KMITL

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

Browse

Search Results

Now showing 1 - 10 of 24

Some of the metrics are blocked by your
consent settings
Deep Learning-Based Heritage Building Assessment with Spatial Context
(2025-01-01)
Rattanapitak, Wirat
;
Khwansuwan, Poon
;
Wangsiripitak, Somkiat
;
Sirikitsathian, Phatthanan
Rapid urbanization threatens architectural heritage in developing regions, where limited conservation experts cannot assess thousands of potentially valuable buildings before irreversible modifications occur. This paper presents an automated screening system for heritage building identification using deep learning and spatial analysis. The proposed framework employs a dual-stream architecture combining YOLOv8 object detectionwith SegFormer semantic segmentation to extract architectural features from building facade photographs. These visual features are integrated with Geographic Information System (GIS) data to incorporate spatial context, recognizing that heritage buildings often cluster in historically significant areas. A hybrid weighting mechanism balances data-driven feature importance (80%) with expert architectural knowledge (20%) to ensure cultural sensitivity. Experimental evaluation on 1,500 buildings in Roi Et Province, northeastern Thailand, demonstrates the system's effectiveness, achieving 87.6% classification accuracy while processing each building in approximately one second. In corporating spatial context improved performance by 6.4% over visual features alone. The transformer-based architecture proved particularly effective at identifying characteristic features such as paired windows and traditional wall patterns that distinguish heritage structures. This work provides a practical tool for large scale preliminary heritage assessment, enabling conservation authorities to efficiently allocate limited expert resources to high priority buildings while maintaining classification reliability suitable for initial screening purposes.
Some of the metrics are blocked by your
consent settings
Deep Ensemble Learning with YOLO and RESNET for Calamansi Ripeness Classification
(2025-01-01)
Komasatid, Kongrith
;
Wangsiripitak, Somkiat
Nowadays, traditional calamansi ripeness classification relies on visual inspection by agricultural workers, which is a labor-intensive task and unpractical for large-scale farms. To turn modernization, this study introduces a digital ripeness classification approach leveraging deep learning ensemble. The initial stage is designed to detect the whole lifecycle from inflorescence to fruit development, while the subsequent stage focuses on efficient ripeness classification. For rapid inference, YOLOv11 (You Look at Once) has been adopted. Additionally, RESNET (Residual Network) has been optimized to enhance the model's accuracy, precision, and recall. To confirm its effectiveness, the proposed model was evaluated against YOLO variants using a custom dataset in a real-world environment. Results indicate that the proposed approach has better performance than YOLOv11 and YOLOv12, achieving Precision, Recall, F1, and mAP @ 0.5 scores of 0.6035,0.7981,0.6872, and 0.623, respectively.
Some of the metrics are blocked by your
consent settings
Ensemble Modeling for Emotion Recognition Using a Frame Attention Network on Faces With and Without Mouth Occlusion
(2025-01-01)
Wangsiripitak, Somkiat
;
Suwannarat, Channarong
;
Tipchatchawalkhun, Punyapan
Facial Expression Recognition (FER) often struggles with partial occlusions, such as when a mask is present. Expressing emotion on the face while speaking is another challenge, as mouth movement conveys both words and emotion. FER models trained without these variations often fail to correctly classify unobserved patterns. Instead of (i) building a single model to learn these variations from the beginning or (ii) training three models, each specifically designed for one scenario (without occlusion, with mouth occlusion, and while speaking without occlusion), we create two models based on a frame attention network with two different datasets: one without any occlusion and the other with mouth occlusion. We then ensemble these models in the FER system, using the average of the scores from both models as the predicted expression. Experiments on the Ck+, Oulu-Casia, and Revdess datasets show that our proposed ensemble model, when tested on samples with no occlusion (regardless of whether the subject is speaking or not) and with mouth occlusion, achieved 52.83% accuracy, outperforming models trained specifically for faces without any occlusion and faces with mouth occlusion by 4.33% and 1.95%, respectively. It can also achieve approximately the same accuracy (0.05% higher) as the model trained on the dataset containing all variations, while saving the time required for retraining the model.
Some of the metrics are blocked by your
consent settings
The Applications of Deep Learning in ECG Classification for Disease Diagnosis: A Systematic Review and Meta-Data Analysis
(2024-08-31)
Khalid, Mudassar
;
Pluempitiwiriyawej, Charnchai
;
Wangsiripitak, Somkiat
;
Murtaza, Ghulam
;
Abdulkadhem, Abdulkadhem A.
