KMITL

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

Browse

Search Results

Now showing 1 - 10 of 14

Some of the metrics are blocked by your
consent settings
Self-attention hierarchical kernel reservoir state network for inland water level prediction
(2026-01-15)
Liu, Zongying
;
Xu, Xiaohan
;
Pasupa, Kitsuchart
;
Loo, Chu Kiong
;
Wei, Yang
Waterway transportation sustainably facilitates global trade through eco-efficient cargo movement, where accurate water level forecasting is critical for ensuring navigational safety and operational continuity. To develop a highly accurate prediction model, it is essential to consider the periodic characteristics of water level data, which often emerge in real-world datasets. This study introduces a novel reservoir state structure based on reservoir computing theory, the Self-attention Hierarchical Kernel Reservoir State Network (SHK-RSN). It employs three primary mechanisms. First, a hierarchical feature extraction method groups training data and extracts high-dimensional features from these groups using the kernel trick in a hierarchical manner. Second, a self-attention weight selection approach is introduced to replace the random weights in the Hierarchical Kernel Reservoir State Network (HK-RSN), improving the rationale for hidden neuron connections and enhancing the interpretability of weight selection. Third, a novel reservoir state structure is proposed to capture periodic information and extract temporal features across periods, enabling the model to capture richer temporal information and identify relationships among periods. Experiments are conducted on one artificial and five real-world time series datasets, with forecast performance evaluated over 1–7 steps. Our proposed model, SHK-RSN, is compared with models based on randomization, the kernel trick, and deep learning. The experimental results demonstrate that SHK-RSN exhibits superior forecasting ability relative to the baselines. It achieves the best Symmetric Mean Absolute Percentage Error (SMAPE) across all datasets in the 1–7 period average among baseline methods, demonstrating a relative improvement of 25.7% to 46.9% over the conventional Echo State Network.
Some of the metrics are blocked by your
consent settings
Deep context-attentive transformer transfer learning for ﬁnancial forecasting
(2025-01-01)
Feng, Ling
;
Sinchai, Ananta
This study presents 2CAT (CNN-Correlation-based Attention Transformer), a deep learning model for financial time-series forecasting. The model integrates signal decomposition, convolutional layers, and correlation-based attention mechanisms to capture temporal patterns. A transfer learning framework is incorporated to enhance generalization across markets through pretraining, encoder freezing, and fine-tuning. Evaluation on six stock indices—Dow Jones Industrial Average (DJIA), Nikkei 225 (N225), Hang Seng Index (HSI), Shanghai Stock Exchange (SSE), Bombay Stock Exchange (BSE), and the Stock Exchange of Thailand (SET)—demonstrates strong predictive accuracy. On DJIA, 2CAT records an MSE of 0.0655, MAE of 0.2023, and R2 of 0.9169, outperforming Deep-Transformer, which yields an MSE of 0.1360 and R2 of 0.8274. The SET index, which posed challenges for previous models, demonstrates notable improvement with 2CAT, achieving an R2 of 0.9094. Wilcoxon signed-rank test confirms statistically significant gains in non-transfer learning scenarios at the 0.05 level. Transfer learning experiments reveal statistically significant improvements, reinforcing the feasibility of cross-market knowledge transfer. An ablation study highlights the impact of architectural refinements and rotary positional encoding, while prediction horizon analysis confirms stable forecasting performance. These results establish 2CAT as a robust financial forecasting framework adaptable to diverse market conditions.
Some of the metrics are blocked by your
consent settings
FACE SPOOFING DETECTION BASED ON DEEP FEATURE EXTRACTION AND INSTANCE-BASED CLASSIFICATION
(2023-02-01)
Claypo, Niphat
;
Jaiyen, Saichon
;
Hanskunatai, Anantaporn
Face recognition is an important task in smart home security for detecting a face or monitoring a person in a live video and verifying the identity of an authentic user. However, there have been spoofing face methods that can trick a face recognition algorithm into wrongly verifying the identity of the person. In this paper, we propose a new hybrid framework for spoofing face detection based on Convolutional Neural Network and Long Short-Term Memory (CNNLSTM) and instance-based learning algorithm. In addition, a new dataset called FSA-CCTV is proposed, which contains face images from CCTV video clips with many types of spoofing attacks. The performance of our method was compared to several other anti-spoofing methods: CNN and RI-LBP, SLRNN, HSV+YCbCr, ResNet50, YCbCr+SVM and YCbCr+KNN. The experimental results show that our method yielded 93.2% of Accuracy, 96.8% of Recall, 94% of Precision, 94.8% of F<inf>1</inf>-score and 0.93 of AUC on the FSA-CCTV dataset. From the experimental results we can conclude that the proposed algorithm outperforms other approaches and yielded the most stable classification accuracy on the proposed dataset.
