components hidden beneath the surface. One of the main issues facing emerging nations like India is the extreme pollution of their waterways. There is pollution in the air. Chemicals are often used as raw materials in industrial operations. A few of them cause pollution and are really dangerous. As a result, there is an increase in the release of dangerous substances into the land, water, and air. Two crucial natural resources—air and water—determine future wealth and stability. at order to assist the authorities at crisis management centers in identifying and foreseeing water and air pollution, we have built the Water and Air Monitoring System. It takes work to keep farmland productive and healthy. It necessitates an understanding of various training and care techniques that have been refined over countless years. In many areas, synthetic fertilizers have supplanted natural fertilizers in recent years, giving farmers more options when it comes to sophisticated land management that maximizes yields. Achieving the soil's natural balance is still crucial for sustainable production, though, and farmers need to be proactive in keeping an eye on the composition of their soil. Technological developments in agriculture assist farmers in obtaining precise information about their farms to enable them to make better decisions. Drone-enabled multispectral remote sensing for soil mapping is one instance of this kind of technology; it can assess acres of farmland utilizing sophisticated. Sampling soil is a crucial method for obtaining data in order to make informed judgments about field fertilizer. Soil sampling may be required every 10 hectares, depending on country regulations. However, this is not enough for precision farming. To collect soil samples from consistent locations while minimizing the number of samples required, the field is divided into smaller areas called management zones.. In this paper, a unique method for autonomously locating soil sample locations is presented.
Design and development of Refreshable Braille Display
(New Horizon College of Engineering, 2026) Ayushi Kumari 1NH22CE004 Saloni Khemka 1NH22CE044 Harshit Rana 1NH22ME025 Mukesh Kumar Pandey 1NH22ME039
Although the world has made remarkable progress in digital accessibility, the visually impaired population continues to face major educational and informational barriers due to the lack of affordable reading devices. According to global studies, less than 10 percent of visually impaired people are Braille literate, and one of the main reasons is the high cost and limited availability of Braille materials and displays. In countries such as India, where the visually impaired population exceeds eight million, only a fraction of educational institutions can afford to install Braille reading systems. Current commercial refreshable Braille displays rely heavily on complex and costly piezoelectric actuators. These actuators provide accurate and reliable tactile movement, but each dot requires an individual piezo crystal, driver circuit, and mechanical support. The overall cost increases exponentially with the number of dots or Braille cells. Additionally, the requirement for specialized manufacturing and imported materials further increases cost and reduces availability. Existing devices also consume relatively high power and are not easily repairable. Many are designed for fixed setups rather than portable use, limiting flexibility for students and professionals who need lightweight, mobile solutions. Furthermore, most systems are proprietary, meaning they cannot be modified or improved easily by researchers or users. This restricts open innovation in the field of assistive technology.
Detecting Dangerous Pests in Farms Using Deep Learning
(NHCE, 2024) E Vanija 1NH20CE012; Yuvaraj K M 1NH20IS170; Sivani Reddy K 1NH20CE049; Sudharshan R 1NH20IS188
This work offers a comprehensive approach that makes use of the YOLO v8 algorithm to effectively detect and manage agricultural pests. The system, embedded in a Streamlit application, allows farmers to upload crop images for real-time pest identification. Lever-aging the YOLO v8 model's accuracy, the system recommends organic pesticides tailored to the identified pests, promoting sustainable agriculture practices. To enhance user ac-cessibility, language translation and audio features are integrated, accommodating di-verse linguistic backgrounds and addressing literacy challenges. Results demonstrate the system's efficacy in pest detection and its potential to empower farmers with advanced technology for informed decision-making in crop protection. Agriculture plays a pivotal role in sustaining human life, but its productivity is consistently threatened by the adverse impact of pests on crops. As the global population grows, ensuring food security becomes an increasingly challenging task. Traditional pest management approaches often fall short in terms of efficiency, precision, and sustainability. This paper addresses this critical issue by proposing an innovative solution that leverages the YOLO v8 algorithm for robust pest detection and an Integrated Pest Management (IPM) system for sustainable agricultural practices. Pest infestations pose a significant threat to global food production, resulting in substantial economic losses and environmental consequences. Conventional pest management methods, often reliant on chemical interventions, raise concerns about environmental sustainability and the long-term health of agricultural ecosystems. Additionally, there exists a need for rapid and accurate pest detection methods to enable timely and targeted responses, minimizing the impact on crop yields.
