Tong-Yuen Chai

Automating failure detection of infrastructure is essential to avoid unexpected downtime, which could impact both operations and user safety. However, frequent replacement of laminated rubber bearing (LRB) as a vital seismic isolation component can be inefficient and inspecting them for faults is often labor-intensive. This is a significant challenge in maintaining structural integrity, especially in critical infrastructure where continuous monitoring is necessary. Recent innovation in deep learning (DL) provides a promising alternative to traditional inspection methods, offering more efficient and accurate assessments. Hence, this paper explores the practical application of DL for detecting internal failure in laminated rubber bearings installed in structures. Neural network models, including a convolutional neural network (CNN), long-short term memory (LSTM), and their combinations (hybrid CNN-LSTM), were employed. An experimental setup was developed to simulate a bridge structure supported by down-scale LRB samples at its base. Samples with internal debonding failure were manufactured by reducing the bonding adhesive, to replicate failure conditions due to shear loading where the LRBs were forced to slide in an extreme condition. The vibration platform was actuated under different levels of frequency. Both Healthy and Faulty LRB conditions data were collected for 10 minutes each, which is adequate for 10 sets of data divided into training, validation, and testing with a fixed 6:2:2 ratio, respectively. Results revealed that CNN outperformed the other two models in average classification accuracy at 5Hz and 10Hz with 97.65% and 91.45%, respectively. Plus, CNN recorded the shortest training period among all models compared, with only 128 seconds at 15Hz, compared to 695 seconds and 1599 seconds owned by LSTM and hybrid CNN-LSTM respectively. In conclusion, neural networks have shown the capability in identifying LRB internal failure. CNN has the advantage in terms of both classification accuracy and training period compared to LSTM and hybrid CNN-LSTM models.

Biometric template protection (BTP) schemes are implemented to increase public confidence in biometric systems regarding data privacy and security in recent years. The introduction of BTP has naturally incurred loss of information for security, which leads to performance degradation at the matching stage. Although efforts are shown in the extended work of some iris BTP schemes to improve their recognition performance, there is still a lack of a generalized solution for this problem. In this paper, a trainable approach that requires no further modification on the protected iris biometric templates has been proposed. This approach consists of two strategies to generate a confidence matrix to reduce the performance degradation of iris BTP schemes. The proposed binary confidence matrix showed better performance in noisy iris data, whereas the probability confidence matrix showed better performance in iris databases with better image quality. In addition, our proposed scheme has also taken into consideration the potential effects in recognition performance, which are caused by the database-associated noise masks and the variation in biometric data types produced by different iris BTP schemes. The proposed scheme has reported remarkable improvement in our experiments with various publicly available iris research databases being tested.

Single-pixel camera is developed to mitigate the constraints faced by the conventional cameras especially in invisible wavelengths and low light conditions. Nyquist–Shannon theorem requires as many measurements as the image pixels to reconstruct images flawlessly. In practice, obtaining more measurements increases the cost and acquisition time, which are the major drawbacks of single-pixel imaging (SPI). Therefore, compressive sensing was proposed to enable image reconstruction with fewer measurements. We present a design of sensing patterns to obtain image information by utilizing spatially variant resolution (SVR) technique in SPI. The proposed method reduces the measurements by prioritizing the resolution in the region of interest (ROI). It successfully achieves the programmable imaging concept where multiresolution adaptively optimizes the balance between the image quality and the measurements number. Results show that SVR images can be reconstructed from significantly fewer measurements yet able to achieve better image quality than uniform resolution images. In addition, the SVR images can be further enhanced by integrating the dynamic supersampling technique. Consequently, the concerns of image quality, long acquisition, and processing time can be addressed. The proposed method potentially benefits imaging applications where the target ROI is prioritized over the background and most importantly it requires fewer measurements.

