Artificial Intelligence for Next Generation Computing, Volume 1

Inhaltsverzeichnis

1. Frontmatter

2. NLP-Based Approach for Efficient Keyword Identification in Unstructured Text

   Minu Choudhary, Sourabh Rungta, Shikha Pandey

   Abstract

   This paper examines the need for efficient keyword extraction methods to improve search accuracy and indexing due to the increasing volume of digital information. Traditional approaches, often based on word occurrence counts, tend to lose the contextual meaning of the text. The current work proposes a four-step NLP-based method for keyword extraction, including tokenization, POS tagging, noun phrase extraction, and chunking. The novel contribution of this research is the design of a grammatical pattern for extracting keywords (in the form of noun phrases) without losing the actual meaning of the text. This approach aims to retain the contextual meaning of the text. When compared to the most popular NLTK tool RAKE, the proposed method delivers better results while preserving the contextual meaning of the input text.

3. Enhancing Cybersecurity Log Analysis with LogGPT: A Novel LLM-Based Approach

   Raj Laxmi Singh, Vaishnavi Jadhav, Mahak Chamria, Vikas Kumar Jain

   Abstract

   Ensuring compliance and finding anomalies in large log collections are major issues in modern cybersecurity environments. This study presents LogGPT, a novel approach that revolutionizes the conventional paradigm of compliance detection and monitoring. Large Language Models (LLMs) are a powerful tool that LogGPT uses to analyze logs without the assistance of other programs. The study examines the inherent drawbacks of the current approaches and offers LogGPT as an advanced substitute. By showcasing LLMs’ greater flexibility and contextual awareness compared to more conventional deep learning models, LogGPT offers improved compliance monitoring accuracy by 14% over models like GPT4ALL, and overall, it achieves an accuracy of 89.43%. Experimental results show the effectiveness of the proposed approach.

4. Development of an Enhanced Pattern Mining Model Leveraging Deep Learning and Genetic Algorithms

   Reena, Akhilesh Tiwari, Saumil Maheshwari

   Abstract

   The need to efficiently and accurately mine frequent patterns from transactional data is, therefore, quite imperative for businesses desiring to understand customer behavior and market trends. Most of the traditional pattern mining algorithms, like Apriori and FP-Growth, suffer from scalability and accuracy issues in processing large and complex datasets and their samples. It addresses these limitations by coming up with an improved algorithm that makes use of machine learning techniques to provide enhancements in efficiency, scalability, and accuracy in pattern discovery. We present a model that puts together the functioning of RNNs with LSTM units, CNNs, and autoencoders. RNNs with LSTM units have been added to model long-term dependencies in sequential data and further the accuracy of pattern identification. Through their convolutional layers, CNNs are used in the detection of local patterns. This brings about efficiency and scalability in parallel processing. Dimensionality reduction is done on transactional data using autoencoders, which in turn makes the key patterns easier to extract, and further improves efficiency in processing. In the pattern extraction phase, the FP-Growth algorithm constructs an FP-tree for the identification of frequent itemsets, whereas the Apriori algorithm iteratively goes through candidate itemsets to make sure that only very relevant and most frequent patterns are extracted. Association rule mining, in this way, generates relationships between items that are then evaluated by support, confidence, and lift measures to optimize such patterns. Genetic algorithms simulate a process of natural selection that evolves candidate solutions to guarantee the optimization of patterns for the most valuable insights in the process. This approach offers overall relief not only from the limitations of the existing algorithms but also garners significant benefits in dealing with large datasets, improved efficiency, scalability, and accuracy. The optimal patterns extracted from them give very deep insight into customer buying behavior and market trends, hence help in data-driven decision-making processes. It is the hybridization of the state-of-the-art machine learning models with traditional pattern mining techniques that is the latest development in this field and it foretells its impactful applications in all types of business genres.

