IAES International Journal of Artificial Intelligence (IJ-AI)
Not a member yet
1769 research outputs found
Sort by
Deep ensemble learning with uncertainty aware prediction ranking for cervical cancer detection using Pap smear images
Get PDFThis paper proposes a novel deep ensemble learning framework designed for the efficient detection and classification of cervical cancer from Pap smear images. The proposed study implements three advanced learning models namely DenseNet201, Xception, and a classical convolutional neural network (CNN) customized with optimal hyperparameters to automate feature extraction and cervical cancer detection process. The proposed study also introduces a novel ensemble learning to enhance the classification of cervical cancer. The proposed ensemble mechanism is based on the confidence aggregation followed by uncertainty quantification and prediction ranking scheme, thus ensuring that more reliable predictions have a proportionally greater influence on the final outcome. The primary goal is to leverage the collective intelligence of the ensemble in a manner that prioritizes reliability and minimizes the impact of less certain predictions. The experimental analysis is carried out on two dataset one with whole slide images (WSI) and another on cropped images. The proposed ensemble model achieves an accuracy rate 100 and 97% for dataset with WSI and with cropped images respectively
Exploring DenseNet architectures with particle swarm optimization: efficient tomato leaf disease detection
Get PDFThe critical challenge of tomato leaf disease demands effective solutions surpassing manual detection limitations, ensuring rapid intervention, optimal crop health, and maximizing yield for farmers. DenseNet, a convolutional neural network (CNN) architecture, is lauded for its adept handling of gradient flow issues by extensive interlayer connectivity. Its application holds significant promise in tackling the intricate task of identifying tomato leaf diseases. This research introduces an innovative methodology employing particle swarm optimization (PSO) to fine-tune the DenseNet architecture and hyperparameter. The proposed approach efficiently converges on optimal configurations encompassing parameters, such as the number of layers in dense blocks, growth rates, dropout rates, activation functions, and optimizers tailored for DenseNet. The DenseNet-PSO model achieves remarkable accuracy and precision in classifying various tomato leaf diseases, outperforming alternative architectures in total parameters, computational efficiency, and overall performance compared with six other architecture models. These outcomes elucidate DenseNet-PSO's efficacy in tomato leaf disease detection and demonstrate
Classification of upper gastrointestinal tract diseases using endoscopic images
Get PDFAutomatic classification and disease detection in medical images, aided by machine learning, provide crucial support to prevent overlooked instances and ensure prompt treatment of diseases. Despite impressive achievements in the field of polyp detection from endoscopic images, classification of other diseases, such as reflux esophagitis, esophageal cancer, gastritis, gastric cancer, and duodenal ulcer, is still subject to significant limitations and remains a challenging area of study because of their different and more challenging characteristics. This paper proposes a method to roughly classify the diseases from the whole images by deep learning. In particular, we focus on identifying hard samples from the training dataset and enriching them with some fundamental augmentation techniques. We then employ a cutting-edge model, specifically ResNet, for the final classification stage. Additionally, we enhance the original ResNet’s loss function by incorporating another loss function called focal loss. These modifications play a crucial role in boosting the accuracy of the ResNet model. Our proposed method outputs the disease category and corresponding heat map showing the area of interest. It achieved very promising accuracy (99.55%) for the classification of five lesions on our self-collected dataset. It serves a dual purpose. Firstly, it aids in the training of novice endoscopists, enabling them to gain valuable experience. Secondly, it offers a rapid solution for annotating extensive volumes of endoscopic image data at the label level
Novel similarity measures for Fermatean fuzzy sets and its applications in image processing
Get PDFDigital imaging is growing in our day-to-day life ranging from selfies to medical imaging. The extended applications of the field open doors for the researchers in the present-day context. The extraction of useful information from digital images is crucial because it depends on the various characteristics of the image. Fuzzy theory provides a better understanding of the image characteristics and, thus extracts meaningful information, even under uncertain situations. The present study reports the Fermatean fuzzy sets (FFSs) application in image processing while proposing similarity measures. These similarity measures highlight the perfect and precise results from an image while using multiple parameters of the image for information extraction. The study concludes that the proposed similarity measures provide a better estimation of data from an image used in image processing problems
Bias in artificial intelligence: smart solutions for detection, mitigation, and ethical strategies in real-world applications
Get PDFArtificial intelligence (AI) technologies have revolutionized numerous sectors, enhancing efficiency, innovation, and convenience. However, AI's rise has highlighted a critical concern: bias within AI algorithms. This study uses a systematic literature review and analysis of real-world case studies to explore the forms, underlying causes, and methods for detecting and mitigating bias in AI. We identify key sources of bias, such as skewed training data and societal influences, and analyze their impact on marginalized communities. Our findings reveal that algorithmic transparency and fairnessaware learning are among the most effective strategies for reducing bias. Additionally, we address the challenges of regulatory frameworks and ethical considerations, advocating for robust accountability mechanisms and ethical development practices. By highlighting future research directions and encouraging collective efforts toward fairness and equity, this study underscores the importance of addressing bias in AI algorithms and upholding ethical standards in AI technologies
Quality and shelf-life prediction of cauliflower using machine learning under vacuum and modified atmosphere packaging