Some of the metrics are blocked by your
consent settings
Sketch image classification using component based k-Nn
(2019-02-01)
Jearasuwan, Suwannee
;
Wangsiripitak, Somkiat
Recently, several researches on recognition of novice-users’ hand-drawn images have been conducted, specifically those on novel feature extraction techniques. Generation of a robust sketch descriptor is one of the most challenging problems in hand-drawn image recognition. Drawing or sketching is a common skill that everyone can do to a varying degree of success. Although an individual May or May not be able to create a beautifully drawn image, any human beings can recognize the shape of each part of a hand-drawn object and the entire object. In this work, we proposed a sketch image classification method that creates an image descriptor from its own components and uses k-NN algorithm for learning/classification. A sketch image can be created simply by drawing some simple geometric shapes. The drawing order, size, and number of the shapes are features that can be extracted and used to recognize an image. After an object in a drawn image has been classified as a car or a human being or anything else, the recognized object can be selected and paired with the motion associated with it. It was found that our proposed method was able to achieve a recognition accuracy rate of 92.33%. We also did a survey with children 7-12 years of age asking them whether they wanted an easy tool that can animate their hand-drawn objects and got almost unanimous affirmative responses from them.
Some of the metrics are blocked by your
consent settings
Comparison of feature extraction for accent dependent Thai speech recognition system
(2018-09-13)
Tantisatirapong, Suchada
;
Prasoproek, Chalisa
;
Phothisonothai, Montri
This paper aims to compare the feature extraction methods for accent dependent Thai speech from three regions including central, southern and northeastern regions. We investigate four frequency analysis methods: i.e., Energy Spectral Density (ESD), Power Spectral Density (PSD), Mel-Frequency Cepstral Coefficients (MFCC) and Spectrogram (SPT). Radial basis function kernel based on support vector machine is used as a classifier with 5-fold cross validation. The isolated speech data sets are recorded from 30 male and 30 female participants speaking the 10 Thai digits from 0 to 9. The MFCC-based feature gives better accuracy than ESD, PSD and SPT respectively. For within the same region, the MFCC-based feature provides average accuracy of 94.9% and 99.1% for male and female voices respectively. For the three regions, the MFCC-based feature provides average accuracy of 89.34% and 93.81% for male and female voices, respectively.
Some of the metrics are blocked by your
consent settings
Alphabetic hand sign interpretation using geometric invariance
(2014-01-20)
Adhan, Suchin
;
Pintavirooj, Chuchart
Hand alphabet is an important sign language for disability people for a long time. This communication is also necessary for a normal people to understand the meaning as well. Hand language interpretation by applying a hand image posture classification is an active research theme to solve obstacle. In this research, we propose a promising technique to apply a B spline curvature concept for supporting a triangular-based feature extraction element in a hand interpretation process. Area, inner angle and adjacent area ratio which derived from a curvature reference set are created a feature string for each alphabet posture in the template. By testing with all 24 hand alphabets, our system provides a promising result in identification satisfactorily.
Some of the metrics are blocked by your
consent settings
Redundancies in linear GP, canonical transformation, and its exploitation: A demonstration on image feature synthesis
(2011-03-01)
Watchareeruetai, Ukrit
;
Takeuchi, Yoshinori
;
Matsumoto, Tetsuya
;
Kudo, Hiroaki
;
Ohnishi, Noboru
