Narrow Results

One application of computer vision is face recognition, which essentially involves the identification of visual patterns. Face recognition is a tool that we often employ for multimedia management, smart card applications, justice reform, and security. The goal of a face recognition system is to automatically identify faces in any image or video using a computer vision domain. There are several methods for the detection of faces from the video; still, the inaccurate detection and computational complexity degrades the recognition precision. Hence, an optimized hybrid deep learning is introduced for the recognition of pose-invariant faces from the image. The pose-invariant face recognition is employed using the proposed ResNet-152 integrated YOLO (Res-YOLONet), wherein the ResNet-152 and YOLOv5 are hybridized together to enhance the recognition accuracy with minimal computational complexity. Besides, the loss function optimization is devised using the proposed Enhanced Fennec Fox (EnFF) algorithm. The proposed EnFF algorithm is designed by integrating the adaptive weighting strategy within the conventional Fennec Fox algorithm for acquiring the global best solution. The loss function optimization using the EnFF enhances the recognition accuracy. The assessment of the proposed EnFF_Res-YOLONet based on Accuracy, Precision, Recall, and Specificity acquires the values of 96%, 94%, 97%, and 96.6%, respectively.

Are there early warnings of an impending financial crisis in China? Our analysis using the Kaminsky–Lizondo–Reinhart (KLR) signal approach reveals that the probability of China having a currency crisis in the 24 months to October 2017 could be increased assuming no remedial action by the authorities to avert an impending crisis. Notwithstanding the above, our analysis shows that nine out of 15 economic indicators are effective in predicting a currency crisis. Loss function of policymakers and evaluation of usefulness are then employed to verify their validity. The results show that bank deposits and M2/international reserves are the most powerful indicators.

The sample size based on the Linex loss function and Blinex loss function is studied in this paper, and the analytical solution of the optimal sample size is deduced on the assumption that the Linex loss function and the normal distribution exist. For the Blinex loss function, an accurate algorithm was presented to obtain the optimal sample size. Furthermore, the optimal sample size is obtained, respectively, by taking Poisson distribution and normal distribution as examples. Due to the wide application of Blinex function in reality, the algorithm presented in this paper has immediate applications.

In this paper, we investigate the problem of personalized ranking from implicit feedback (PRIF). It is a more common scenario (e.g. purchase history, click log and page visitation) in recommender systems. The training data are only binary in these problems, reflecting the users’ actions or inactions. One shortcoming of previous PRIF algorithms is noise sensitivity: outliers in training data might bring significant fluctuations in the training process and lead to inaccuracy of the algorithm. In this paper, we propose two robust PRIF algorithms to solve the noise sensitivity problem of existing PRIF algorithms by using the pairwise sigmoid and pairwise fidelity loss functions. These two pairwise loss functions are flexible and can easily be adopted by popular collaborative filtering models such as the matrix factorization (MF) model and the K-nearest-neighbor (KNN) model. A learning process based on stochastic gradient descent with bootstrap sampling is utilized for the optimization. Experiments are conducted on practical datasets containing noisy data points or outliers. Results demonstrate that the proposed algorithms outperform several state-of-the-art one class collaborative filtering (OCCF) algorithms on both the MF and KNN models over different evaluation metrics.

Pedestrian detection provides manager of a smart city with a great opportunity to manage their city effectively and automatically. Specifically, pedestrian detection technology can improve our secure environment and make our traffic more efficient. In this paper, all of our work both modification and improvement are made based on YOLO, which is a real-time Convolutional Neural Network detector. In our work, we extend YOLO’s original network structure, and also give a new definition of loss function to boost the performance for pedestrian detection, especially when the targets are small, and that is exactly what YOLO is not good at. In our experiment, the proposed model is tested on INRIA, UCF YouTube Action Data Set and Caltech Pedestrian Detection Benchmark. Experimental results indicate that after our modification and improvement, the revised YOLO network outperforms the original version and also is better than other solutions.

Many classification algorithms aim to minimize just their training error count; however, it is often desirable to minimize a more general cost metric, where distinct instances have different costs. In this paper, an instance-based cost-sensitive Bayesian consistent version of exponential loss function is proposed. Using the modified loss function, the derivation of instance-based cost-sensitive extensions of AdaBoost, RealBoost and GentleBoost are developed which are termed as ICSAdaBoost, ICSRealBoost and ICSGentleBoost, respectively. In this research, a new instance-based cost generation method is proposed instead of doing this expensive process by experts. Thus, each sample takes two cost values; a class cost and a sample cost. The first cost is equally assigned to all samples of each class while the second cost is generated according to the probability of each sample within its class probability density function. Experimental results of the proposed schemes imply 12% enhancement in terms of $F$-measure and 13% on cost-per-sample over a variety of UCI datasets, compared to the state-of-the-art methods. The significant priority of the proposed method is supported by applying the pair of $T$-tests to the results.

