Aircraft Fuselage Defect Detection using Deep Neural Networks

Malekzadeh, Touba; Abdollahzadeh, Milad; Nejati, Hossein; Cheung, Ngai-Man

doi:10.48550/arxiv.1712.09213

Cited by 8 publications

(13 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In a few-data regime, conventional learning methods mostly fail due to overfitting. Fine-tuning a pre-trained network [5,6,7,8] sometimes prevents overfitting but at the cost of computation [9]. Therefore, recent successful approaches tackle this problem by meta-learning [10].…”

Section: Related Workmentioning

confidence: 99%

Revisit Multimodal Meta-Learning through the Lens of Multi-Task Learning

Abdollahzadeh¹,

Malekzadeh²,

Cheung³

2021

Preprint

Self Cite

View full text Add to dashboard Cite

Multimodal meta-learning is a recent problem that extends conventional few-shot meta-learning by generalizing its setup to diverse multimodal task distributions. This setup makes a step towards mimicking how humans make use of a diverse set of prior skills to learn new skills. Previous work has achieved encouraging performance. In particular, in spite of the diversity of the multimodal tasks, previous work claims that a single meta-learner trained on a multimodal distribution can sometimes outperform multiple specialized meta-learners trained on individual unimodal distributions. The improvement is attributed to knowledge transfer between different modes of task distributions. However, there is no deep investigation to verify and understand the knowledge transfer between multimodal tasks. Our work makes two contributions to multimodal meta-learning. First, we propose a method to quantify knowledge transfer between tasks of different modes at a micro-level. Our quantitative, task-level analysis is inspired by the recent transference idea from multi-task learning. Second, inspired by hard parameter sharing in multi-task learning and a new interpretation of related work, we propose a new multimodal meta-learner that outperforms existing work by considerable margins. While the major focus is on multimodal meta-learning, our work also attempts to shed light on task interaction in conventional meta-learning. The code for this project is available at https://miladabd.github.io/KML.Multimodal Model-Agnostic Meta-Learning (MMAML) [1] proposes a framework to better handle multimodal task distributions and achieves encouraging performance. As one of the most intriguing findings, MMAML claims that a single meta-learner trained on a multimodal distribution can sometimes outperform multiple specialized meta-learners trained on individual unimodal distributions. This was observed in spite of the diversity of the multimodal tasks. In [1], this observation is attributed to knowledge transfer across different modes of multimodal task distribution.

show abstract

Section: Related Workmentioning

confidence: 99%

Revisit Multimodal Meta-Learning through the Lens of Multi-Task Learning

Abdollahzadeh¹,

Malekzadeh²,

Cheung³

2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…The key challenge of all the above automated techniques is developing defect detection algorithms that are able to perform with accuracy and repeatability. Several attempts have been made and most of them can be divided into the following two categories: the ones that use more traditional image processing techniques [5,[16][17][18] and the ones that use machine learning [19][20][21][22][23][24]. In the first category, image features such as convexity or signal intensity [5] are used.…”

Section: Introductionmentioning

confidence: 99%

“…These techniques have very good accuracy in the test data but are failing to effectively generalize and need continuous tuning. On the other hand, algorithms using convolutional neural networks (CNN) have showed good results in defect detection [19][20][21]25]. In [19,20], CNNs are used as feature extractors and then either a single shot multibox detector (SSD) or a support vector machine (SVM) are used for the classification.…”

Section: Introductionmentioning

confidence: 99%

“…On the other hand, algorithms using convolutional neural networks (CNN) have showed good results in defect detection [19][20][21]25]. In [19,20], CNNs are used as feature extractors and then either a single shot multibox detector (SSD) or a support vector machine (SVM) are used for the classification. The use of faster CNN is also proposed for classification and localization [22].…”

