International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
Leveraging CIELAB Segmentation and CNN for Wheat Fungi Disease Classification
Get PDFWheat is the third most harvested and consumed grain globally, but a significant portion of its production is wasted due to diseases. Fungal infections caused by pathogenic fungi are particularly harmful, greatly reducing crop yields. Manual visual inspection of large fields is slow, exhausting, and requires specialized expertise. This research introduces a novel combination of image augmentation, CIELAB segmentation, and a fine-tuned pre-trained CNN, achieving an unprecedented 98.43% accuracy in wheat fungal disease classification, addressing gaps in current detection methods and promoting sustainable agriculture. To conduct this research, datasets from Kaggle were merged and meticulously validated to create a comprehensive set with five classes: healthy wheat and four fungal diseases. Preprocessing steps included resizing, contrast enhancement and noise removal to ensure uniform and high-quality images followed by rigorous image augmentation techniques to expand and diversify the dataset ultimately enhancing the deep learning model\u27s robustness and accuracy. The CNN model, trained over 80 epochs achieved an impressive 98.43% accuracy in classifying wheat fungal diseases. With a precision of 98.47% and an F1 score of 98.43% the model demonstrated strong positive classification accuracy. Additionally, a recall of 98.43% and specificity of 98.47% indicated its effectiveness in identifying true positive cases and accurately detecting disease presence or absence
A Detecting Land Use Land Cover Changes Induced by the Dynamics of River Indus, Pakistan, from 1972 -2022, Using Remote Sensing and GIS Techniques
No full textIntroduction/Importance of Study: This study evaluates the shifting of the Indus River and its impact on land use and land cover from 1972 to 2022 using Geographic Information System (GIS) and Remote Sensing (RS) techniques.
Novelty Statement: This research uniquely addresses the intricate relationship between river shifting and LULC changes, providing new insights into flood management and land use planning. Chronic alluvium erosion due to the river\u27s fast flow has led to poverty among residents and annual national asset losses, affecting the economy.
Material and Method: Using satellite images from 1972 to 2022, the research employed GIS and remote sensing techniques to analyse river sinuosity, channel migration, erosion, and accretion patterns, along with LULC changes. Methods included calculating the Indus River\u27s sinuosity index, assessing channel and bank migration, and applying the Normalized Difference Water Index and maximum likelihood classification for accurate LULC assessment.
Result and Discussion: Long-term results indicated that river erosion influenced land area, increasing settlement areas, decreasing vegetation, and causing fluctuations in barren land, water bodies, and agricultural land. Built-up areas expanded considerably, indicating population growth in floodplains. Erosion and deposition have notably affected agricultural and settlement areas, leading to socio-economic stress and internal migration. Satellite images taken during spring and dry seasons (March to May) showed minimal stream flow due to lower rainfall. Maximum erosion and management plans are critical for Reaches A, B, C, H, I, and J from 1972 to 2022. Minor embankment improvements are necessary for these reaches, as initial migration occurred on the right side for Reaches A, B, and C, shifted to the left from D to G, and affected both sides from H to J.
Concluding Remark: This research highlights how important GIS and remote sensing are for studying river changes and their effects on land use. It provides valuable information to help make better decisions about managing floods and planning land use
Analyzing the Shadows: Machine Learning Approaches for Depression Detection on Twitter
Get PDFDepression is a leading cause of disability worldwide, affecting approximately 4.4% of the global population. It can escalate from mild symptoms to severe outcomes, including suicide, if not treated early. Thus, developing systematic techniques for automatic detection is crucial. Social media platforms like Facebook, Twitter, TikTok, Snapchat, and Instagram provide users with the means to share personal feelings and daily activities, offering valuable insights into their thoughts and behaviors. This research aims to identify users who publicly disclosed their diagnosis and collect their data from Twitter. We created three different datasets, each varying in the number of tweets stored based on criteria discussed later. We selected six classifiers for analysis: Support Vector Machine (SVM), Logistic Regression, Random Forest, Max Vote Ensemble, Bagging, and Boosting. We conducted two analyses. In the first, textual data was converted into embeddings using the Bag of Words approach before analysis. In the second, a multivariate analysis, we trained algorithms on multi-dimensional data. Our findings revealed that Logistic Regression outperformed other techniques on smaller datasets. However, the Boosting algorithm yielded the best results on a dataset of 3,200 tweets, and the Bagging algorithm excelled when trained on 3,200 tweets of multivariate data. Overall, nearly all algorithms performed well on the 3,200-tweet datasets