Conventional imaging sensors often reach their limits in terms of resolution and dynamic range. In addition, conventional imaging in invisible wavelengths is relatively more expensive and complicated. As an alternative, single-pixel cameras allow reduction of cost and complexity that would be otherwise required in a conventional multi-nixel camera. In terms of digital imaging, Nyquist-Shannon theorem states that to stably recover an image without introducing perceptible errors, the number of measurements and the number of image pixels are required to be at least the same. As the number of image pixels is ever increasing, increasing the number of measurements to fulfill Nyquist-Shannon theorem's requirements has become increasingly challenging. Since in many cases increasing the number of measurements means that the cost and time required are increasing accordingly as well. Therefore a mean to recover images from a number of measurements less than the number of pixels (sub-Nyquist measurements) is needed. The objective of this paper is to present and compare single-pixel imaging via compressive sensing/sampling (CS) and spatially-variant resolution (SVR) single-pixel imaging. Both methods are capable of recovering images stably from sub-Nyquist measurements. The measurements and reconstructions of images were done in simulations. SVR single-pixel imaging reduces the number of measurement by sacrificing the resolution of the peripheral regions. This realizes the programmable imaging concept where multi-resolution can be adaptively applied to optimize the balance between image quality and number of measurement. This could benefit some imaging applications where a target in the region of interest (the fovea) is prioritized over the rest of the region (such as the background).

Iris has been found to be unique and consistent over time despite its random nature. Unprotected biometric (iris) template raises concerns in security and privacy, as numerous large-scale iris recognition projects have been deployed worldwide—for instance, susceptibility to attacks, cumbersome renewability, and cross-matching. Template protection schemes from biometric cryptosystems and cancelable biometrics are expected to restore the confidence in biometrics regarding data privacy, given the great advancement in recent years. However, a majority of the biometric template protection schemes have uncertainties in guaranteeing criteria such as unlinkability, irreversibility, and revocability, while maintaining significant performance. Fuzzy commitment, a theoretically secure biometric key binding scheme, is vulnerable due to the inherent dependency of the biometric features and its reliance on error correction code (ECC). In this paper, an alignment-free and cancelable iris key binding scheme without ECC is proposed. The proposed system protects the binary biometric data, i.e., IrisCodes, from security and privacy attacks through a strong and size varying non-invertible cancelable transform. The proposed scheme provides flexibility in system storage and authentication speed via controllable hashed code length. We also proposed a fast key regeneration without either re-enrollment or constant storage of seeds. The experimental results and security analysis show the validity of the proposed scheme.

The objective of this paper is to propose a method to control a two-link planar manipulator using a sliding mode controller with higher order sliding surface. Earlier approaches of varying sliding surface are time dependent and this may not yield optimal performance as the rotating surface does not depend on the dynamic states of the system. In this paper, the proposed control law alters the sliding surface based on the error states to drive the trajectory approaching the sliding phase quicker. A new sliding manifold is established and tuned until error decreases to zero. Stability is ensured through Lyapunov theorem and the trajectory is driven towards a designed sliding surface. The performance of the proposed controller is evaluated and compared against the conventional sliding mode controller as well as the previous approach of rotating sliding surface controller. Results showed that the proposed controller improved the respond speed and shortened the reaching phase. Although the chattering phenomenon remains, it enhanced the flexibility to adapt to the variation of system settings (e.g. torque limit).

Iris feature is widely used in recognition due to its uniqueness and reliability among other biometrics. Iris pattern is generally complex containing randomness and distinctive features such as furrows, freckles and crypts. It is still an uncertainty how useful each of these features will be when it comes to recognition. In this paper, we formulate the recognition problem in a semi-naïve Bayesian classification framework to maintain simplicity and robustness. A nonhierarchical structure is adopted to classify random feature sets in different ferns. Each feature set is a new binary representation and returns the probability that it belongs to any of the classes learned during training. The proposed method can synthesize multiple views of the iris features from a training image as they would appear under different scales and perspectives. The outcomes are then combined in a semi-naïve Bayesian manner. This new approach has been evaluated on CASIA iris database and benchmarked with Random Forest classifier and Hamming distance. The proposed method shows shorter processing time and lower equal error rate compared to other methods.

The recent proposed Indexing-First-One (IFO) hashing is a latest technique that is particularly adopted for eye iris template protection, i.e. IrisCode. However, IFO employs the measure of Jaccard Similarity (JS) initiated from Min-hashing has yet been adequately discussed. In this paper, we explore the nature of JS in binary domain and further propose a mathematical formulation to generalize the usage of JS, which is subsequently verified by using CASIA v3-Interval iris database. Our study reveals that JS applied in IFO hashing is a generalized version in measure two input objects with respect to Min-Hashing where the coefficient of JS is equal to one. With this understanding, IFO hashing can propagate the useful properties of Min-hashing, i.e. similarity preservation, thus favorable for similarity searching or recognition in binary space. © (2017) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.