Activiteit

Using simulation results, this paper explores the performance of turbo codes, low-density parity-check (LDPC) codes, and polar codes on an additive white Gaussian noise (AWGN) channel in the context of intersymbol interference, a factor affecting the initial signal. By using an equalizer at the receiver, the adverse effects of inter-symbol interference (ISI) were eliminated. In practice, two kinds of equalizers were employed: zero-forcing (ZF) and minimum mean square error (MMSE). This analysis considers scenarios with perfect channel estimation, as well as estimation using the least squares method. The signal-to-noise ratio, influencing modifications to the bit error rate, was used for performance measurement; the MMSE equalizer’s performance exceeded that of the ZF equalizer in this situation. Previous channel equalization methodologies, while established, haven’t extensively addressed equalization strategies within the realm of turbo codes, particularly LDPC and polar codes for channel encoding. This research, in this regard, constitutes a contribution to the field of digital communications.

According to collected data, roughly 60 to 70 percent of the world’s population encounters low back pain (LBP) sometime during their life span, usually occurring in their young to middle-aged years. Individuals experiencing these effects face a heightened risk of diminished quality of life, increased absenteeism from work, and escalated medical expenses. We propose a new rehabilitation method that facilitates ongoing data gathering and analysis, enabling a more personalized treatment strategy. Dynamic spine correction (DSC) of the lumbar spine (L1-L5) during torso movement is evaluated by this method, designed for postural neuromuscular re-education in low back pain. Spinal mobility evaluation was conducted on 54 patients, all aged between 18 and 40 years and without low back pain. Measurements of the DSC device’s rotational position were taken by using 12-bit rotary position sensors, specifically the AS5304. Exercise-related lumbar spine rotation exhibited a statistically significant (p < 0.0001) difference, showing a higher mean rotation to the right (478 ± 224) than to the left (299 ± 144). In a similar vein, the greatest degree of rightward rotation was higher than the leftward rotation (1135 333 versus 742 144; p < 0.00001). The study’s findings, in terms of measurements, can be instrumental in future therapeutic applications of the device.

A laser three-dimensional (3D) projection system, an auxiliary tool, is employed in intelligent manufacturing. The practical deployment of this system is achieved through a positioning system. This study proposes a laser 3D projection system calibration method, facilitated by a binocular vision approach. The significance of binocular vision’s positioning function in calibration procedures was investigated. Two innovative methods for calibrating laser-based 3D projection systems were suggested, each dependent upon the binocular vision approach for precise positioning. A strategy involving simplified calculation models exists, and another utilizes data to derive the conversion relationship. The conversion relationship between the components of the projection system was determined directly through experimental calibration using data. In the experiment, the simplicity of the proposed calibration method was impressively demonstrated. The system using binocular vision, a 3D laser projection system, saw a reduction in calculation time. The average calibration error at a working distance between 18 and 22 meters was found to be 0.38 millimeters.

While face recognition technology is currently implemented in industrial settings, it still faces obstacles, including accurate verification and identification across varying angles and positions. The application of facial recognition to surveillance videos utilizing mugshots from diverse viewpoints, complementing traditional frontal and profile images collected by law enforcement, requires more in-depth research. In an effort to alleviate the shortage of databases for the study of this problem, we introduce the Face Recognition from Mugshots Database (FRMDB). This documentation set comprises 28 mugshots and 5 surveillance videos, captured from various perspectives, featuring 39 different individuals. Evaluating the impact of employing mugshots from various viewpoints on face recognition in surveillance video frames is the intended function of the FRMDB. To gain a first impression of the FRMDB’s performance, we ran accuracy tests using VGG16 and ResNet50 CNNs, pre-trained on the VGGFace and VGGFace2 datasets to extract facial features. We evaluated our data against the findings of the Surveillance Cameras Face Database (SCFace), a comparable dataset featured in the related literature. The outcomes, while illustrating the attributes of the proposed database, highlight that the subset of mugshots, including the frontal picture and the right profile view, obtains the lowest accuracy score among the evaluated. Consequently, further investigation is warranted to determine the optimal quantity of mugshots for facial recognition analysis in surveillance video frames.

