IAES International Journal of Artificial Intelligence (IJ-AI)
Not a member yet
1769 research outputs found
Sort by
Automatic detection of dress-code surveillance in a university using YOLO algorithm
Get PDFDress-code surveillance is a field that utilizes an object detection model to en- sure that people wear the proper attire in workplaces and educational institutions. The case is the same within universities, where students and staff must adhere to campus clothing guidelines. However, campus security still enforces univer- sity student clothing manually. Thus, this experiment creates an object detection model that can be used in the campus environment to detect if students are wear- ing appropriate clothing. The model developed for this research has reached an f1-score of 45% with an overall 51.8% mean accuracy precision. With this, the model has reached a satisfactory state with room for further improvements
Financial text embeddings for the Russian language: a global vectors-based approach
Get PDFThe article presents a software implementation of the linguistic embedding method for the Russian language, based on the global vectors for word representation (GloVe) model. The GloVe method allows to obtain word vectors that reflect their semantic and syntactic properties. The resulting vector model can be used in various natural language processing (NLP) tasks, such as machine translation and text clustering. The article describes the architecture of software that implements a method similar to the GloVe algorithm for Russian-language financial texts. The mechanisms used to train the model as well as to compute word vectors are described. Testing with typical classification methods demonstrated that the developed program generates accurate vector representations of Russian-language texts, proving effective in various NLP tasks. This work is one of the first studies devoted to the software implementation of the GloVe method for the Russian language using learning algorithms based on sparse matrices. The results of this study can be used in various NLP tasks, such as machine translation and text clustering
Hadamard Walsh space based hybrid technique for image data augmentation
Get PDFImage data augmentation (IDA) is common when deep learning is used for image classification to address the issue of overfitting. Overfitting occurs when the datasets are small and the deep learning models have a huge capacity. Overfitting models have low training errors but high validation errors and result in poor generalization. Several methods have been researched in this context, but frequency domain-based methods are less explored. In this research, we have explored the Hadamard and Walsh space and developed two hybrid technique for IDA. The proposed techniques use a combination of Hadamard/Walsh transform and geometrical transformations. Empirical study is carried out using the VGG-16 model for image classification on the CIFAR-10 dataset and the results are compared with existing methods. The analysis of the results shows that the proposed techniques improve the evaluation parameters significantly. Further, analysis of training loss vs. validation loss shows that the proposed Hadamard-based hybrid methods have better generalization ability than the proposed Walsh-based hybrid method
Neural networks based-simple estimated model for greenhouse gas emission from irrigated paddy fields
Get PDFThe current study aims to develop a simple model for estimating greenhouse gas emissions originating from paddy fields, utilizing backpropagation neural networks. The model integrated three input parameters: soil moisture, soil temperature, and soil electrical conductivity (EC), while generating estimations for two output parameters: methane (CH4) and nitrous oxide (N2O) emissions. The model was put into practice across three different irrigation systems, i.e., continuous flooded (FL), wet (WT), and dry (DR) regimes. For model training and validation, the input parameters were measured by a single 5-TE sensor. Concurrently, CH4 and N2O emissions were determined utilizing a closed chamber, and gas samples were subjected to laboratory analysis. Findings unveiled that the developed model accurately estimated CH4 and N2O emissions, demonstrating commendable coefficient of determination (R2) values ranging from 0.60 to 0.97 for validation process. Notably, the WT irrigation system exhibited the highest precision, boasting R2 values of 0.97 for CH4 and 0.73 for N2O estimation, respectively. Conversely, the FL irrigation system has the lowest accuracy with R2 values of 0.66 and 0.60. Despite variances in accuracy across irrigation systems, the overall performance remained deemed acceptable, warranting the model's applicability for estimating greenhouse gas emissions under diverse irrigation scenarios
Deep learning architectures for location and identification in storage systems
Get PDFThis document exposes the application of two deep learning models based on ResNet-18 architectures, intended for the location and identification of products in storage areas. One model obeys a tree structure and the other a structure under an ouroboron cycle. The performance of both models is evaluated using the metrics of training time, processing time and level of learning precision, which allows recommendations to be made regarding which one should be used for order preparation purposes, based on multilevel feature extraction. The total training time of the first model is 34.65 minutes and the second 40.43 minutes. The analysis of results allowed the detection parameters to be adjusted, finally with the refined models, through confusion matrices, precision results greater than 90% and processing times are obtained, which for model 1 is 6.8565 seconds and for model 2 is 4.884 seconds. For practical purposes, training times are not relevant, as are the precision and processing times for selecting the most convenient model according to the end user's objectives
Improving lithium-ion battery reliability through neural network remaining useful life prediction
