KMITL

Permanent URI for this communityhttps://dspace.kmitl.ac.th/handle/123456789/1

Browse

Search Results

Now showing 1 - 9 of 9

Some of the metrics are blocked by your
consent settings
Comparative Analysis Of Convolutional Neural Networks, Long Short-Term Memory Networks, and Bert For Text-Based Emotion Classification
(2026-06-01)
Visutsak, Porawat
;
Tongbai, Tanajak
;
Ongrungruaeng, Duongduen
;
Phongwuttisak, Nuttiruj
;
Wiriya, Surapong
This paper presents a detailed experimental study and comparative analysis of three popular deep learning architectures (CNN, LSTM, and BERT) for emotion classification in written messages. Using a publicly available dataset of six unique emotional states (sadness, joy, love, anger, fear, and surprise), an effective ablation study was conducted to determine optimal architectural configurations, including a sequence length of 66 tokens and an embedding size of 200. To validate the results of a comparative analysis of model performance, a bootstrap technique (30 trials) and the Wilcoxon Signed-Rank test were used to eliminate potential bias. As shown by experiments, the tuned BERT architecture (with a learning rate of 2e-5) produced the most accurate and reliable result of 93.50% in classifying emotional states from texts. Moreover, with an appropriate sequence length configuration, the LSTM network (89.92%) significantly outperformed the CNN (89.65%), confirming the need to account for long-range dependencies in emotion classification. Overall, the research results show the key importance of hyperparameter tuning and the ability to handle complex information for emotion identification.
Some of the metrics are blocked by your
consent settings
Mathematical Benchmarking of Convolutional Neural Networks for Thai Dialect Recognition: A Spectrogram Texture Classification Approach
(2026-03-01)
Visutsak, Porawat
;
Ongrungruaeng, Duongduen
;
Wiriya, Surapong
;
Ryu, Keun Ho
This study rigorously evaluates 13 Convolutional Neural Network (CNN) architectures for Thai dialect recognition. By treating Automatic Speech Recognition (ASR) as a computer vision texture classification task, we processed an extensive 840-h dataset from the Spoken Language Systems, Chulalongkorn University (SLSCU) corpus. Raw audio from four major dialects—Central, Northern (Khummuang), Northeastern (Korat), and Southern (Pat-tani)—was transformed into 2D Mel-spectrograms using the Short-Time Fourier Transform (STFT). We analyzed a diverse range of architectures, including the VGG, Inception, ResNet, DenseNet, and MobileNet families, to establish the optimal trade-off between mathematical complexity and spectral feature extraction. Our experimental results identify NASNet-Mobile as the most effective model, achieving a macro-average F1-score of 0.9425. The analysis suggests that NASNet’s search-optimized cell structure is uniquely capable of capturing the multiscale texture of phonetic formants. In contrast, we observed a catastrophic mode collapse in VGG16 (32.97% accuracy), likely due to excessive parameter bloat, while Xception and MobileNetV2 maintained robust generalization. Confusion matrix analysis reveals high acoustic distinctiveness for Southern Thai (96.7% recall), whereas Northern Thai exhibits significant spectral overlap with Central Thai. These results support the hypothesis that CNNs interpret spectrograms as textures rather than discrete objects, positioning NASNet-Mobile as a high-performance, low-latency baseline for edge-device deployment in resource-constrained environments.
Some of the metrics are blocked by your
consent settings
Text analytics of IMDB reviews using latent Dirichlet allocation
(2025-08-29)
Wiriya, Surapong
;
Thongbai, Tanajak
;
Kamolsin, Chiranun
;
Chomjan, Suchada
;
Visutsak, Porawat
This study utilizes a seven-step text analytics methodology in MATLAB to analyze IMDB review data. The methodology includes: 1) Data Loading and Preprocessing, involving loading the data and preprocessing text using functions such as converting text to lowercase, tokenization, punctuation removal, stop word removal, short and long word removal, and lemmatization. 2) Exploratory Data Analysis, using word clouds to visualize the most frequent words. 3) Bag-of-Words Model Creation, including splitting the data into training and validation sets and removing infrequent words and empty documents. 4) Topic Modeling with LDA, testing different solvers and evaluating performance using perplexity. 5) Optimal Topic Number Selection, optimizing the number of topics by comparing validation perplexities. 6) Final Topic Model Training, training a final LDA model and assessing its performance. and 7) Topic Interpretation and Analysis, involving visualizing word clouds and finding relevant reviews for specific words.
Some of the metrics are blocked by your
consent settings
Baseline Performance of Pre-trained Models on Movie Genre Classification from Spectrograms
(2025-04-01)
Visutsak, Porawat
;
Treeraphapkajondet, Kavin
;
Sakphet, Visaroot
;
Nitinuntatip, Wachirawit
;
Satthong, Pawwinkan
This study investigates the use of deep learning for classifying movie genres based on audio spectrograms. We construct a dataset of movie trailers, transform them into spectrograms, and label them by genre. Then, we utilize MATLAB's pre-trained convolutional neural networks (CNNs) for classification, comparing the performance of 9 different architectures, including MobileNet-v2, RestNet-18, DenseNet-201, Places365-GoogLeNet, VGG-16, VGG-19, Inception-RestNet-v2, Inception-v3, and NASANet-Mobile. We evaluated all models based on their ability to classify movie trailers into five genres: action, romance, drama, comedy, and thriller. Our results, based on accuracy and F1-score across genres, indicate that VGG16 achieves the highest overall performance with an accuracy of 86.27%, an F1-score of 86.69%, a recall of 86.87%, and a precision of 87.28%. This research demonstrates the potential of leveraging pre-trained CNNs, particularly VGG-16, for effcient and effective audio-based genre classification in movie trailers.
Some of the metrics are blocked by your
consent settings
Genre Classification of Movie Trailers using Spectrogram Analysis and Machine Learning
(2024-01-01)
Visutsak, Porawat
;
Pensiri, Fuangfar
;
Netisopakul, Ponrudee
;
Punsathit, Navapoom
