KMITL

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

Browse

Search Results

Now showing 1 - 10 of 10

Some of the metrics are blocked by your
consent settings
Compact Thai Sign Language Translation by Deep Learning
(2024-01-01)
Jearanaitanakij, Kietikul
;
Choojan, Piyathida
;
Thongtem, Piyada
Sign language translation is a challenging problem in natural language processing. Its principle involves machine translation from sign language images to spoken language text. Designing a good translation is not a trivial task since there are a large number of both input image pixels and output classes. We propose the deep learning model to translate static gestures of Thai sign language (TSL) to the corresponding Thai spoken words. The main objective is to design a compact model that delivers high performance so that it can be implemented on mobile devices. Several mobile convolutional neural networks (CNN) are investigated to find the best backbone architecture. We also attach additional layers to the selected CNN architecture to fine-tune its performance. The experiments on the dataset collected from twenty-four volunteers indicate excellent results; in terms of precision, recall, and f1-score, of the proposed model. The comparisons with the state-of-the-art models and the feature visualizations from convolution layers endorse its effectiveness.
Some of the metrics are blocked by your
consent settings
Deep Learning Aided Robust RSRP Prediction in Cellular Networks
(2024-01-01)
Wongphatcharatham, Tanutsorn
;
Phakphisut, Watid
;
Jaruvitayakovit, Tanun
;
Boonkajay, Amnart
;
Huang, Jiajia
We propose a transfer learning enhanced hybrid model for robust reference signal received power (RSRP) prediction. The hybrid model comprises an expected RSRP estimation based on transmit power, 3-D antenna gain models, path loss, and a deep learning (DL) for predicting an error from ground-truth measurement. The DL architecture consists of regression neural network (NN) and convolutional neural network (CNN). Besides cell site configuration and the long-term evolution (LTE) measurement report from user equipments (UEs), the expected RSRP and geospatial data e.g. building percentage and clutter index are considered. Since trained model may not perform well in new environment, it requires tedious work and long time to collect data at a new cell site. Therefore, we use transfer learning (TL) to apply the trained model to the other areas, which have differences in environment information and antenna configurations, by transferring the knowledge acquired from trained model. The results of the trained area show that root mean square error (RMSE) and mean absolute error (MAE) are approximately 2.92 and 2.01, respectively. For the other area, TL have improved MAE approximately 1 to 2.
Some of the metrics are blocked by your
consent settings
Spacing and Stressing Extracting of Sentences System
(2024-01-01)
Ngodngamjaras, Peerach
;
Teethawatthanakorn, Napatrapee
;
Anuntachai, Anuntapat
Word stressing and spacing to convey the meaning is an important part of speaking English. Audiences are more likely to engage with a speaker who uses proper word stress and spacing. Stress can be observed from louder pronunciation, and higher frequency. Space can be observed from breathing and the length of time before starting a new sentence. These things, if those who would like to practice English are inexperienced or do not listen to native speakers often, they may not be able to remember the points of stressing and spacing correctly. As a result, speech does not flow smoothly. Nowadays, there are no tutorials or visualizations that clearly point out where the spacing is. For this reason, the organizer decided to create a program that would identify word stress and spacing to display on a dashboard that would be developed into a website in the future.
Some of the metrics are blocked by your
consent settings
Deep Learning Based Automobile Identification Application
(2023-01-01)
Chantamit-o-Pas, Pattanapong
;
Sangaroon, Pattanayu
;
Srisura, Jukkapat
Today, the high competition among domestic automobile manufacturers is intense situation than previous years. This result gives advantages in a good variety of brands, models, engine sizes and appearances. This can cause some critical issues in recognizing and recalling a car by manufacturer. In addition, an owner may modify some parts of original vehicle such as the head bumper, the rear bumper, and the head light. This modification also affects the people who are looking for pre-owned cars. Despite the fact, the details are mismatch with the vehicle registration book that issued by the Department of Land Transport. From this incident, the researchers implemented a convolutional neural network (CNN) in the identification of vehicle characteristics to reduce the ambiguity for each car’s models. The researchers conducted experiments using five algorithms. SVM, ResNet34, ResNet50 and Inception-ResNetV2. The researchers set up a library of two car models, Toyota Hilux and Honda Civic sedan and Civic Hatchback, including models from past ten years ago until the present. The images are of 224 × 224 pixels. The data are categorized into two sets, a training set has 1,449 images which is counted as 80% of total images and a testing set is having 362 images which is about 20% of total. The total images are 1,811 and 26 Classes. Our experiments compared the accuracies of SVM, ResNet34, ResNet50, and Inception-ResNetV2, which came out to be 21.4%, 55.5%, 66.6%, and 92.8% respectively. As a result, Inception-ResNetV2 outperforms among all other methods.
Some of the metrics are blocked by your
consent settings
Reference Signal Received Power Prediction Using Convolutional Neural Network with Residual Loss
(2023-01-01)
Ngenjaroendee, Thearrawit
;
Phakphisut, Watid
;
Wijitpornchai, Thongchai
;
Areeprayoonkij, Poonlarp
;
Jaruvitayakovit, Tanun
In this paper, LTE measurement reports collected from user equipments are used to generate the residual loss, which can represent the loss value of each grid. The residual loss and geospatial data are used in the learning process of convolutional neural network (CNN). We also use the site configuration and three-dimensional antenna pattern. Thus, the neural network and convolutional neural network are proposed to construct deep learning to predict the reference signal received power (RSRP) in Bangkok, Thailand. The results show that residual loss can improve the efficiency of prediction.