The supremacy of deep learning in artificial intelligence (AI) contexts, including image and speech recognition, computer vision, and medical imaging, among others, has established it as AI’s dominant approach. Several studies have been conducted on the use of deep learning in physiological signals, especially in ECG signals, in recent years, but there has been a lack of comprehensive review on the use of deep learning in ECG for biometric systems. This review is divided into two main sections: it provides a comprehensive bibliographic review of deep learning for ECG classification towards assisting in disease diagnosis in the first part while presenting an overview of the field, pioneers, and landmark studies. The second part offers comprehensive information on the subject, starting with the mathematical background of deep learning algorithms, the ECG signal processing, and the function of the heart. Using a PRISMA framework, 309 research papers were initially identified through specified keywords. After applying inclusion criteria, 90 articles were retained for detailed analysis, excluding 24 documents based on exclusion criteria EC1 and the remainder due to EC2. Key findings reveal that deep learning models achieve an average accuracy improvement of 10-15% over traditional methods, with convolutional neural networks (CNNs) and recurrent neural networks (RNNs) demonstrating superior performance in capturing complex ECG patterns. Through ECG databases, deep learning algorithms, assessment frameworks, metrics, and code availability, this review designs a systematic view from different perspectives to highlight the trends, challenges, and opportunities of deep learning for ECG arrhythmia classification. This paper’s goal is to contribute to the knowledge of both new and experienced researchers and practitioners in the field so that they can learn and understand the various processes involved in ECG signal processing using deep learning.
Some of the metrics are blocked by your
consent settings
Vision-Based Human Movement Matching for Muay Thai Training Support System
(2024-01-01)
Wangsiripitak, Somkiat
;
Pipatbowornkul, Napong
;
Wedsuwan, Nawin
;
Wattanacheep, Bhattarabhorn
An algorithm is proposed to evaluate Muay Thai trainees based on body movement matching with the trainer, aiming to support Muay Thai training. Skeletons of the trainee in image sequences captured by a monocular camera are extracted by a pre-trained deep-learning model. They are then synchronized with those of the trainer by matching keyframes and subsequently other image frames between both sequences using dynamic time warping. In each matched frame, the skeleton lines (bones) are compared by measuring the angle deviation of each bone from the trainee to the corresponding bone of the trainer. The similarity score of each bone pair is computed based on a cosine similarity metric. Simultaneously, a predefined upper-value limit of the maximum expected difference in bone comparison is applied to enhance the difference in similarity scores between excellent and underperformed postures, making the evaluation results more distinctive. Each matched frame is then scored by weighted averaging of bone scores, allowing the evaluation to be adaptively focused on significant parts of the body according to movement types. The overall score of the sequence is an average over all frames. Experimental results demonstrate that the proposed scoring method is flexible for use with an off-the-shelf camera and adaptive to a marking scheme that often changes with training strategies and patterns. Prior matching of keyframes reduces the mismatching of similar but non-corresponding frames. The score is also normalized to penalize misoriented bones, resulting in even lower scores for trainees with less skill.
Some of the metrics are blocked by your
consent settings
Learning Extended Term Frequency-Inverse Document Frequency (TF-IDF++) for Depression Screening From Sentences in Thai Blog Post
(2023-01-01)
Khunruksa, Sahussawud
;
Wangsiripitak, Somkiat
This paper proposed the method of depression screening from a sentence in Thai blog posts. Three classifiers based on a decision tree, linear SVC, and logistic regression were used to create classification models; each learned from extended term frequency-inverse document frequency (TF-IDF++) which is a feature vector created from a term frequency-inverse document frequency (TF-IDF), part-of-speech, and statistics of sentences such as word counts of selected terms. Our experiments showed that the model based on logistic regression achieves the top average score with a precision of 78.32%, a recall of 78.26%, and an f1-score of 78.27%. The proposed method outperforms the Thai BERT model by 0.75%, 0.77%, and 0.76%, respectively. Our investigation also showed that excessive confidence in the Thai BERT model tends to classify a sample with high probability. This also happens in case of an incorrect prediction; the error in such a case becomes noticeably higher than that of the wrong prediction in our proposed logistic regression-based model.
Some of the metrics are blocked by your
consent settings
