KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/kmitl/1

Browse

Search Results

Now showing 1 - 10 of 11

Some of the metrics are blocked by your
consent settings
Advancing enterprise credit rating through multi-head self-attention and transfer learning
(2026-01-01)
Zhang, Rui
;
Li, Qing
;
Chen, Jiangchuan
;
Li, Jie
;
Sakdanuphab, Rachsak
This study introduces the multi-head self-attention rotary-positional encoding Informer (MHSA-RPEI), a novel enterprise credit rating model that leverages self-correlation weights and positional information of financial indicators to enhance rating accuracy. Corporate credit rating presents challenges such as temporal dynamics, complex non-linear relationships, and cross-regional variations, which traditional and deep learning models struggle to address. The MHSA-RPEI model overcomes these limitations through four innovations: (1) objective MHSA-RPEI model, which is specifically designed for credit rating, achieving high precision using only financial data and serving as an objective and validated tool; (2) revolutionary rotary positional encoding (RoPE), which enhances temporal modeling, improving dynamic risk assessment; (3) strategic transfer learning (TL), which facilitates cross-regional adaptations, mitigating data scarcity and domain variations; and (4) optimized informer architecture, which improves long-term risk forecasting based on extensive financial time series. Experiments on the U.S., Chinese, and Japanese financial data showed that MHSA-RPEI achieved 92.16% accuracy (U.S.), 90.32% (China), and 92.50% (Japan), with area under the receiver operating characteristic curve scores ranging from 88.47% to 91.95%, surpassing traditional and state-of-the-art models. Furthermore, TL experiments highlighted the exceptional adaptability of the MHSA-RPEI for cross-regional applications, yielding performance gains of 5–6%. Ablation studies confirmed that the RoPE and self-correlation weight modules were crucial to the effectiveness of the model.
2
Some of the metrics are blocked by your
consent settings
Time Series Forecasting Using Transfer Learning with an Attention-Revamped Transformer: A Case Study of Financial Instruments
(2025-12-01)
Feng, Ling
;
Sinchai, Ananta
Accurate financial time series forecasting is essential for informed investment and risk management decisions. Traditional methods, including statistical techniques such as SMA, ARIMA, VAR, and LASSO, and deep learning models like LSTM and Transformer, often fall short of capturing the complex seasonal and cyclical dynamics inherent in financial data. To overcome the noted shortcomings, this study introduces a transfer learning approach utilizing an enhanced Transformer model with correlation-based attention mechanisms. This proposed model significantly improves its capacity to capture long-term dependencies and perform robust cross-market predictions. Initially trained on the Dow Jones Index, it demonstrates superior transferability to diverse asset classes, including stock indices, commodities, and cryptocurrencies. Experimental evaluations across multiple metrics, including MSE, MAE, MHD, and R<sup>2</sup>, reveal that the proposed model consistently outperforms benchmarks. Notably, it achieves outstanding predictive accuracy in BTC (MSE: 0.0352) and SET (MSE: 0.1701), establishing a strong foundation for advanced transfer learning applications in financial forecasting across varied markets.
Some of the metrics are blocked by your
consent settings
Design of an integrated model for pothole detection and repair optimization using multimodal transformers and hybrid deep learning
(2025-12-01)
Meshram, Kundan
;
Saurabh, Aryan
;
Kharole, Vinay
;
Chatrabhuj
;
Mishra, Umank
The detection and timely repair of potholes are crucial for maintaining road safety and minimizing vehicle damage. However, existing methods often suffer from limitations such as reliance on single-modal data, poor generalization across diverse environments, and suboptimal resource management. To address these challenges, we propose a comprehensive framework for enhanced pothole detection and repair optimization using advanced deep learning techniques. Our approach integrates four key methodologies: Multimodal Enhanced Pothole Detection with Person-Level Data (M-E-Pot holeNet), Hybrid Machine Learning-Deep Learning for Classification (Hybrid-Pot holeNet), Deep Reinforcement Learning for Pot hole Detection and Repair Optimization (DRL-Pot holeOpt), and Transfer Learning for Pothole Detection in Diverse Environments (TL-Pot holeAdaptNet). M-E-Pot holeNet employs a Self-Supervised Multimodal Transformer (SSMT) to fuse camera, accelerometer, and crowdsourced smartphone data, achieving robust detection with a 97 % accuracy and under 2 % false positive rate. Hybrid-Pot holeNet combines Graph Attention Networks (GAT) and XGBoost, modeling spatial road features to classify potholes with 95 % accuracy and an F1-Score of 0.92. DRL-Pot holeOpt uses Soft Actor-Critic (SAC) with Bayesian Optimization to efficiently schedule repair tasks, reducing repair costs by up to 20 % and crew travel time by 15–25 %. Finally, TL-Pot holeAdaptNet leverages Domain-Adversarial Neural Networks (DANN) to ensure cross-domain adaptability, with 90 % accuracy in new environments and a 40–50 % reduction in domain discrepancy. This multi-faceted approach addresses the limitations of previous work by providing scalable, real-time, and resource-optimized solutions for pothole detection and maintenance, offering significant improvements in accuracy, cost efficiency, and adaptability.