5. An End-to-End Brain–Computer Interface for Mental Workload Classification Based on EEGNet Model and Data Augmentation

   Rishabh Kholiya, Mitul Kumar Ahirwal

   Abstract

   Mental workload or cognitive load can be stated as how much mental resources are required to complete a task. High mental workload for a prolonged period can lead to mental fatigue, poor decision making, mental illness, etc. That shows the importance of monitoring mental workload. Electroencephalography (EEG) signals have been used to measure the brain activity. The dataset used in this paper is of mental workload during arithmetic task and based on number of tasks performed; subjects were classified into good counter and bad counter (Zyma et al. in Data 4:14, 2019). In this work, convolution neural network (CNN) model called as EEGNet is utilized for classification of mental state. In original dataset, average classification accuracy was 61.78% ± 0.076, to improve the accuracy data augmentation through Synthetic Minority Oversampling Technique (SMOTE) is used. After data augmentation, average classification accuracy reached to 99.56% ± 0.005.

6. Vaquita Dolphin Detection on Remote Sensing Data Using YOLOV8

   Jyothi Priya Inturi, Sunny Nalluri, Sri Lakshmi Dammu, Hamsa Kumari Kovvasu

   Abstract

   Dolphins belong to the group of marine mammals called cetaceans. They hold a key place at the top of the marine food chain and are essential to preserving the ecosystem’s general equilibrium. However, various factors such as gillnet entanglement, climate change, fisheries bycatch, and plastic pollution have led to the endangerment of certain dolphin species, such as the Vaquita dolphins. Therefore, there is an urgent need for detection and classification methods to conserve these species and protect their populations. The application of cutting-edge deep learning models such as You Only Look Once (YOLOv8) for object detection is selected to detect Vaquita dolphins in satellite images. This methodology holds promise for enhancing our understanding of dolphin populations and informing conservation efforts in critical habitats. By leveraging YOLOv8 for detection, we have achieved promising results in identifying Vaquita dolphins with higher accuracy. The results demonstrated promising accuracy and performance metrics, with the YOLOv8 model achieving mAP of 0.927.

7. Enhancing Surveillance Systems with Deep Learning-Based Anomaly Detection

   M. Evany Anne, M. Brindha, N. Sivakumaran

   Abstract

   An anomaly in surveillance is defined as any suspicious activity in video footage. An object with an unusual motion pattern or trajectory entering the frame is considered an anomaly. Examples include bike accidents or a person driving at high speed in a crowded place. This paper focuses on detecting anomalies in single-scene surveillance footage, as it is particularly useful for real-world applications where cameras are located at a fixed position. An unsupervised approach is used, employing an autoencoder and decoder model to minimize the reconstruction error between the input and generated output. The model is trained, and the reconstruction error and training loss are observed. After training, regularity scores are calculated for each test case. The proposed model outperforms other models in many scenarios, achieving a precision of 0.87 at a standard threshold of 0.75, which is better compared to other encoder-decoder models. However, the accuracy of the model decreases as the threshold increases.

8. Semantic Segmentation of Indian Road Scene Images with a Focus on Small Objects

   J. Umamahesh, C. V. Jawahar

   Abstract

   Semantic segmentation is a well-known task in computer vision, with many applications in autonomous navigation. This task aims to partition the image (set of pixels) into multiple labeled subsets of pixels, often referred to as regions. Such a segmentation is key to the scene understanding. This also results in localization of the objects in the image. There exists many research papers that study semantic segmentation in outdoor (often road) scenes. Semantic segmentation of small objects is specially challenging, and critically useful. However, this did not get sufficient attention in the past. The problem is challenging due to (i) availability of only small number of pixels on the small objects, (ii) class imbalance in learning, etc. These make the today’s deep learning architectures less effective in performance when it comes to small objects. This can be directly observed from the fact that semantic segmentation performance on small objects in many popular datasets has low accuracy. In this paper, we investigate the challenges associated with and the directions for design of algorithms for segmenting small objects. It is well known that loss functions affect performance in semantic segmentation. The segmentation of small objects also depends heavily on the loss functions used in training the deep learning-based solutions. We also investigate this in this paper. This problem is more severe in Indian situations where many small objects are often seen on the road. The small objects (such as pedestrians) are also very important to segment out for autonomous navigation and driver-assistance systems. In the Indian driving scenario, the Indian Driving Dataset (IDD) provides a class of annotated small objects. These are captured and annotated in an unstructured environment. Thus, we focus primarily on segmentation of small objects in IDD.

