Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
5 results
Search Results
Item Optimizing Object Detection in YOLOv5 using Adaptive Genetic Algorithm: A Study on Population Sizes for Enhanced Accuracy(Institute of Electrical and Electronics Engineers Inc., 2024) Naik, P.M.; Rudra, B.In recent years, deep learning-based object detection models like YOLOv5 have demonstrated remarkable efficiency in accurately identifying objects within images. However, achieving optimal performance still presents challenges. Many researchers have employed hybrid models to enhance the mean average precision (mAP) in the YOLOv5 architecture. However, only a few have explored hyperparameter-based optimization techniques to improve mAP. We use an adaptive genetic algorithm-based approach for analyzing Arecanut X-ray images to improve detection accuracy within the YOLOv5 framework. Balancing population size and computational efficiency is crucial for effective exploration and exploitation of the solution space in genetic algorithms without incurring unnecessary computational costs. This research investigates the influence of various population sizes on mAP using a genetic algorithm. Our findings indicate that a population size of 50 is ideal for achieving the best mAP among the explored sizes, when evolved for 300 generations. © 2024 IEEE.Item Classification of Arecanut X-Ray Images for Quality Assessment Using Adaptive Genetic Algorithm and Deep Learning(Institute of Electrical and Electronics Engineers Inc., 2023) Naik, P.M.; Rudra, B.The traditional approach for analyzing the quality of arecanuts is based on their external appearance. However, using machine learning and deep learning techniques, automated classifications were performed. But the true quality can only be analyzed when the internal structure of the arecanut is examined. Therefore, we use the X-ray imaging technique to determine the internal quality of arecanuts. We prepared a novel dataset of arecanut X-ray images and used a YOLOv5 based deep learning architecture for classification. The present study employs an adaptive genetic algorithm based approach for hyperparameter optimization to enhance the mean average precision (mAP) using a light weight model generated using a ghost network and a feature pyramid network (FPN). We have achieved the highest mAP of 97.84% using our method with a lower model size of 15 MB. Our method has excelled in detecting the arecanut compared to cutting-edge object detection algorithms such as YOLOv3, YOLOv4, Detetron, YOLOv6, YOLOv8, and YOLOX. We also acknowledged the performance enhancement using the adaptive genetic algorithm on the Pascal VOC 2007 image dataset. Despite of significant computational requirements for executing genetic algorithms, we proved that genetic algorithms can boost mAP. Additionally, the methodology developed in this investigation produced multiple models with the best mAP featuring optimized hyperparameters. This methodical strategy is helpful for the design of an automatic, non-destructive, integrated X-ray image based classification system. This system has the potential to revolutionize the quality assessment of arecanuts by offering a more efficient evaluation method. © ; 2023 The Authors.Item Quantum-inspired Arecanut X-ray image classification using transfer learning(John Wiley and Sons Inc, 2024) Naik, P.; Rudra, B.Arecanut X-ray images accurately represent their internal structure. A comparative analysis of transfer learning-based classification, employing both a traditional convolutional neural network (CNN) and an advanced quantum convolutional neural network (QCNN) approach is conducted. The investigation explores various transfer learning models with different sizes to identify the most suitable one for achieving enhanced accuracy. The Shufflenet model with a scale factor of 2.0 attains the highest classification accuracy of 97.72% using the QCNN approach, with a model size of 28.40 MB. Out of the 12 transfer learning models tested, 9 exhibit improved classification accuracy when using QCNN models compared to the traditional CNN-based transfer learning approach. Consequently, the exploration of CNN and QCNN-based classification reveals that QCNN outperforms traditional CNN models in accuracy within the transfer learning framework. Further experiments with qubits suggest that utilising 4 qubits is optimal for classification operations in this context. © 2024 The Author(s). IET Quantum Communication published by John Wiley & Sons Ltd on behalf of The Institution of Engineering and Technology.Item Framework for Lightweight Deep Learning Model Using YOLOv5 for Arecanut Grade Assessment(Springer, 2024) Naik, P.; Rudra, B.Arecanut grading using Machine Learning (ML) approaches can be less ideal in real-world scenarios that require real-time operations due to the need for rapid processing and adaptation to varying conditions. This paper proposes an efficient and lightweight deep learning model for real-time Arecanut detection suitable for industrial applications. We introduce a novel Arecanut image dataset and explore the YOLOv5 architecture to achieve a lightweight and efficient model for both grading and detection. Our investigation reveals that a hybrid model utilizing GhostNet in Darknet-53 convolutions and a Feature Pyramid Network (FPN) in the YOLOv5 neck structure achieves optimal performance. This model achieves the best mean Average Precision (mAP) of 98.85% while maintaining a compact model size of only 1.9 MB. We compared our design with the latest YOLO models, and ours performed better in detecting Arecanuts. This accurate and cost-effective solution empowers Arecanut producers with the ability to identify top-grade Arecanuts, leading to economic benefits and mitigating health concerns associated with manual grading processes. Finally, we propose a framework for integrating this model into the agro-industry for future real-world implementation. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2024.Item Deep learning-based arecanut detection for X-ray radiography: improving performance and efficiency for automated classification and quality control(Taylor and Francis Ltd., 2025) Naik, P.M.; Rudra, B.X-ray radiography is a valuable, non-destructive tool and can be used to examine the internal components or quality attributes of agricultural commodities, including arecanut. The true quality of an arecanut can be determined using destructive methods through visual inspection. However, dissected arecanuts do not have a shelf life. There is no non-destructive method available for grading arecanuts. We employ X-ray imaging as an aid to conduct internal examinations of arecanuts, allowing for thorough inspection without causing damage. A custom X-ray image dataset of arecanuts is created for automated interpretation of grades. We developed a hybrid arecanut grading model using YOLOv5s architecture incorporating the Stem, the GhostNet and the Transformer blocks. The proposed hybrid architecture outperforms in comparison with state-of-the-art models with a mean average precision (mAP) of 97.30%. The proposed 9.5 MB lightweight model can easily fit into X-ray devices, making it ideal for detecting arecanut grades in the industry. This method could transform the standards of quality inspection for arecanut. Its incorporation could establish a new industry benchmark for unparalleled quality assessment using X-ray technology as a non-destructive tool. © 2024 Informa UK Limited, trading as Taylor & Francis Group.