Detection of Autism Spectrum Disorder Using Machine Learning
(NHCE, 2024) Darshna V 1NH20CE009; M kumar Venkat 1NH20ME075; Omini rao N 1NH20CE029; Prajwal 1NH20ME085
The exponential disorder known as autism spectrum disease has a profound impact on how individuals behave and engage in their communities. Children diagnosed with autism spectrum disorder may find it difficult to interact socially and learn new words. Unusual behavior that parents observe in their autistic children includes poor motor skills and repetitive movements of their hands and head, as well as violent head and body jerks. While there is no known cure for ASD, many children's lives can be greatly enhanced by early intervention. For this reason, ASD is called a lifelong illness. Depending on the illness's complexity, severity, and symptoms, multiple factors can contribute to ASD. Environment and genetics play important roles. Early autism prediction has become effortless because of artificial intelligence and machine learning (ML). Even though several studies have been carried out utilizing diverse methodologies, these inquiries have not produced any definitive results regarding the capacity to anticipate autism characteristics across different age groups. Although the model can be applied to both toddlers and adults, the primary focus of this project is the early detection of autism disorder in toddlers. ASD symptoms typically appear between the ages of 12 and 18 months, so when they are identified early on, a toddler's communication skills can be improved through therapy.
Detection of Fake Profile on Social Media
(NHCE, 2025) Melbin Biju 1NH20CE027; Sanjay G 1NH20CE040; Alistair Kissinger 1NH20CE067; Prince Enoch 1NH20CE068
In Python, fake profile detection entails detecting and flagging profiles on social media or other platforms that display questionable or fraudulent behavior. These personas may impersonate others, send spam, distribute misinformation, or participate in fraud. Machine learning methods, natural language processing (NLP) for text analysis, image recognition for profile photos, network analysis, and anomaly detection are techniques used to detect bogus profiles. Python offers a range of libraries and tools to implement fake profile detection, including: Scikit-learn: Implementing machine learning models like decision trees, random forests, or SVMs. Natural Language Toolkit (NLTK): Useful for text and sentiment analysis. TensorFlow or PyTorch: Ideal for deep learning-based approaches like neural networks. OpenCV: Useful for image processing and analysis. NetworkX: For analyzing the connections and networks between profiles. The process typically involves collecting data from profiles, extracting relevant features (text content, posting frequency, connection patterns), and training a model to distinguish between genuine and fake profiles.
Disease Prediction using Machine Learning
(NHCE, 2024) Kammara Trivikram 1NH20CE022; Manobhi Ram Reddy B 1NH20CE026; Shashank S 1NH20CE045; Sushmitha H 1NH20CE053
In contemporary healthcare systems, the timely and accurate diagnosis of diseases represents a critical pillar for effective treatment and the containment of healthcare costs. Addressing this imperative, our project endeavors to develop a sophisticated web application that harnesses the power of machine learning algorithms to anticipate diseases based on user-provided symptoms. By doing so, we aim to facilitate prompt medical intervention and improve healthcare outcomes. Central to our solution is the implementation of an ensemble model that combines multiple machine learning classifiers, including Logistic Regression, Random Forest, KNN, MLP, and SVM, with XGBoost serving as the meta-learner. This ensemble approach capitalizes on the diverse strengths of these algorithms to achieve an exceptional accuracy rate of 97%. Through meticulous testing and validation procedures, we have ensured the reliability and efficacy of our predictive model. The core functionality of our Flask-based web application revolves around providing users with a seamless and intuitive platform for disease prediction and access to tailored healthcare recommendations. Key features include user authentication mechanisms to safeguard personal data, a user-friendly interface for inputting symptoms, and integration with the Google Maps API to offer personalized recommendations of nearby medical specialists. This integration enhances user experience by facilitating convenient access to healthcare services. One of the distinguishing aspects of our project is its emphasis on proactive healthcare management through early disease detection. By leveraging machine learning algorithms to analyze symptom datta, our application enables users to receive timely alerts regarding potential health risks. This proactive approach empowers individuals to take proactive measures to safeguard their health and seek medical assistance when needed, ultimately contributing to improved health outcomes
Emotion Recognition with deep learning
(New Horizon College of Engineering, 2026) Shashank TJ 1NH23CE403 Niskarsh Naulakha 1NH22CE031 Naveen Kumar 1NH23CE402 Sanjay A 1NH22CE046