9. Unveiling Dementia’s Early Signals: Deep Learning Meets Image Processing

   Maddela Likitha, Sunny Nalluri, Inti Alekhya

   Abstract

   Dementia is a progressive disease that impairs cognitive functions such as memory, thinking, and communication, affecting over 40 million individuals worldwide, according to the World Health Organization. Early diagnosis is essential for effective management and treatment. In this study, we present a Custom Convolutional Neural Network (CNN) model designed to analyze Magnetic Resonance Imaging (MRI) scans for dementia detection. The model classifies dementia into four stages: Non-demented, Very Mildly Demented, Mildly Demented, and Moderately Demented. Our system incorporates advanced preprocessing techniques, such as image resizing, Gaussian filtering for noise reduction, and normalization, to enhance image quality and ensure consistent inputs for the CNN. Performance evaluation of the Custom CNN was carried out using key metrics like accuracy, precision, recall, and the confusion matrix to measure classification performance. The Custom CNN achieved an accuracy of *98.18%*, demonstrating its effectiveness in accurately detecting dementia stages. The confusion matrix further confirmed the model’s high precision in classification, showing significant improvements over existing models. This result underscores the potential of the proposed system to support early diagnosis, leading to timely interventions and improved patient outcomes. Our Custom CNN model offers a substantial advancement in dementia detection, contributing to both research and clinical practice in addressing the global challenge of dementia.

10. An Approach for Cyclone Tracking and Monitoring

    C. A. Rishikeshan, R. Jayanthi, Snehasis Ghosh, Navoneel Mondal, Srijanbroto Deb

    Abstract

    Accurate monitoring and forecasting of tropical cyclones is essential to reduce their devastating impacts on vulnerable coastal regions. This study presents an approach using convolutional neural network (CNN) that uses deep learning to detect and classify cyclones from remote sensing images. It uses a custom designed CNN architecture; the model effectively captures the distinct features between cyclone activity and normal weather conditions. The network architecture consists of convolutional layers that extract high-level semantic features which is followed by a fully connected layer for final classification. Performance evaluation is conducted through accuracy measurements and confusion matrix analysis. The proposed CNN model is novel compared to existing methods for its ability to retrieve meaningful contextual information (high-level features) from remotely sensing images, effectively distinguishing cyclone activity from normal weather patterns, offering improved accuracy (92.5%) in detection compared to traditional approaches. Despite challenges such as limited labelled storm data, the model shows strong potential to improve storm detection in different regions. This research helps in advancing deep learning applications in disaster risk management, providing a valuable tool to enhance storm monitoring and preparedness efforts.

11. Punjabi Speech Corpus with Tonal Characteristics

    Jaspreet Kaur Sandhu, Munish Kumar, Amitoj Singh

    Abstract

    In recent years, the Automatic Speech Recognition (ASR) system for Punjabi has gained significant popularity. Over the past decade, considerable efforts have been devoted into improving Punjabi speech recognition accuracy. New methods and optimized architectures have been proposed to increase the Punjabi ASR recognition rate. Despite that, the main obstacle pointed out is the non-availability of a standardized Punjabi speech corpus. Therefore, decision regarding designing a good standardized speech corpus for the Punjabi language has been initiated to eliminate the problem related to non-availability of standardized Punjabi speech corpus. This newly designed corpus covers all the vital characteristics of a good speech corpus, is large in size, and aids to boost the Punjabi speech recognition research. The new Punjabi speech corpus comprises 119,500 utterances and 65 h of recording. In total, 180 speakers’ voices from the Malwa, Majha, Doaba, and Powadh region of Punjab, in India are recorded and considered. Both male and female speakers of ages 20–45 are involved. Annotation of recordings is performed with the help of the PRAAT toolkit after the recording part is completed.