The recording of a coplanar XXY stage’s movement in this study employed a charge-coupled device (CCD) image feedback control system for visual recognition. The feedback loop for the stage’s location is established through image placement, and the image compensation parameter controls the adjustment of the stage’s position. The constrained image resolution led to an average positioning error of 6712 meters for the optimized control parameter, a root mean square error of 2802 meters, and a settling time of roughly 7 seconds. Long short-term memory (LSTM) deep learning models possess the capability to detect long-term dependencies and sequential state data, which is crucial for determining the subsequent control signal. In pursuit of improved positioning, an LSTM model was constructed for stage motion, employing a dial indicator accurate to 1 meter for recording XXY position information. Following the removal of the assistive dial indicator, a novel LSTM-based XXY feedback control system was subsequently implemented to mitigate positioning inaccuracies. Morphing control signals are determined not just by time, but also by the cyclical nature of the LSTM learning algorithm. Repeated back-and-forth, forward, and backward movements were performed in response to point-to-point commands. Experimental observations, using the LSTM model, produced an average positioning error of 2085 meters, along with a root mean square error of 2681 meters and a settling time of 202 seconds. LSTM-based control of the stage demonstrated higher control accuracy and a reduced settling time in comparison to the CCD imaging system, as determined by three positioning indices.

Smart vending machines, a kind of unmanned retail, are being transformed by the burgeoning fields of mobile payment, Internet of Things, and artificial intelligence, opening a new vista. Nonetheless, the scarcity of data gathered from vending machine activities does not support the creation of autonomous vending systems. The methodology of this paper involves using machine learning on limited datasets to pinpoint the spiral rack’s location, particularly the end of the spiral rack. This location is crucial in preventing product issues during dispensing within vending machines. To achieve this, we propose a k-means clustering approach for dividing small datasets that are unevenly distributed both in quantity and in attributes, a consequence of real-world limitations and design choices, and create a remarkably lightweight convolutional neural network (CNN) to serve as a classification model, enabling real-time applications. Our proposed approach of splitting the data and utilizing the CNN demonstrates a visually effective strategy. The resulting model’s resilience to product variations achieves a perfect 100% accuracy. A portable handheld device, using a single-board computer, is created and the trained model is incorporated to demonstrate the feasibility of a real-time application.

Progress in 3D shape acquisition methods over the past decades notwithstanding, these techniques still represent a substantial barrier for a wide range of 3D face-based applications, thereby prompting significant research. Beside that, advanced 2D data generation models rooted in deep networks face limitations in their direct application to 3D objects due to the inherent dimensionality differences between 2D and 3D data. Two novel sampling methods for 3D facial data are proposed in this work. These methods aim to represent 3D faces as matrix-like structures suitable for deep neural networks. (1) A geometric sampling method, exploiting intersections of iso-geodesic and radial curves, generates structured 3D face representations. (2) The second method is a depth-like map sampling technique derived from the average depth values of grid cells on the facial front. Unstructured 3D face models can be connected to powerful deep networks for unsupervised 3D face generation using the sampling methods above. The above-mentioned techniques effectively generate structured 3D facial representations, which are then adaptable for use in a Deep Convolution Generative Adversarial Network (DCGAN) for 3D face generation. This method produces superior 3D faces, exhibiting a variety of expressions. We verified the efficacy of our generative model by producing a substantial number of 3D faces with a range of expressions, all through the use of the two novel down-sampling techniques.

The synthesis of tungsten oxide thin films, with variations in thickness, crystallinity, and morphology, was achieved through a multi-step process encompassing electron-beam deposition, thermal annealing, and acid boiling. Optical sensing materials, comprising films with varied surface morphologies and coated with gold nanoparticles, were assessed for hydrogen detection. The film’s structural, morphological, and optical properties were determined via a multi-faceted analysis encompassing X-ray diffraction, scanning electron microscopy, ellipsometry, and UV-VIS spectroscopy. obinutuzumab inhibitor The hydrogen-air response was marked by good selectivity, outperforming other reducing agents like ammonia and carbon monoxide.