This paper concerns redundancies in representation of linear genetic programming (GP). We identify the causes of redundancies in linear GP and propose a canonical transformation that converts original linear representations into a canonical form in which structural redundancies are removed. In canonical form, we can easily verify whether two representations represent an identical program. We then discuss exploitation of the proposed canonical transformation, and demonstrate a way to improve search performance of linear GP by avoiding redundant individuals. Experiments were conducted with an image feature synthesis problem. Firstly, we have verified that there are really a lot of redundancies in conventional linear GP. We then investigate the effect of avoiding redundant individuals. The results yield that linear GP with avoidance of redundant individuals obviously outperforms conventional linear GP. © 2010 Springer Science+Business Media, LLC.
Some of the metrics are blocked by your
consent settings
EMG signal feature extraction based on Wavelet transform
(2010-07-30)
Mahaphonchaikul, K.
;
Sueaseenak, D.
;
Pintavirooj, C.
;
Sangworasil, M.
;
Tungjitkusolmun, S.
In this paper, a multi-channel electromyogram acquisition system using programmable system on chip (PSOC) microcontroller was used to obtain the surface of EMG signal. Two pairs of single-channel surface electrodes were used to measure and record the EMG signal on forearm muscles. Then, different levels of Daubechies Wavelet family were performed to analyze the EMG signal. Finally, features in terms of root mean square, logarithm of root mean square, centroid of frequency, and standard deviation were used to extract the EMG signal. The experimental results show that root mean square feature extraction method exhibits better performance for extracting the EMG signal compared to the other features. In the future, our method can be utilized to control a mechanical arm in real-time processing.
Some of the metrics are blocked by your
consent settings
Feature extraction in medical ultrasonic image
(2007-01-01)
Udomhunsakul, Somkait
;
Wongsita, Pichet
This paper presents a method for speckle noise reduction as well as feature extraction in medical ultrasonic images to effectively reduce the speckle noise and extract the object in ultrasonic images. In noise reduction process, the logarithm transform of the ultrasonic image is analyzed into wavelet domain by using 2D stationary wavelet transform (SWT). Next, the Weiner filter is used to apply over areas in each subband (HH, HL, LH and LL). Finally, the inverse wavelet transform is computed and applying the exponential. In feature extraction process, first the denoised image is enhanced by histogram equalization technique. Haar filter is used to extract the object. Moreover, nonmaxima suppression technique is adopted to get the edge localization. Finally, the adaptive hysteresis thresholding is applied to get the final result. The experiments show that the proposed algorithm can be detected well-localized and thin edges.
Some of the metrics are blocked by your
consent settings
2D/3D vision-based mango's feature extraction and sorting
(2006-01-01)
Chalidabhongse, Thanarat
;
Yimyam, Panitnat
;
Sirisomboon, Panmanas
This paper describes a vision system that can extract 2D and 3D visual properties of mango such as size (length, width, and thickness), projected area, volume, and surface area from images and use them in sorting. The 2D/3D visual properties are extracted from multiple view images of mango. The images are first segmented to extract the silhouette regions of mango. The 2D visual properties are then measured from the top view silhouette as explained in [7]. The 3D mango volume reconstruction is done using volumetric caving on multiple silhouette images. First the cameras are calibrated to obtain the intrinsic and extrinsic camera parameters. Then the 3D volume voxels are crafted based on silhouette images of the fruit in multiple views. After craving all silhouettes, we obtain the coarse 3D shape of the fruit and then we can compute the volume and surface area. We then use these features in automatic mango sorting which we employ a typical backpropagation neural networks. In this research, we employed the system to evaluate visual properties of a mango cultivar called "Nam Dokmai". There were two sets total of 182 mangoes in three various sizes sorted by weights according to a standard sorting metric for mango export. Two experiments were performed. One is for showing the accuracy of our vision-based feature extraction and measurement by comparing results with the measurements using various instruments. The second experiment is to show the sorting accuracy by comparing to human sorting. The results show the technique could be a good alternative and more feasible method for sorting mango comparing to human's manual sorting. © 2006 IEEE.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results