;
Rojsuwan, Pongkorn
This study investigates the application of machine learning for movie trailer genre classification using spectrogram analysis. Spectrograms, visual representations of a signal's frequency content over time, are extracted from movie trailers. Convolutional Neural Networks (CNNs), known for their image recognition capabilities, are employed to analyze these spectrograms and identify patterns that distinguish between different genres. Our approach is compared to a Random Forest model for performance evaluation. The models are trained on a dataset of movie trailers categorized into five genres: action, romance, drama, comedy, and thriller. Librosa, a Python library, is utilized for audio pre-processing, while the overall training process is conducted within the Python environment. This paper explores the potential of machine learning in conjunction with spectrogram analysis for effective movie trailer genre classification.
Some of the metrics are blocked by your
consent settings
FORECASTING CBOE OPENING VALUES: AN EFFECTIVE ARIMA APPROACH WITH TRANSFORMATION AND MODEL DIAGNOSTICS
(2024-01-01)
Visutsak, Porawat
;
Wongpanti, Ratchanon
;
Netisopakul, Ponrudee
This study investigates the use of an ARIMA model, coupled with Monte Carlo simulation, to forecast the opening value of a Volatility Index (VIX) time series. The data obtained from the Chicago Board Options Exchange (CBOE) for the years 1992–2019 have been transformed into stationary data using a detrend method and first-order difference. The Augmented Dickey-Fuller (ADF) test is used to ensure the data are adequately transformed. The autocorrelation function (ACF) and partial ACF (PACF) are then used to identify series with serial correlation and determine whether an autoregressive (AR) model is appropriate. Significant moving average (MA) lags are also determined for model identification. The Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) are used to find the best-fit ARIMA model. Among the evaluated ARIMA models, which ranged from a random walk ARIMA(0, 1, 0) to ARIMA(2, 1, 2), the ARIMA(2, 1, 2) model was found to be the most optimal, exhibiting the lowest AIC and BIC values. This model was then used to forecast the opening value for the year 2014, using 2013 data as the real data. The generated ARIMA(2, 1, 2) model demonstrates reasonable alignment with the actual 2014 data, suggesting its potential for forecasting in this context. Additionally, Monte Carlo simulations are employed to assess the model’s robustness by generating a range of potential outcomes, providing a more comprehensive understanding of the uncertainty associated with the forecasts.
Some of the metrics are blocked by your
consent settings
The smoothed 3D skeleton for animation
(2009-12-01)
Visutsak, Porawat
;
Prachumrak, Korakot
Skeleton is at the main interest of 3D character animation. In this paper, we use a recent skeleton pruning method based on the medial axis transform and the geodesic distance function that originates one-voxel thick, graph-like skeleton. Unfortunately, the location of skeleton joints does not usually match the anatomical joint of the model. We introduce a novel method for smoothing the pruned skeleton for realistic character animation by using the filtering process for adjusting the locations of consecutive points along the skeleton. Chord-to-point distance accumulation then be applied, and the smoothed skeleton is split in order to create segments and joints corresponding to its shape. The new skeleton can be regenerated later on. Therefore, the new skeleton produced from the proposed method can capture the essential shape characteristics in a compact form, while preserving the meaningful anatomical information of the 3D character models. The demonstration of the approach with several examples is also provided. © 2009 IEEE.
Some of the metrics are blocked by your
consent settings
Knowledge-based approach for 3D skeleton extraction
(2009-12-01)
Visutsak, Porawat
;
Boonjing, Veera
;
Prachumrak, Korakot
A 3D skeleton is a one-voxel thick, graph-like structure, widely used in the area of a character animation. In this paper, we propose a novel method for the 3D skeleton extraction of the polyhedral models based on a priori knowledge of the learned skeleton. The method has several steps. First, the octree of the input model is calculated and used to compare with the octree of the 3D models stored in the octrees database. By comparing the octree similarities, if the searched results exactly match with the input octree, the corresponding skeleton will be retrieved from the skeletons database. Otherwise, the method finds the list of the close match (the octree similarity ratio which the value is greater than 0.8) in order to use to estimate the new skeleton. In the worst case, if the input octree does not match with any case in the octrees database, the method computes the new skeleton and stores it in the skeletons database. The method is fast and efficient because it is not necessary to extract the skeleton from every input model. Thus, the computational time of our method depends only on the time of the octree similarity calculation, and the time for searching the similar octree in the octrees database. Several examples show the results obtained with our approach. Copyright © 2009 ACM.
Some of the metrics are blocked by your
consent settings
The 3D skeleton pruning for removing undesired joints
(2009-09-21)
Visutsak, Porawat
;
Prachumrak, Korakot
The skeleton is a low-dimension shape representation of 3D objects useful for different areas, such as machine vision, image processing, computer graphics, character animation, and etc. The most common techniques for skeleton computing are based on the Reeb graph and the shortest path finding. Using only the shortest path algorithms for extracting the critical points and constructing the Reeb graph over the suiface of the model may generate unwanted skeleton joints. A new skeleton pruning method for removing the spurious skeleton joints is introduced that is based on the medial axis transform and the geodesic distance fUnction. Consequently, the generated skeleton can capture in a simple and meaningful way the essential shape of the 3D objects in a compact form.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results