Evaluating Faster R-CNN and YOLOv8 for Traffic Object Detection and Class-Based Counting
Get PDFReal-time traffic object detection is a critical component necessary for achieving a fully autonomous traffic system. Traffic object detection, along with background classification, is a significant area of research aimed at enhancing safety on the roads and reducing accidents by accurately identifying vehicles. This research aims to develop an accurate and efficient system for traffic object detection and classification in real-time traffic environments. It also seeks to minimize false positives and negatives, ensuring that no objects are overlooked in the detection of classes such as cars, buses, bicycles, motorcycles, and pedestrians. This research aims and focuses on the two following deep learning technologies: YOLO stands for (You Only Look Once) and Faster R- CNN stands for (Region-based Convolutional neural network). YOLO, initially designed as the single-stage approach, emphasizes speed; therefore, it is best suited for real-time uses. However, Faster R-CNN which is a two-stage detector gives better results in object detection and is highly accurate. Both models are trained and tested on the same data set containing 5712 trained images, 570 validation images, and 270 test images using a workstation with RAM 32 GB and NVIDIA GeForce RTX 4080 Super GPU through the help of CUDA version 12.4 to provide the end evaluating results. Since Faster RCNN is a very intensive model it took 22 hours to complete 3 epochs with an accuracy of 55.2% to train the model and YOLO finished the training within 10 epochs with the [email protected] value of 0.931 of all classes. Our results of traffic object real-time detection indicated that YOLO was vastly better and quicker than Faster R-CNN
VDMF: VANETs Detection Mechanism Using Fog Computing for Collusion and Sybil Attacks
Get PDFVehicular Ad Hoc Networks (VANETs) have evolved as a key component of the intelligent transportation system, enhancing road safety and traffic efficiency. It is crucial to secure sensitive information, and detection of incident response, whenever malicious activity is observed. Key components of VANETs include vehicles, Roadside Units (RSUs), and Fog servers (FS). Despite this, the open and evolving nature of VANETs introduces substantial security challenges, including exposure to malicious attacks like Sybil and collusion attacks. The proposed technique addresses the crucial security vulnerabilities in VANETs by developing a robust and efficient fog computing-based mechanism for detecting and mitigating Sybil and collusion attacks. The proposed approach emphasizes minimizing computational and communication overheads while ensuring timely and accurate detection and response to malicious activities. The results show that the proposed technique provides less communication and computational overheads in sparse and dense scenarios with enhanced security
AI-Powered Classification for Cheating Detection in Offline Examinations Using Deep Learning Techniques with CUI Dataset
Get PDFSupervising students during examinations is a very demanding process, and real-time supervision by human proctors proves to be challenging. This method is time-consuming and involves the extra work of monitoring several students concurrently. Automating exam activity recognition is perceived as one of the ways to address these challenges. In this work, we designed, implemented, and tested an accurate deep learning-based system that classifies student activities during examinations using a pre-trained ResNet50 model. This new concept helps expedite the rate at which exams are monitored and supervised, minimizing the role of human proctors. An amalgamated dataset was used, and the model works with six types of student behaviors, incorporating dropout layers for better performance. The Adam optimizer was employed for fine-tuning the learning process, and k-fold cross-validation was utilized to ensure the model\u27s robustness. The system achieved a high training accuracy of 96%, with 57% of all the documents producing output close to 1 for all categories, demonstrating high precision rates. The results indicate that the proposed method is reliable for future automated proctoring systems. Supervisors can focus on more critical tasks, such as addressing student concerns, rather than constantly monitoring every student\u27s movement. Moreover, the automation system provides consistent and unbiased supervision, eliminating human errors and fatigue that could otherwise affect the monitoring process. This ensures a fairer examination environment where all students are treated equally under constant vigilance. The system can also be scaled to handle larger examination rooms or remote testing, providing flexibility in its deployment
Exploring Character-Based Stylometry Features Using Machine Learning for Intrinsic Plagiarism Detection in Urdu