Some of the metrics are blocked by your
consent settings
Comparison of image enhancement techniques and CNN models for COVID-19 classification using chest x-rays images
(2022-01-01)
Kanjanasurat, Isoon
;
Domepananakorn, Nontacha
;
Archevapanich, Tuanjai
;
Purahong, Boonchana
This paper compares two image enhancement techniques with five convolutional neural network (CNN) models to classify Covid-19 chest x-ray images. a contrast limited adaptive histogram (CLAHE) and gamma correction which is method to improve image histogram are compared with the original chest x-ray image. We use five publicly available pre-trained CNN models to detect COVID-19: MobileNet, MobileNetV2, DenseNet169, DenseNet201, and ResNet50V2. Our procedure was validated using the COVID-19 radiography database, which is a freely accessible resource. MoblileNet with gamma correction is well-suited for COVIC-19 classification, achieving an accuracy score of 87.53 percent on the first epoch and 95.46 percent after training 100 epochs with the shortest computation time.
Some of the metrics are blocked by your
consent settings
Thai Recipe Retrieval Application Using Convolutional Neural Network
(2022-01-01)
Phophan, Thitiwut
;
Khuthanon, Rungwaraporn
;
Chantamit-o-Pas, Pattanapong
Due to the COVID-19, self-catering captured the interest of many people. This paper proposes a novel mobile application, which can share recipes and recognition material to help individuals with low prior cooking skill. It offers good, practical knowledge and can help to build cooperative teams in the cooking community among novice cooks. Choosing the ingredients for cooking can be difficult. This is especially true because of Thai vegetables look similar such as white and sweet basil particularly for new cooks not familiar with their other characteristics. This research introduces a mobile application, Kin Rai Dee App, which is based on sharing recipes and recognition material by using Roboflow with a pretrained model. To develop Thai vegetable image classification in our mobile application, the Convolutional Neural Network technique and a Thai vegetable dataset is used to evaluate the performance of our classification model. This dataset is composed of two sources including (1) Thai herb dataset from Kaggle website and (2) our own images. Therefore, there are totally 12 classes in the Thai vegetable dataset with image’s resolutions of 224 × 224 pixels. The result for image training is implemented through machine learning and Roboflow methods. The experiments process has training results accuracy at 85% and testing result at 15% in both models. The performance of our model has proven that it can achieve the result with confidence values 100% and 99.21% for specific Thai vegetables.
Some of the metrics are blocked by your
consent settings
Low resolution image area classifier based on Convolutional Neural Network
(2021-05-19)
Ngernplubpla, Jaturon
;
Warunsin, Kulwarun
;
Chitsobhuk, Orachat
Deep learning techniques are widely implemented in computer vision applications. The Convolutional Neural Networks (CNN) is a deep learning class that is the most effective in categorizing the statistical characteristics of images. It is often a challenging task to classify the frequency level region in various low-resolution image. In this research, we proposed the CNN for classification of gradient profile priors by learning on several gradient characteristics such as horizontal gradient acceleration, vertical gradient acceleration, the Relational Gradient Direction and Edge Sketch Image. This technique is used multiple building blocks to designed features through backpropagation with automatic and adaptive spatial hierarchies learning. The performance comparison was improved in classification of the frequency level area in various low-resolution image input that was illustrated in the experimental results which evaluate with several predictive and conventional classification techniques.
Some of the metrics are blocked by your
consent settings
Analyze traffic conditions and events with sound processing
(2020-10-13)
Anuntachai, Anuntapat
;
Pavaranchanakul, Kittituch
Traffic monitoring is a quiddity to understand occurrence of road transportation, for example, route decision in each festival, driving behavior, passerby quantity, traffic density and road contamination from road traffic, especially the accidents. Accident randomly happen in anytime and it dominates the high portion cause of death in each year. Accident is a cause of traffic jam. The assistance may come lately because no one cannot be noticed this event immediately. The traffic monitoring by applying sound analysis processing for accident sound with Convolutional Neural Network (CNNs) can significantly improve efficiency of traffic monitoring.
Some of the metrics are blocked by your
consent settings
Automatic labeling for thai news articles based on vector representation of documents
(2018-08-13)
Jirasirilerd, Wiphada
;
Tangtisanon, Pikulkaew
Nowadays, the most powerful news source in the world comes from online media on the Internet. The information comes from the SNS, video clips, audio clips or various news websites. In this competitive world, many news websites are mainly focused on publishing their contents to the website as fast as they can without taking time to label them correctly. This leads to a problem where readers cannot find news that they are interested in from a large amount of information on the website. In this paper, we propose a method to automatically label articles on the Thai language website using distributed representation of documents. The semantic similar words are extracted from paragraph vectors of each category of news and assign them as labels. We apply the convolutional neural network with binary classification approach to separate words from sentences and the result of the experiments indicated that our method can be applied to automatically label Thai news article effectively.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results