1
Some of the metrics are blocked by your
consent settings
Quality Classification of Sunglasses Lens by Deep Learning
(2025-01-01)
Theskham, Charin
;
Jearanaitanakij, Kietikul
Sunglasses lens are medical devices designed to correct human vision while also providing protection against ultraviolet (UV) radiation. Sunglasses lenses come in various colors, each offering different light-filtering properties. Currently, the quality classification of sunglasses lenses during the production process still relies on human vision. Therefore, this research aims to study the design and quality classification of sunglasses lens with both evenness and unevenness colors using machine vision and deep learning techniques. However, from the review of existing studies on sunglasses lens quality inspection, informal research has been conducted in this area. This work is considered a new contribution with potential for practical application in the industry. In this study, a Convolutional Neural Network (CNN) will be used. To save research time, the researchers employed transfer learning techniques using models such as VGG16, VGG19, InceptionV3, Xception, DenseNet121, ResNet50, EfficientNetB0, EfficientNet -B7, and EfficientNetV2L. The classification results are divided into 2 categories for evenness colors lens and unevenness colors lens. The dataset used real photos of sunglasses lens from Hoya Lens Thailand as the data source. A dataset of 1,250 real images of sunglasses lens was used, comprising 625 images of evenness colors lens and 625 images of unevenness colors lens. Data augmentation was performed by rotating the images 90, 180, and 270 degrees, as well as vertically flipping the sunglasses lens images. This process yielded a total of 10,000 images, with 5,000 images each for lenses with evenness and unevenness colors. The results of applying Transfer Learning of each model for classifying the quality of sunglasses lens that the DenseNet121 model achieved the highest performance, with an accuracy of 82.23 % and precision of 82.32 %
Some of the metrics are blocked by your
consent settings
Advanced White Blood Cells Detection and Analysis with VGG16 Transfer Learning
(2024-01-01)
Eamkong, Anoma
;
Pintavirooj, Chuchart
;
Tungjitkusolmun, Supan
White blood cells (WBCs) provide a significant role in the immune system, and precise classification and quantification are critical for detecting a variety of diseases. This work employs the VGG16 model combined with transfer learning to improve WBC classification and counting in blood smear pictures. We utilized a dataset of 1,651 images, including samples from normal bone marrow and chronic myeloid leukemia (CML) cases, to train and evaluate the model. The workflow incorporates data augmentation, Otsu thresholding, and morphological operators to improve segmentation accuracy. The VGG16-based model achieved a high accuracy of 92.37%, with validation accuracy reaching 96.41 %. Performance metrics were evaluated using accuracy, precision, recall, and F -measure, highlighting that eosinophils provided the highest accuracy, while neutrophils demonstrated the best precision. The model effectively distinguishes between normal and elevated WBC counts, as evidenced by the results from normal and CML blood smears. Despite these promising results, further refinement is suggested. Future work may focus on specific WBC types, such as monocytes or neutrophils, to improve precision and balance memory efficiency. This approach has the potential to advance diagnostic accuracy and efficiency in medical image analysis.
1
Some of the metrics are blocked by your
consent settings
Subcellular Protein Patterns Classification Using Extreme Gradient Boosting with Deep Transfer Learning as Feature Extractor
(2024-01-01)
Phankokkruad, Manop
;
Wacharawichanant, Sirirat
Proteins are essential structural and functional components of human cells. Understanding and identifying proteins can provide valuable insights into their structure, function and role in human body. Subcellular proteins provide the expression that characterizes the many proteins and their conditions across cell types. This work proposed a classification model for subcellular protein patterns using XGBoost with transfer learning of CNN as the feature extractor. In the model training process, we used ResNet50, VGG16, Xception, and MobileNet as the pre-trained models based on the transfer learning technique to extract different features. The proposed model was used to classify subcellular proteins into 28 patterns. The XGBoost with ResNet50, VGG16, MobileNet, and Xception model achieved an accuracy level of 92.20% 92.77%, 91.63%, and 91.44%, respectively. The XGBoost with ResNet50, VGG16, MobileNet, and Xception model obtained an F1 score of 0.9179, 0.9233, 0.9131, and 0.9152, respectively. Considering the F1 score, All XGBoost with transfer learning of CNN models gave a high score. Therefore, all evaluation parameters clearly demonstrate the high performance of the subcellular protein pattern classification model.
1
Some of the metrics are blocked by your
consent settings
