KMITL

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

Browse

Settings

Author: search.filters.author.Woraratpanya, KuntpongSubject: search.filters.subject.Deep learning

Search Results

Now showing 1 - 5 of 5
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    Evaluation of deep learning algorithms for semantic segmentation of car parts
    (2022-10-01)
    Pasupa, Kitsuchart
    ;
    Kittiworapanya, Phongsathorn
    ;
    Hongngern, Napasin
    ;
    Woraratpanya, Kuntpong
    Evaluation of car damages from an accident is one of the most important processes in the car insurance business. Currently, it still needs a manual examination of every basic part. It is expected that a smart device will be able to do this evaluation more efficiently in the future. In this study, we evaluated and compared five deep learning algorithms for semantic segmentation of car parts. The baseline reference algorithm was Mask R-CNN, and the other algorithms were HTC, CBNet, PANet, and GCNet. Runs of instance segmentation were conducted with those five algorithms. HTC with ResNet-50 was the best algorithm for instance segmentation on various kinds of cars such as sedans, trucks, and SUVs. It achieved a mean average precision at 55.2 on our original data set, that assigned different labels to the left and right sides and 59.1 when a single label was assigned to both sides. In addition, the models from every algorithm were tested for robustness, by running them on images of parts, in a real environment with various weather conditions, including snow, frost, fog and various lighting conditions. GCNet was the most robust; it achieved a mean performance under corruption, mPC = 35.2, and a relative degradation of performance on corrupted data, compared to clean data (rPC), of 64.4%, when left and right sides were assigned different labels, and mPC = 38.1 and rPC = 69.6 % when left- and right-side parts were considered the same part. The findings from this study may directly benefit developers of automated car damage evaluation system in their quest for the best design.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    A data generation framework for extremely rare case signals
    (2021-08-01)
    Chalongvorachai, Thasorn
    ;
    Woraratpanya, Kuntpong
    Unlike data augmentation, data generation for extremely rare cases is an approach that can spawn a significant number of high-quality samples based on very few original data. This could be useful in anomaly detection and classification tasks that have the limitation of publicly available datasets for research purposes. Though some other approaches have attempted to solve this problem, such as data augmentation techniques, there was nothing to ensure the characteristics of synthesized samples. Previously, we initiated a framework, called Data Augmentation and Generation for Anomalous Time-series Signals (DAGAT), that was in cooperation with important components: Data Augmentation, Variational Autoencoder (VAE), Data Picker (DP), Signal Fragment Assembler (SFA), and Quality Classifier (QC). And then, an upgraded framework, called An Advanced Data Generation for Anomalous Signals (ADGAS), was introduced to eliminate the limitations of DAGAT; those are uncontrollable outputs and the possibility of bad data included in a training set. By reforming DAGAT architecture, ADGAS achieves a better outcome of generated samples. Nonetheless, ADGAS could be improved through better SFA, DP, and QC. Hence, this paper proposed a Data Generation Framework for Extremely Rare Case Signals. The proposed framework is achievable in generating reliable data for various objectives. We challenged this framework by using the 1D-CNN to serve as the performance evaluator in multi-class anomalous classifications and using the water treatment and water distribution testbed (SWaT and WADI) as the real-world anomaly datasets. The result shows that it surpasses other baseline methods of anomaly data augmentation and data generation techniques.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    ADGAS: An Advanced Data Generation for Anomalous Signals
    (2021-01-01)
    Chalongvorachai, Thasorn
    ;
    Woraratpanya, Kuntpong
    Anomaly detection using deep learning approaches is still challenging, especially in the case of data limitations. A small number of samples for training deep learning models typically result in poor performance of detection and classification. Previously, data augmentation was one of the methods used to solve this problem. The data augmentation with rotation, permutation, time warping, and their combination can increase the performance of anomaly classification. However, this method is still limited and does not guarantee that the generated output data have adequate varieties and keep original characteristics of data. Our recent work, data augmentation and generation for anomalous time series signals (DAGAT) was proposed to expand the space of possible augmented data by implementing vanilla augmentation on various domains in conjunction with variational autoencoder (VAE). Nonetheless, the DAGAT still has barriers, which are an uncontrollable number of target results, a missed opportunity of integrating multiple augmentation characteristics in latent space, and a possibility of including any bad data for training in VAE. To overcome these limitations, this paper proposed an advanced data generation for anomalous signals (ADGAS). By focusing on the quality of generated data, one more quality classifier (QC) was added as a prepossessing step of VAE. In this way, the experimental results showed that convolutional neural networks (CNNs), used as a performance tester, trained with the generated datasets of ADGAS achieved better accuracy in classifying anomalous events when compared to models trained with a combination of rotation, permutation, and time warping data augmentation methods.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    3DVAE-ERSG: 3D Variational Autoencoder for Extremely Rare Signal Generation
    (2021-01-01)
    Chalongvorachai, Thasorn
    ;
    Woraratpanya, Kuntpong
    Data generation is not data augmentation. Our data generation is a new technique that can synthesize a dataset from a very small number of samples and ensure the quality of its outputs. Recently, this concept has been proposed and applied in a framework called Data Generation Framework for Extremely Rare Case Signals (DGERS) to solve a problem of a limited number of anomaly signals. With the power of DGERS consisting of principal components, including various data augmentation techniques on diverse domains, Signal Fragment Assembler (SFA), Variational Autoencoder (VAE), Data Picker (DP), and Quality Classifier (QC), the generated dataset had the good quality, when evaluated with a performance tester. Nevertheless, the DGERS has not used the full potential of VAE yet. The previous framework used the VAE latent space in only two dimensions. To use a higher potential of VAE, this paper proposed a 3D Variational Autoencoder for Extremely Rare Signal Generation (3DVAE-ERSG). This method increases the dimension of the latent space from 2D to 3D. We also proposed the 3D Data Picker for data exploration. To test this hypothesis, we experimented with the same datasets that the DGERS method was tested before. The results show that our 3DVAE-ERSG can outperform the baseline in most cases.
  • Loading...
    Thumbnail Image
    Some of the metrics are blocked by your 
    consent settings
    Item type:Item,
    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.