Get PDFThe reliable performance of lithium-ion batteries is crucial for the safe and efficient operation of electrical systems, particularly in electric vehicles. To mitigate the risk of battery failure due to degradation, accurate forecasting of the remaining useful life (RUL) is imperative. In this study, we propose employing various recurrent neural network (RNN) methods, including RNN, gated recurrent unit (GRU), and long short-term memory (LSTM), to enhance RUL prediction accuracy for lithium-ion batteries. Our approach aims to provide reliable, accurate, and simple estimates of remaining battery life, facilitating effective management of electric vehicle power systems and minimizing the risk of failure. Performance evaluation metrics such as mean absolute error (MAE), R-squared (R²), mean absolute percentage error (MAPE), and root mean squared error (RMSE) are utilized to assess prediction accuracy. Experimental validation conducted using the NASA lithium-ion battery dataset demonstrates the superiority of LSTM in reducing prediction error and enhancing RUL prediction performance compared to alternative approaches. These findings underscore the potential of neural network methodologies in advancing battery management practices and ensuring the longevity and reliability of lithium-ion battery systems
Review of recent advances in non-invasive hemoglobin estimation
Get PDFHemoglobin is essential for diagnosing conditions like anemia and respiratory issues. Traditionally, the assessment of hemoglobin necessitates invasive techniques that involve blood draws, which can induce discomfort and present possible complications for patients. Recent advancements in non-invasive technologies have light-emitting diode (LED) to the development of smartphone applications and machine learning algorithms that allow real-time hemoglobin level estimation, eliminating the need for blood sampling. This not only improves patient comfort but also enhances access to ongoing health monitoring. This review aims to delve into the newest developments in smartphone-oriented strategies for hemoglobin estimation, highlighting their importance within contemporary healthcare practices and the potential implications they might have for more expansive clinical applications. Technological advancements have combined smartphones and artificial intelligence (AI) for non-invasive hemoglobin estimation, offering a promising alternative to traditional methods. These solutions optimize data collection and analysis processes, enhance diagnoses' accuracy, and facilitate timely medical interventions. Advancements in technology have revolutionized medical diagnostics, particularly in estimating hemoglobin levels non-invasively. AI methodologies have demonstrated significant results in accurately forecasting hemoglobin concentrations through a variety of analytical strategies. Future research should focus on the best configurations for these networks and the physiological concepts underpinning spectral data interpretations
A transfer learning-based deep neural network for tomato plant disease classification
Get PDFThe agriculture sector plays a significant role in Morocco's economy, and tomato farming is an essential component of this industry. However, tomato plants are prone to various diseases that can adversely affect productivity and quality. A novel approach to detect tomato plant diseases is proposed int this study, by modeling and developing a transfer learning-based convolution neural network (CNN) model that processes real-time images. The model is trained and validated with a deep CNN using a private dataset of 18,159 annotated tomato leaf images collected from experimental farms over five months. The performance of our residual neural network (ResNet-50) model is evaluated using stochastic gradient descent (SGD) and adaptive moment estimation (Adam) optimizers to demonstrate superior efficiency. Farmers can simply send images of their tomato leaves through our platform, and the trained model will identify accurately the disease. The developed model demonstrates exceptional performance, achieving a 0.96 F1 score and an 97% accuracy rate when tested on a dataset generated from real-world fields. This approach not only improves disease detection but also contributes to sustainable farming practices and enhanced productivity
Literature review on forecasting green hydrogen production using machine learning and deep learning
Get PDFGreen hydrogen is a sustainable and clean energy source, for this purpose, it conducts the global energy transition. The integration of artificial intelligence (AI), especially machine learning (ML) and deep learning (DL) with the process of green hydrogen production is essential in enhancing its production. This literature review studies in detail the intersection between AI and green hydrogen. Firstly, it concentrates on ML and DL algorithms used in forecasting green hydrogen production. Secondly, it presents an analysis of the studies released from 2021 to March 2024. Finally, the focus is on the results realized by the ML and DL algorithms proposed by the studies reviewed. This study provides a summary that explains the trends and methods used, as well as highlights the gaps and the opportunities in the field of AI and green hydrogen production. This liternature review presents a solid foundation for future research initiatives in this field
Survey and comparative analysis of phishing detection techniques: current trends, challenges, and future directions
Get PDFIn the age of digital communication, scams such as phishing continue to be a problem, necessitating the need for ever-more-advanced detection techniques to safeguard sensitive data. Examining several methods now in use, this review article groups them according to the application (email, web server, mail server, or browser-based). It explores the advantages and disadvantages of behavior-based, heuristic-based, machine learning (ML)-based, and signature-based techniques and offers a comparative evaluation of their efficacy. The essay delves deeper into the latest developments in phishing detection research, such as ML-powered social media exploration and real-time website analysis. The evaluation goes beyond just identifying detecting techniques; it also includes a data-driven analysis. In particular, random forest and support vector machines are ML algorithms that regularly produce results with high accuracy for detecting phishing attempts. Metrics like as recall, F1-score, and precision show how well these algorithms. Furthermore, specialised techniques such as heuristic-based and cantina-based approaches provide remarkable performance, underscoring the possibility of additional research in this field. Future research explores improved phishing detection through: better accuracy with ML, integrating new technologies, analyzing user behavior. A hybrid approach combining these techniques offers a stronger defense