Rice Diseases Recognition Using Transfer Learning from Pre-trained CNN Model
(2023-01-01)
Hamhongsa, Wittawat
;
Wiangsripanawan, Rungrat
;
Thorncharoensri, Pairat
This research aims at applying the well-known pre-trained convolutional neural network (CNN) image classification algorithms such as InceptionV3, Xception, ResNetV2, InceptionResNetV2, and DenseNet to classify the five diseases of rice's leaves in Thailand. Data sets used are from 3 sources: UCI database, Rice Leaf Disease Image Samples dataset and images collected by authors in Thailand between 2018–2020. Our initial experimental result shows that using the pre-trained CNN models to classify the rice leaves disease seems to be possible with a greater number of images required. Therefore, the image data augmentation technique is used to add the number of images into the dataset. The experimental result with data augmentation shows that it could increase rice disease classification efficiency up to 16.533% (especially for the ResNet50V2 model).
1
Some of the metrics are blocked by your
consent settings
A CNN-BASED MULTI-MODEL ENSEMBLE METHOD FOR INDOOR AND OUTDOOR MULTI-VIEW STEREO RECONSTRUCTION
(2022-01-01)
Wattanacheep, Bhattarabhorn
;
Chitsobhuk, Orachat
Camera poses estimation is a critical process that ensures the success of Three-Dimensional (3D) modelling. We present a Convolutional Neural Network (CNN)-based multi-model ensemble method for indoor and outdoor multi-view stereo reconstruction capable of learning across multiple domains, including images from both indoor and outdoor environments. Each domain’s images have distinct properties and shooting view-points, which leads to difficulty in efficient learning such a large difference and requires large amount of computational resources. In order to reduce complexity of the end-to-end single model, the proposed model is divided into multiple learning agents consisting of domain-specific agents and domain relationship agent. The domain-specific agent is trained independently on its own set of unique image characteristics, for example, one for indoor datasets and another for outdoor datasets. The domain relationship agent then ensembles and analyzes the multiple domain features and finalizes the estimation. In terms of average root mean square error, we compare the performance of the combined domain single model with the suggested ensemble CNN model. The experimental results indicate that the proposed model outperforms the others, with rotation and translation prediction errors of 0.112012266.
1
Some of the metrics are blocked by your
consent settings
Car Damage Detection and Classification
(2020-07-01)
Kyu, Phyu Mar
;
Woraratpanya, Kuntpong
Nowadays, the proliferation of automobile industries is directly related to the increasing number of car incidents. So, insurance companies are facing many simultaneous claims and solving claims leakage. The sense of Artificial Intelligence (AI) based on machine learning and deep learning algorithms can help to solve these kinds of problem for insurance industries. In this paper, we apply deep learning-based algorithms, VGG16 and VGG19, for car damage detection and assessment in real-world datasets. The algorithms detect the damaged part of a car and assess its location and then its severity. Initially, we discover the effect of domain-specific pre-trained CNN models, which are trained on an ImageNet dataset, and followed by fine-tuning, because some of the categories can be fine-granular to get our specific tasks. Then we apply transfer learning in pre-trained VGG models and use some techniques to improve the accuracy of our system. We achieve the accuracy of 95.22% of VGG19 and 94.56% of VGG16 in the damaged detection, the accuracy of 76.48% of VGG19 and 74.39% of VGG16 in damage localization, the accuracy of 58.48% of VGG19 and 54.8% of VGG16 in damage severity with the combination of transfer learning and L2 regularization. From their results, the performance of VGG19 is better than VGG16. After analyzing and implementing our models, we find out that the results of using transfer learning and L2 regularization can work better than those of fine-tuning.
2
Some of the metrics are blocked by your
consent settings
An Empirical Study of Deep Neural Networks for Glioma Detection from MRI Sequences
(2020-01-01)
Coupet, Matthieu
;
Urruty, Thierry
;
Leelanupab, Teerapong
;
Naudin, Mathieu
;
Bourdon, Pascal
Gliomas are the most common central nervous system tumors. They represent 1.3% of cancers and are the 15th most common cancer for men and women. For the diagnosis of such pathology, doctors commonly use Magnetic Resonance Imaging (MRI) with different sequences. In this work, we propose a global framework using convolutional neural networks to create an intelligent assistant system for neurologists to diagnose the brain gliomas. Within this framework, we study the performance of different neural networks on four MRI modalities. This work allows us to highlight the most specific MRI sequences so that the presence of gliomas in brain tissue can be classified. We also visually analyze extracted features from the different modalities and networks with an aim to improve the interpretability and analysis of the performance obtained. We apply our study on the MRI sequences that are obtained from BraTS datasets.
1

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results