Section: Introductionmentioning

confidence: 99%

“…The existence of large datasets in concrete structures has allowed CNNs to show excellent results in defect detection in concrete structures. On the other hand, in aircraft structures, the results are promising but are still not very accurate [18] or they deal with only the problem of defect recognition [20]. In this paper, we propose a two-step classification process of an ensemble of machine learning classifiers for both defect recognition and classification.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Defects Recognition Algorithm Development from Visual UAV Inspections

Avdelidis

Tsourdos

Lafiosca

et al. 2022

Sensors

View full text Add to dashboard Cite

Aircraft maintenance plays a key role in the safety of air transport. One of its most significant procedures is the visual inspection of the aircraft skin for defects. This is mainly carried out manually and involves a high skilled human walking around the aircraft. It is very time consuming, costly, stressful and the outcome heavily depends on the skills of the inspector. In this paper, we propose a two-step process for automating the defect recognition and classification from visual images. The visual inspection can be carried out with the use of an unmanned aerial vehicle (UAV) carrying an image sensor to fully automate the procedure and eliminate any human error. With our proposed method in the first step, we perform the crucial part of recognizing the defect. If a defect is found, the image is fed to an ensemble of classifiers for identifying the type. The classifiers are a combination of different pretrained convolution neural network (CNN) models, which we retrained to fit our problem. For achieving our goal, we created our own dataset with defect images captured from aircrafts during inspection in TUI’s maintenance hangar. The images were preprocessed and used to train different pretrained CNNs with the use of transfer learning. We performed an initial training of 40 different CNN architectures to choose the ones that best fitted our dataset. Then, we chose the best four for fine tuning and further testing. For the first step of defect recognition, the DenseNet201 CNN architecture performed better, with an overall accuracy of 81.82%. For the second step for the defect classification, an ensemble of different CNN models was used. The results show that even with a very small dataset, we can reach an accuracy of around 82% in the defect recognition and even 100% for the classification of the categories of missing or damaged exterior paint and primer and dents.

show abstract

Fair Generative Models via Transfer Learning

Teo

Abdollahzadeh

Cheung

2023

AAAI

View full text Add to dashboard Cite

This work addresses fair generative models. Dataset biases have been a major cause of unfairness in deep generative models. Previous work had proposed to augment large, biased datasets with small, unbiased reference datasets. Under this setup, a weakly-supervised approach has been proposed, which achieves state-of-the-art quality and fairness in generated samples. In our work, based on this setup, we propose a simple yet effective approach. Specifically, first, we propose fairTL, a transfer learning approach to learn fair generative models. Under fairTL, we pre-train the generative model with the available large, biased datasets and subsequently adapt the model using the small, unbiased reference dataset. We find that our fairTL can learn expressive sample generation during pre-training, thanks to the large (biased) dataset. This knowledge is then transferred to the target model during adaptation, which also learns to capture the underlying fair distribution of the small reference dataset. Second, we propose fairTL++, where we introduce two additional innovations to improve upon fairTL: (i) multiple feedback and (ii) Linear-Probing followed by Fine-Tuning (LP-FT). Taking one step further, we consider an alternative, challenging setup when only a pre-trained (potentially biased) model is available but the dataset that was used to pre-train the model is inaccessible. We demonstrate that our proposed fairTL and fairTL++ remain very effective under this setup. We note that previous work requires access to the large, biased datasets and is incapable of handling this more challenging setup. Extensive experiments show that fairTL and fairTL++ achieve state-of-the-art in both quality and fairness of generated samples. The code and additional resources can be found at bearwithchris.github.io/fairTL/.

show abstract

Aircraft Fuselage Defect Detection using Deep Neural Networks

Cited by 8 publications

References 0 publications

Revisit Multimodal Meta-Learning through the Lens of Multi-Task Learning

Revisit Multimodal Meta-Learning through the Lens of Multi-Task Learning

Defects Recognition Algorithm Development from Visual UAV Inspections

Fair Generative Models via Transfer Learning

Contact Info

Product

Resources

About