With the continuous development of autonomous driving technology, realizing high-precision road obstacle detection is crucial to ensure traffic safety and driving experience. However, traditional obstacle detection methods often perform poorly in complex driving scenarios, such as obstacle movement and occlusion. To cope with this problem, this study proposes a road obstacle detection method based on a two-stream convolutional neural network model, aiming to overcome the limitations of traditional methods in capturing spatiotemporal features and handling complex situations. Our research approach is based on the following innovations: First, we introduce a dual-stream convolutional neural network structure, where one stream is used to extract spatial features from the contour information of the obstacle frames, and the other stream extracts temporal features from the temporal stream information. This dual-stream structure can fully capture the appearance and dynamic information of the obstacles, thus improving detection accuracy. Second, we design a feature fusion module to fuse the two features to obtain richer obstacle features. In addition, we propose a new loss function, i.e. clustering loss, for better optimizing the model training process, reducing intra-class variation, and increasing inter-class differences, thus improving the generalization performance of the model. In the experimental section, we conducted extensive experimental analysis using Citycapes and BDD100K datasets. The experimental results show that our model achieves significant performance improvement compared to both the traditional convolutional neural network method and the YOLOv5 method in a variety of scenarios such as obstacle stationary, obstacle moving, and obstacle with occlusion. Specifically, our method improves the recognition rate up to 4.8% to 14.5% on the Citycapes dataset and 6.5% to 12.8% on the BDD100K dataset, respectively, under different scenarios. In addition, our model also exhibits more advantages on small datasets, showing higher generalization ability and robustness.

Wood lumber is widely used in the construction and furniture manufacturing industries. In order to solve the problems of poor recognition, low efficiency and few detection types of manual and traditional processing methods, this paper proposes a model SGM-YOLO for the detection of surface defects in wood lumber. The SGM-YOLO model references a new backbone feature network, SL-backbone, to enhance the model’s ability to detect defects of different sizes. And, a new GVE-neck layer structure will be proposed in this paper, which reduces the parameters as well as the accuracy. In addition, the Normalized Weighted Distance Loss (NWD) small target detection algorithm is combined with the MPDIOU boundary loss function to replace the original loss function to further enhance the small target detection capability. Experiments show that SGM-YOLO achieves an average recognition accuracy of 77.4% for wood lumber defects, compared with the original model YOLOv8, the mAP is improved by 3.8% and the FPS is improved by 4.4, while the number and size of parameters are reduced, which provides better detection of several defects that are difficult to be identified. The methods presented in this paper were also applied to the YOLOv5 model, yielding positive results, to confirm its generalizability. The results demonstrate the high application value of the SGM-YOLO model in the wood lumber processing and manufacturing industry’s final product inspection.

Head Pose Estimation (HPE) has a wide range of applications in computer vision, but still faces challenges: (1) Existing studies commonly use Euler angles or quaternions as pose labels, which may lead to discontinuity problems. (2) HPE does not effectively address regression via rotated matrices. (3) There is a low recognition rate in complex scenes, high computational requirements, etc. This paper presents an improved unconstrained HPE model to address these challenges. First, a rotation matrix form is introduced to solve the problem of unclear rotation labels. Second, a continuous 6D rotation matrix representation is used for efficient and robust direct regression. The RepVGG-A2 lightweight framework is used for feature extraction, and by adding a multi-level feature fusion module and a coordinate attention mechanism with residual connection, to improve the network’s ability to perceive contextual information and pay attention to features. The model’s accuracy was further improved by replacing the network activation function and improving the loss function. Experiments on the BIWI dataset 7:3 dividing the training and test sets show that the average absolute error of HPE for the proposed network model is 2.41. Trained on the dataset 300W_LP and tested on the AFLW2000 and BIWI datasets, the average absolute errors of HPE of the proposed network model are 4.34 and 3.93. The experimental results demonstrate that the improved network has better HPE performance.