Get PDFPlagiarism detection in natural language processing (NLP) plays a crucial role in maintaining textual integrity across various domains, particularly for low-resource languages like Urdu. This study addresses the emerging challenge of intrinsic plagiarism detection in Urdu, an area with limited research due to the scarcity of datasets and model resources. To bridge this gap, our research investigates the use of character-based stylometric features in combination with machine learning (ML) and deep learning (DL) models specifically designed for Urdu text analysis. We conducted a series of experiments to evaluate the performance of several classifiers, including Random Forest, AdaBoost, K-Nearest Neighbor (KNN), Decision Tree, Gaussian Naive Bayes, and Long Short-Term Memory (LSTM) networks. Our results show that KNN and LSTM achieved the highest accuracy at 74%, with KNN outperforming the others in terms of F1-score (64.3%), highlighting its balanced performance across accuracy, precision, and recall. AdaBoost followed closely with an accuracy of 73% and a precision of 77.5%, although its F1-score was slightly lower at 63.6%. These findings emphasize the need for specialized approaches in NLP for Urdu, demonstrating that tailored ML and DL techniques can significantly improve intrinsic plagiarism detection in low-resource languages
Machine Learning in Livestock Management: A Systematic Exploration of Techniques and Outcomes
Get PDFThis Systematic Literature Review (SLR) examines the growing field of leveraging Machine Learning (ML) to improve livestock productivity. Through a meticulous analysis of peer-reviewed articles, the study categorizes research into key domains such as disease detection, feed optimization, and reproductive management. Various ML algorithms, including supervised, unsupervised, and reinforcement learning, are evaluated for their efficacy in enhancing herd health and management. The review also addresses the role of diverse data sources, such as sensor technologies and electronic health records and discusses the socio-economic and ethical implications of ML adoption in livestock farming. Insights into scalability, economic viability, and future research directions contribute to a comprehensive understanding of the current background and pave the way for sustainable and technologically advanced livestock management practices. This review serves as a valuable resource for researchers, practitioners, and policymakers in shaping the future of precision agriculture in improving livestock productivity
Towards End-to-End Speech Recognition System for Pashto Language Using Transformer Model
Get PDFThe conventional use of Hidden Markov Models (HMMs), and Gaussian Mixture Models (GMMs) for speech recognition posed setup challenges and inefficiency. This paper adopts the Transformer model for Pashto continuous speech recognition, offering an End-to-End (E2E) system that directly represents acoustic signals in the label sequence, simplifying implementation. This study introduces a Transformer model leveraging its state-of-the-art capabilities, including parallelization and self-attention mechanisms. With limited data for Pashto, the Transformer is chosen for its proficiency in handling constraints. The objective is to develop an accurate Pashto speech recognition system. Through 200 hours of conversational data, the study achieves a Word Error Rate (WER) of up to 51% and a Character Error Rate (CER) of up to 29%. The model\u27s parameters are fine-tuned, and the dataset size increased, leading to significant improvements. Results demonstrate the Transformer\u27s effectiveness, showcasing its prowess in limited data scenarios. The study attains notable WER and CER metrics, affirming the model\u27s ability to recognize Pashto speech accurately. In conclusion, the study establishes the Transformer as a robust choice for Pashto speech recognition, emphasizing its adaptability to limited data conditions. It fills a gap in ASR research for the Pashto language, contributing to the advancement of speech recognition technology in under-resourced languages. The study highlights the potential for further improvement with increased training data. The findings underscore the importance of fine-tuning and dataset augmentation in enhancing model performance and reducing error rates
Algorithmic Implementation and Evaluation for Image Segmentation Techniques
Get PDFThis research conducts a comprehensive comparative analysis of five prominent image segmentation algorithms, including Thresholding, K-Means Clustering, Mean Shift, Graph-Based Segmentation (Watershed), and U-Net (Deep Learning). The study employs a diverse set of five images and associated masks to rigorously evaluate algorithmic performance using key metrics such as Jaccard Index, Dice Coefficient, Pixel Accuracy, Hausdorff Distance, and Mean Intersection over Union. The findings reveal that the Threshold Algorithm consistently outperformed its counterparts, achieving perfect scores in Jaccard Index, Dice Coefficient, Pixel Accuracy, and Mean Intersection over Union, while minimizing Hausdorff Distance to 0. This emphasized its exceptional accuracy, precision, and agreement with ground truth segmentation, positioning it as an optimal choice for applications demanding precise segmentation, such as medical imaging or object detection. The research underscores the need to carefully consider specific application requirements and tradeoffs when selecting an algorithm, offering valuable guidance to researchers and practitioners in the field of image segmentation. The standardized approach outlined in the material and methods section ensures fair comparisons, making this study a valuable resource for informed decision-making in diverse imaging applications