Multimodal Biometrics Recognition Using a Deep Convolutional Neural Network with Transfer Learning in Surveillance Videos
(2022-07-01)
Aung, Hsu Mon Lei
;
Pluempitiwiriyawej, Charnchai
;
Hamamoto, Kazuhiko
;
Wangsiripitak, Somkiat
Biometric recognition is a critical task in security control systems. Although the face has long been widely accepted as a practical biometric for human recognition, it can be easily stolen and imitated. Moreover, in video surveillance, it is a challenge to obtain reliable facial information from an image taken at a long distance with a low-resolution camera. Gait, on the other hand, has been recently used for human recognition because gait is not easy to replicate, and reliable information can be obtained from a low-resolution camera at a long distance. However, the gait biometric alone still has constraints due to its intrinsic factors. In this paper, we propose a multimodal biometrics system by combining information from both the face and gait. Our proposed system uses a deep convolutional neural network with transfer learning. Our proposed network model learns discriminative spatiotemporal features from gait and facial features from face images. The two extracted features are fused into a common feature space at the feature level. This study conducted experiments on the publicly available CASIA-B gait and Extended Yale-B databases and a dataset of walking videos of 25 users. The proposed model achieves a 97.3 percent classification accuracy with an F1 score of 0.97and an equal error rate (EER) of 0.004.
Some of the metrics are blocked by your
consent settings
Traffic Light and Crosswalk Detection and Localization Using Vehicular Camera
(2022-01-01)
Wangsiripitak, Somkiat
;
Hano, Keisuke
;
Kuchii, Shigeru
An improved convolutional neural network model for traffic light and crosswalk detection and localization using visual information from a vehicular camera is proposed. Yolov4 darknet and its pretrained model are used in transfer learning using our datasets of traffic lights and crosswalks; the trained model is supposed to be used for red-light running detection of the preceding vehicle. Experimental results, compared to the result of the pretrained model learned only from the Microsoft COCO dataset, showed an improved performance of traffic light detection on our test images which were taken under various lighting conditions and interferences; 36.91% higher recall and 39.21% less false positive rate. The crosswalk, which is incapable of detection in the COCO model, could be detected with 93.37% recall and 7.74% false-positive rate.
Some of the metrics are blocked by your
consent settings
Improvement of Text-Independent Speaker Verification Using Gender-like Feature
(2021-01-21)
Kiawjak, Pornprom
;
Wangsiripitak, Somkiat
;
Pasupa, Kitsuchart
Text-independent speaker verification is a task of verifying a speaker identity from a characteristic of voice. We proposed the combined deep Convolutional Neural Network (CNN) consisting of (i) the first CNN trained to achieve gender classification which is then used to create a gender-like embedding and (ii) the last CNN trained with one additional input, the gender-like feature (embedding) from the first, to classify each speaker. The classification layer of the last CNN is removed to allow the remaining combined deep CNN for one-shot learning and verification of unobserved speaker. Our proposed CNN could obtain better results compared to VGGVox (ResNet-50) by 0.40% of Equal Error Rate (EER) on average. Additionally, we investigated results based on the scenario that the gender is known; the evaluation was performed only on utterance pairs that comply with the scenario. The EER rate of such case that only gender of claimed identity is known is 0.52% lower than that of VGGVox (ResNet-50) on average of two genders. In a more specific situation that the gender of person making a claim is also known, two dedicated networks were retrained for female and male, in addition to our first network which was trained for both. It is interesting that, when compared to the first network, the female network achieved less EER on female-female verification, while the network dedicated for male performed worse. Nevertheless, our two dedicated networks outperformed VGGVox (ResNet-50) by 0.88% of EER.
Some of the metrics are blocked by your
consent settings
Human height estimation using visual geometry and feature learning
(2021-01-01)
Dokthurian, Siriporn
;
Rattanapitak, Wirat
;
Wangsiripitak, Somkiat
Many existing video surveillance systems use human characteristics like face, height, and gait to identify a person. This paper proposes a human height estimation approach using visual geometry and feature learning that makes an estimate from a video clip of a person. An experiment was conducted to evaluate the performance of the approach. The approach achieved an average percentage final height estimate of 100.59 % (actual height = 100%), better than a previously reported estimate of 98.8% in the literature achieved by another approach. A successful further development of this approach would directly benefit forensic science investigators.

KMITL

Browse

Filters

Settings

Sort By

Results per page

Search Results