12. Leaf Disease Detection using Deep Learning: A Survey

    Pradeep Nazareth, S. U. Deepika, Rakshitha, M. N. Varshini, Vidyalaxmi D. Navalur

    Abstract

    Demand for global food productivity increased from past many years. At the same time, agriculture is facing number of challenges from pests and diseases. Thus, it is necessary to predict and detect various diseases in agriculture. The efficiency and accuracy of predicting and detecting agricultural diseases will play a vital role in improving productivity. Technology play a major role in predicting and detecting agricultural diseases. Deep learning (DL) technology is one of the potential technology, that can be used in the field of agriculture to predict and detect various diseases. This paper focuses on reviewing the use of various DL techniques to predict and detect the crop leaf diseases. Further, discusses the on current trends, challenges, and future research directions in using deep learning for crop leaf disease identification and broader plant disease and pest management strategies.

13. Machine Learning Integration in Bamboo Treatment and Seasoning for RCC Compound Wall Panels

    Shilpa Kewate, Pankaj Kewate

    Abstract

    This research explores the use of machine learning (ML) to optimize bamboo treatment and seasoning for reinforced cement concrete (RCC) compound wall panels. The study focuses on improving key processes such as drying, splitting, and bending of bamboo, along with its treatment for waterproofing, anti-termite protection, and enhanced bonding with concrete. By analyzing environmental data and monitoring in real-time, ML models help to fine-tune these processes, ensuring consistent quality and durability of bamboo reinforcement. Predictive modeling enables the adjustment of treatment conditions and parameters, reducing waste and improving energy efficiency. Machine learning can enhance the effectiveness of bamboo processing, leading to stronger and more reliable RCC panels. This approach not only supports sustainable construction practices but also improves the performance and longevity of bamboo-reinforced concrete structures.

14. Biorthogonal Wavelet-Based Image Enhancement for Accurate Pneumonia Diagnosis from Lung CT Scans Using Deep Learning

    S. Rohith Kumar, S. Vineth Ligi, R. Kumar, Samiappan Dhanalakshmi

    Abstract

    Identifying pneumonia in lung CT scans is complex as the hidden symptoms of the disease are not evident and barely distinguishable by the naked eye, leading to probable misdiagnosis or loss of cases. Further, it is rather challenging to distinguish between the diseases because of the complex structure of the lungs and the variety of their pathology. This study’s originality is found in an image preprocessing approach that enhances the diagnosis of pneumonia from lung radiographs. Specifically, a wavelet-based image enhancement technique is used to improve the visibility of infection details in the images, the preprocessed images are fed into several deep learning networks, which classify the outputs into three categories: Pneumonia due to COVID-19, pneumonia due to other etiologies, and no findings. Furthermore, to measure the effectiveness of the suggested technique and the outcomes of this proposed preprocessing method, it experiments on various deep learning models such as DenseNet, ResNet50, and GoogLeNet. As a result, DenseNet achieved a better accuracy of 99.20%, which is higher than ResNet50 97.93%, and GoogLeNet 97.59%. The observations indicate that the proposed preprocessing leads to greater sensitivity and better results in diagnosing pneumonia, further stressing the applicability of this method to aid in early and accurate diagnosis potentially.

15. YOLOV8 for Urban EV Detection: Sustainable Mobility Management

    Tarushi Singh, Samiappan Dhanalakshmi

    Abstract

    This paper explores the use of YOLOv8 algorithm for object detection to identify electric vehicles (EVs) from other conventional diesel/petrol vehicles in real time in urban parking infrastructure. By leveraging a dataset with 1200 high-resolution images, augmented to 3353 images, the model was trained and evaluated under various conditions like diverse lighting, background clutter, and different vehicle orientations. The results showcased the algorithm’s remarkable performance, with a mean average precision (mAP) of 91% and precision and recall rates of 91.8%. These metrics underscore YOLOv8’s validity and accuracy in intricate, real-world scenarios, making it a compelling and powerful tool for optimizing parking facilities and managing EV charging stations. The framework also plays a critical role in developing sustainable urban mobility by effectively assigning dedicated charging stations and facilitating the harmonious integration of EVs into current infrastructure. This paper highlights the rising significance of EV detection in environment friendly urban development, providing insights into practical applications and future research directions to further increase intelligent parking solutions. YOLOv8 provides an encouraging solution to the challenges of real-time EV detection, contributing substantially to sustainable urban planning and transportation systems.