Get PDFEnsuring the freshness and quality of cauliflower during storage and transportation is essential due to its high perishability. This study harnesses the power of machine learning to predict the quality and shelf-life of cauliflower under cost-effective vacuum and modified atmosphere packaging (MAP) techniques. By investigating key parameters such as total soluble solids (TSS), pH, weight loss, and color change, a significant impact on post-packaging quality was identified. To address the challenge of accurate color change measurement, an innovative method utilizing a bilateral filter for noise reduction and particle swarm optimization (PSO) with Markov random field (MRF) segmentation was developed. TSS, weight loss, and color change were identified as key parameters, and leveraging these parameters, artificial neural networks (ANN) were employed to create highly precise predictive models, achieving R-squared values of 0.952 for TSS, 0.992 for weight loss, and 0.981 for color change. This approach not only enhances the efficiency and sustainability of food production and distribution but also minimizes food waste and maximizes profitability for cauliflower in global markets through the use of cost-effective packaging solutions
Optimizing deep learning models from multi-objective perspective via Bayesian optimization
Get PDFOptimizing hyperparameters is crucial for enhancing the performance of deep learning (DL) models. The process of configuring optimal hyperparameters, known as hyperparameter tuning, can be performed using various methods. Traditional approaches like grid search and random search have significant limitations. In contrast, Bayesian optimization (BO) utilizes a surrogate model and an acquisition function to intelligently navigate the hyperparameter space, aiming to provide deeper insights into performance disparities between naïve and advanced methods. This study evaluates BO's efficacy compared to baseline methods such as random search, manual search, and grid search across multiple DL architectures, including multi-layer perceptron (MLP), convolutional neural network (CNN), and LeNet, applied to the Modified National Institute of Standards and Technology (MNIST) and CIFAR-10 datasets. The findings indicate that BO, employing the tree-structured parzen estimator (TPE) search method and expected improvement (EI) acquisition function, surpasses alternative methods in intricate DL architectures such as LeNet and CNN. However, grid search shows superior performance in smaller DL architectures like MLP. This study also adopts a multi-objective (MO) perspective, balancing conflicting performance objectives such as accuracy, F1 score, and model size (parameter count). This MO assessment offers a comprehensive understanding of how these performance metrics interact and influence each other, leading to more informed hyperparameter tuning decisions
Real-time age-range recognition and gender identification system through facial recognition
Get PDFFacial recognition and age estimation are being implemented in apparel retailing which is undergoing significant changes due to fashion and technology. To improve interaction with customers and refine marketing strategies. The paper proposes an approach based on a Siamese neural network and the use of tools such as MediaPipe for face detection and DeepFace for age and gender estimation. In addition, the four stages of the research work, real-time image capture, ID assignment, facial feature extraction, and data storage, are described. Early approaches to age estimation were based on biometric features, such as eyes, nose, mouth, and chin, resulting in limited accuracy and low performance in older adults. To improve accuracy, additional elements, such as the presence of wrinkles, were considered and a diverse database of images was used. The proposed methodology achieves a positive result for real-time age classification and gender ID. The results include information on gender, age, ID, time and date for each person identified
Hybrid methods to identify ovarian cancer from imbalanced high-dimensional microarray data
Get PDFScientists have used microarray data to identify healthy people and patients with various types of cancer, including ovarian cancer. Ovarian cancer is the most dangerous of all types of cancer that attacks the female reproductive organ. The right combination of methods is needed to identify ovarian cancer from microarray data because that type of data is high-dimensional and imbalanced. This research aims to propose two hybrid methods which are a combination of infinite feature selection (IFS) as features selector with classification and regression tree (CART) as a classifier. IFS can work with two separate scenarios, namely supervised infinite feature selection (SIFS) and unsupervised infinite feature selection (UIFS). This research also compares the performance of the two hybrid methods proposed (SIFS-CART and UIFS-CART) with CART without IFS. The data used is OVA_ovary that has 10937 columns and 1545 rows. The results shows that SIFS-CART achieves maximum performance using 1000 features and UIFS-CART 5000 features. CART without IFS uses all 10935 features. The balanced accuracy results show SIFS-CART can outperform CART without IFS and UIFS-CART. Using less features to get highest balanced accuracy results, SIFS is more effective in performing feature selection on the OVA_ovary dataset compared to UIFS
Hybrid object detection and distance measurement for precision agriculture: integrating YOLOv8 with rice field sidewalk detection algorithm
Get PDFThis study aims to propose a new approach to semantic segmentation of sidewalk images in rice fields using the YOLOv8 algorithm, with the objective of enhancing agricultural monitoring and analysis. The experimental process involved preparing the development environment, extracting data from JSON, and training the model using YOLOv8. Evaluation reveals consistent and accurate sidewalk detection with a confidence score of 0.9-1.0 across various environmental conditions. Confusion matrix and precision-recall analysis confirmed the robustness and accuracy of the model. These findings validate the effectiveness of the approach in detecting and measuring sidewalks with high precision, potentially improving agricultural monitoring. The novelty of this study lies in the utilization of the RIFIS-D algorithm as an integral part of a hybrid approach with YOLOv8. This hybridization enriches the model with additional capability to detect the distance between the sidewalk and the tractor, addressing specific needs in agricultural applications. This contribution is significant in the advancement of automatic navigation and monitoring technology in agriculture, enabling the implementation of more sophisticated and efficient systems in field operations