Emotion recognition using deep learning has emerged as a pivotal area of research, leveraging advanced neural network architectures to analyse and interpret human emotions from various modalities, including speech, text, and facial expressions. This mini project explores the integration of deep learning techniques in emotion recognition systems, focusing on their effectiveness, challenges, and applications. The project begins with a comprehensive literature review, highlighting the evolution of emotion recognition technologies and the role of deep learning in enhancing accuracy and efficiency. Traditional methods often relied on handcrafted features and shallow learning algorithms, which limited their performance. In contrast, deep learning models, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have demonstrated superior capabilities in automatically extracting relevant features from raw data, thus improving recognition rates. A significant portion of the project is dedicated to the implementation of a deep learning model for emotion recognition. The chosen model is trained on a diverse dataset comprising audio, visual, and textual inputs, allowing for a multimodal approach. Data preprocessing techniques, such as normalization and augmentation, are employed to enhance the robustness of the model. The training process involves optimizing hyperparameters and employing techniques like dropout and batch normalization to prevent overfitting. The evaluation of the model's performance is conducted using various metrics, including accuracy, precision, recall, and F1-score. The results indicate that the deep learning model outperforms traditional methods, achieving high accuracy in recognizing emotions such as happiness, sadness, anger, and surprise. Additionally, the project discusses the importance of dataset diversity and quality in training effective emotion recognition systems.
FairCourt: A System to Evaluate Fairness in Legal Judgments
(New Horizon College of Engineering, 2026) Pranava Gopavaram 1NH22CE036 Priyanka S Reddy 1NH22CE039 Keerthana N 1NH22CE020 Bhavishya Likitha Kondaveeti 1NH22CE007
The Indian judiciary, though constitutionally mandated to uphold justice and equality, faces critical challenges that undermine its effectiveness and credibility. These include inconsistent judicial interpretations, prolonged delays due to case backlogs, and growing concerns over biases related to caste, religion, gender, and socioeconomic status. Additionally, the lack of transparency in judicial appointments and perceived instances of corruption have contributed to a decline in public trust. To address these systemic issues, there is a pressing need for a data-driven, transparent, and scalable solution that can objectively evaluate the fairness of court judgments. FairCourt is an Al-powered framework designed to assess judicial fairness using advanced Natural Language Processing (NLP), fairness metrics, and explainable Al. The system architecture consists of six core modules: data acquisition from public legal platforms such as Indian Kanoon, NJDG, and eCourts, preprocessing to clean and structure unstructured legal texts, semantic analysis using transformer-based models like BERT and LegalBERT; fairness assessment through metrics like statistical parity and equality of opportunity; explainability using tools such as SHAP and LIME to provide interpretable justifications, and a dashboard to visualize fairness trends across judges, regions, and case types. By identifying patterns of bias and inconsistency in judgments, FairCourt aims to promote judicial accountability, improve public trust, and support policy reforms. Aligned with Sustainable Development Goal 16, the project advances the rule of law, reduces bias and corruption, and supports the development of transparent and inclusive judicial institutions
FINGERPRINTINSIGHT: Unveiling Crime Pattern Through Deep Fingerprint Analysis
(NHCE, 2024) Revappa 1NH20CE037; Sachin B Handiganur 1NH20CE038; Sachin Y M 1NH20CE052; Sumit S Purandare 1NH20CE052
Fingerprint images in crime scene are important clues to solve serial cases. Crime scene fingerprint identification system using deep machine learning with Convolutional Neural Network (CNN). Images are acquired from crime scene using methods ranging from precision photography to complex physical and chemical processing techniques and saved as the database. The images collected from the crime scene are usually incomplete and hence difficult to categorize. The fingerprint recognition system is divided into three stages that are fingerprint image pre-processing, feature extraction and matching. Suitable enhancement methods are required for pre-processing the fingerprint images. The output of this stage will be passed to feature extraction stage which is extract the minutiae point (ridge ending, Bifurcation) from thinning fingerprint image, then the false minutiae removal is applied to extract real minutiae. The features of pre- processed data are fed into the CNN as input to train and test the network.

Browse

Browsing Project Reports by Title

Results Per Page

Sort Options