PAI-Excel - Research

Punjab AI Excellence is mentoring students to do top quality AI Research. We support school students with introductory courses, and guide university students to master practical AI. Very proud to showcase the AI research from batch of interns who are from major universities of Punjab. Students tackled some very hard problems, and worked on writing research papers for practical AI applications.

Air Quality Prediction Using Machine Learning, August 2025. (Read PDF)

This paper presents a machine learning-based approach for forecasting air quality by predicting Air Quality Index (AQI) values and their corresponding health-related categories—’Good’, ’Moderate’, and ’Unhealthy’. Leveraging environmental and pollutant data, the system aims to provide early warnings about hazardous air conditions and associated health risks, thereby enabling proactive decision-making for individuals and authorities. Two distinct datasets were utilized for model development: the UCI Air Quality dataset and a city-specific dataset of Delhi sourced via Kaggle. Work done by Ekampreet Kaur of Dr. B.R.Ambedkar National Institute of Technology, Jalandhar.

Human Activity Detection Using IMUs, August 2025. (Read PDF)

In this project, we use two different popular datasets, UCI HAR and PAMAP2, to create a rich and more meaningful dataset to train a machine learning model. Both datasets include time-series recordings from different activities, captured at different body locations and under varies conditions. To align the data and ensure consistency, six common activities were selected from both datasets and performed preprocessing steps such as resampling, time window segmentation, normalization, mapping and more. We then trained deep learning model including CNN, LSTM, hybrid CNN-LSTM and CNN-BiLSTM. Combining more than one dataset improves the model’s ability to generalize and recognize activities more accurately. Also, makes the model train with different time-series pattern rather than sticking to one dataset. Work done by Sajandeep Singh of Guru Nanak Dev University.

Vision AI for Plant Disease Detection in Capsicum, Potato, and Tomato, August 2025. (Read PDF)

This paper proposes a deep learning-based Vision AI system for the early and accurate detection of plant diseases in three important crops—Capsicum, Potato, and Tomato. The model utilizes convolutional neural networks (CNNs) and transfer learning to analyze leaf images and classify them as healthy or infected, further identifying the specific type of disease present. Work done by Anshpreet Kaur of Chitkara University.

Plant Disease Detection Using Deep Learning and MobileNet, August 2025. (Read PDF)

This report presents the development of an AI-based plant disease detection system using deep learning techniques, specifically Convolutional Neural Networks (CNNs) and transfer learning. The system was trained on a balanced, augmented multi-crop dataset, achieving over 95% accuracy in both individual crop and multi-class classification tasks. Work done by Ankur Paul of Guru Nanak Dev Engineering College, Ludhiana.

Road Segmentation System Using Deep Learning with focused Edge Detection, August 2025. (Read PDF)

This report presents the development of a binary road segmentation system using the CamVid dataset and a U-Net architecture for autonomous driving applications. Although the initial model achieved high accuracy (~97–98%), it struggled with blurry road boundaries, which are crucial for real-world tasks like lane following and obstacle avoidance. To address this, an edge-aware training strategy was implemented by incorporating Canny edge detection and a custom weight map, improving the model’s performance and increasing the Intersection over Union (IoU) metric from 87% to over 92%. The final model demonstrates both accuracy and precise edge delineation, providing a strong foundation for future work in real-time road segmentation. Work done by Manpreet Singh of Khalsa College, Amritsar.

TextileVision Chatbot Using Retrieval-Augmented Generation (RAG), August 2025. (Read PDF)

This paper introduces TextileVision, a domain-specific chatbot designed to assist with accessing technical knowledge in the textile industry. Using a Retrieval-Augmented Generation (RAG) framework, the chatbot leverages semantic search with FAISS, pipeline orchestration with LangChain, and large language models to generate accurate responses based on a curated textile dataset. Initially developed using Hugging Face, the chatbot was later integrated into a website for textile industry operations, improving information access and reducing reliance on domain experts. TextileVision showcases the potential of AI to bridge knowledge gaps and support digital transformation in traditional industries like textiles. Work done by Navjot Kaur Cheema, Baba Banda Singh Bahadur Engineering College.

AI-Based Waste Classification using Deep Learning, August 2025. (Read PDF)

This study presents a deep learning-based approach to automating the classification of solid waste images into five categories using computer vision. A baseline CNN was trained from scratch and later outperformed by a fine-tuned EfficientNet-B0 transfer learning model. Using a combined dataset of over 5,500 images sourced from TrashNet and a manually curated custom dataset, the final EfficientNet model achieved a test accuracy of 87.08%, with balanced precision and recall across all classes. The study also tackles challenges such as class imbalance, dataset diversity, and limited computational resources. The trained model is designed for real-time deployment and is compatible with edge devices using TensorFlow Lite, contributing a scalable and practical solution for smart waste management systems. Work done by Preetinderjeet Singh, Guru Nanak Dev Engineering College, Ludhiana.

Plant Disease Detection using Deep Vision Models, August 2025. (Read PDF)

The timely detection of plant diseases is crucial for food security and sustainable agriculture. In this work, we present a deep learning pipeline built on the EfficientNetV2B0 architecture, tailored to classify 37 plant disease categories from a diverse dataset of over 31,000 leaf images. We employ a three-phase training strategy: initial head training, selective fine-tuning of top layers, and full-network fine-tuning, while handling class imbalance via computed class weights. Our best model achieves a test accuracy of 94.09% and a macro F1-score of 0.8411. The compact TensorFlow Lite export (~9 MB) enables real-time inference on edge devices, making this solution practical for field deployment. Work done by Aryan Prajapati Guru Nanak Dev Engineering College, Ludhiana.

AI-Based Plant Disease Detection Using Deep Learning, August 2025. (Read PDF)

Plant diseases pose a significant threat to crops, leading to substantial losses in both quality and yield. Early detection of these diseases is crucial for farmers to protect their plants and maintain productivity. This study explores the use of deep learning techniques to identify diseases in common crops such as tomato, potato, and sugarcane. By testing various models, including a custom convolutional neural network (CNN), pre-trained EfficientNetB0, and MobileNetV2, the results show that MobileNetV2, with its lightweight architecture, achieves high accuracy across different disease categories. These findings highlight the potential of AI-driven models as valuable tools in agriculture, especially in resource-limited settings. Work done by Mehakpreet Kaur, Guru Nanak Dev University.

Detection of Plant Diseases Using AI, August 2025. (Read PDF)

This project demonstrates how AI-based models can be used to identify plant diseases at an early stage using leaf images. The system is designed using Convolutional Neural Networks (CNN) with support from pre-trained models such as ResNet-50. Datasets such as PlantVillage, PlantDoc, and the New Plant Disease Dataset were used to train the model. The objective is to equip farmers and agricultural professionals with an accessible, image-driven solution to detect crop diseases quickly and take corrective actions. This work bridges the gap between AI and agriculture, contributing to food security and smarter farming practices. Work done by Tejnoor Singh, Akal University.

Page updated

Google Sites

Report abuse