16. Enhanced Alzheimer’s Disease Classification Using CNN and SMOTE with Optimizer Evaluation

    Chitresh Singhal, Nilesh Kumar Pandey, Rekh Ram Janghel

    Abstract

    Alzheimer’s disease is a significant neurological condition that result in memory loss and a steady deterioration in cognitive function. It primarily affects those 65 and older. Early diagnosis is crucial for efficient illness management and better patient outcomes. This paper presents a through deep learning approach that begins with the over-sampling synthetic minority method (SMOTE) to balance the Alzheimer’s Disease Neuroimaging Initiative’s (ADNI) dataset which improves the model’s ability to learn from minority classes. Following that, a 2D convolutional neural network (CNN) is used to categorize MRI images into four groups: non-demented, very mild demented, mild demented, and moderate demented. The model underwent rigorous experimentation, assessing various optimizers, dropout rates, and architectural configurations. Finally, the 4 Convo + 4 Pool layer configuration combined with a dropout rate of 0.3 and Adam optimizer achieved an impressive classification accuracy of 99.4%. These findings illustrate the effectiveness of the proposed methodology in accurately detecting Alzheimer’s disease and demonstrate its potential for clinical implementation.

17. Deep Learning for Brain Tumor Detection Xception CNN-based MRI Image Classification

    Narsingh Nath Bauddha, Rakesh Kumar Khare

    Abstract

    Premature recognition of brain tumors is crucial. Only biopsies can classify brain cancers, requiring invasive brain surgery. Brain malignancies can be identified and classified by medical professionals with the use of computationally focused procedures. Here, we present a deep learning approach based on a comparison of several recent studies that used a variety of machine learning techniques to diagnose three different tumor types using magnetic resonance brain images: glioma, meningioma, and pituitary gland in addition, no tumors were also included in the analysis. This makes it possible for doctors to accurately diagnose malignancies in their early stages. In this study, 7023 MRI brain pictures were utilized in this investigation. The Xception network creates a novel convolution neural network (CNN) once the images have been preprocessed and enhanced. Multiple convolution layers with 3 * 3 kernel functions are present in the network. To avoid overfitting, batch normalization layers were utilized, and the ReLU function served as each layer’s activation function. The Adamax optimization function was used with a 0.001 rate of learning to maximize efficiency. The suggested model’s validation accuracy was determined to be 99.70%, while its training accuracy was found to be 99.98%. The current investigation demonstrates that the suggested CNN Xception Net model has achieved the best classification accuracy for brain tumors. This model obtained excellent performance and optimal execution time compared with other CNN and machine learning approaches used in earlier research. Radiologists and doctors can utilize this recommended network in healthcare facilities to diagnose brain tumors.

18. Machine Learning Model to Study Single-Cell Osteoarthritic Chondrocytes via Raman Spectroscopy

    Gavish Uppal, Nisha, Tarun Goyal, Anup Kumar, Rajesh Kumar

    Abstract

    Raman spectroscopy (RS) has become an imperative instrument in biomedical research. It is a nondestructive and label-free laser spectroscopic technique, which can identify chemical composition of a biological sample by detecting multiple molecular components simultaneously without the need or minimal sample preparation under physiological condition. A major limitation while analyzing Raman spectra of a biological sample is the overlapping nature of vibrational modes of different molecules at a single spectral peak. Machine learning (ML) techniques encompass various multivariate and statistical analysis methods, which can be broadly classified into supervised and unsupervised techniques. Multivariate curve resolution-alternating least square (MCR-ALS) technique is a multivariate ML model that can potentially deconvolute overlapping Raman vibrational modes and extract the meaningful spectra that can be associated with biochemical components along with their concentrations. In this study, Raman spectral data were acquired from osteoarthritic chondrocytes. The spectra of three cellular components associated with DNA, proteins, and lipids were extracted using MCR-ALS technique, and it was observed that concentration of DNA and proteins decreased while concentration of lipids increased with progression of osteoarthritic grades. The study indicated that combined approach of Raman-MCR analysis can open new possibilities for improved Raman spectral analysis, which might assist further in providing the meaningful interpretation of complex Raman spectra originated from the biological samples.