In order to address the problems of traditional inpainting algorithm models, such as the inability to automatically identify the specific location of the area to be restored, the cost of inpainting and the difficulty of inpainting, and the problems of structural and texture discontinuity and poor model stability in deep learning-based image inpainting, this paper proposes an image inpainting based on a contextual coherent attention. This paper designs a network model based on generative adversarial networks. First, to improve the global semantic continuity and local semantic continuity of images in image inpainting, a contextual coherent attention layer is added to the network; second, to solve the problems of slow convergence and insufficient training stability of the model, a cross-entropy loss function is used; finally, the trained generator is used to repair images. The experimental results are compared using PSNR and SSIM metrics, compared with the traditional GAN model, our model has a 3.782dB improvement in peak signal-to-noise ratio and a 0.025% improvement in structural similarity. The experimental results show that the image inpainting method in this paper has better performance in terms of image edge processing, pixel continuity and overall image structure.

Pavement crack condition is a vitally important indicator for road maintenance and driving safety. However, due to the interference of complex environment, such as illumination, shadow and noise, the automatic crack detection solution cannot meet the requirements of accuracy and efficiency. In this paper, we present an extended version of U-Net framework, named MSK-UNet, for pavement crack to solve these challenging problems. Specifically, first, the U-shaped network structure is chosen as the framework to extract more hierarchical representation. Second, we introduce selective kernel (SK) units to replace U-Net’s standard convolution blocks for obtaining the receptive fields with distinct scales. Third, multi-scale input layer establishes an image pyramid to retain more image context information at the encoder stage. Finally, a hybrid loss function including generalized Dice loss with Focal loss is employed. In addition, a regularization term is defined to reduce the impact of imbalance between positive and negative samples. To evaluate the performance of our algorithm, some tests were conducted on DeepCrack dataset, AsphaltCrack300 dataset and Crack500 dataset. Experimental results show that our approach can detect various crack types with diverse conditions, obtains a better performance in precision, recall and $F_1$-score, with 97.43%, 96.95% and 97.01% precision values, 82.51%, 93.33% and 87.58% recall values and 95.33%, 99.24% and 98.55% $F_1$-score values, respectively.

Rising acceptance of machine learning driven decision support systems underscores the need for ensuring fairness for all stakeholders. This work proposes a novel approach to increase a Neural Network model’s fairness during the training phase. We offer a frame-work to create a family of diverse fairness enhancing regularization components that can be used in tandem with the widely accepted binary-cross-entropy based accuracy loss. We use Bias Parity Score (BPS), a metric that quantifies model bias with a single value, to build loss functions pertaining to different statistical measures — even for those that may not be developed yet. We analyze behavior and impact of the newly minted regularization components on bias. We explore their impact in the realm of recidivism and census-based adult income prediction. The results illustrate that apt fairness loss functions can mitigate bias without forsaking accuracy even for imbalanced datasets.

Steam coal is the blood of China industry. Forecasting steam coal prices accurately and reliably is of great significance to the stable development of China’s economy. For the predictive model of existing steam coal prices, it is difficult to dig the law of nonlinearity of power coal price data and with poor stability. To address the problems that steam coal price features are highly nonlinear and models lack robustness, Laplacian kernel–log hyperbolic loss–Ridge regression (LK-LC-Ridge-Ensemble) model is proposed, which uses ensemble learning model for steam coal price prediction. First, in each sliding window, two kinds of correlation coefficient are employed to identify the optimal time interval, while the optimal feature set is selected to reduce the data dimension. Second, the Laplace kernel functions are adopted for constructing kernel Ridge regression (LK-Ridge), which boosts the capacity to learn nonlinear laws; the logarithmic loss function is introduced to form the LK-LC-Ridge to enhance the robustness. Finally, the prediction results of each single regression models are utilized to build a results matrix that is input into the meta-model SVR for ensemble learning, which further develops the model performance. Empirical results from three typical steam coal price datasets indicate that the proposed ensemble strategy is reliable for the model performance enhancement. Furthermore, the proposed model outperforms all single primitive models including accuracy of prediction results and robustness of model. Grouping cross-comparison between the different models suggests that the proposed ensemble model is more accurate and robust for steam coal price forecasting.

To reach an optimal acceptance sampling decision for products, whose lifetimes are Burr type XII distribution, sampling plans are developed with a rebate warranty policy based on truncated censored data. The smallest sample size and acceptance number are determined to minimize the expected total cost, which consists of the test cost, experimental time cost, the cost of lot acceptance or rejection, and the warranty cost. A new method, which combines a simple empirical Bayesian method and the genetic algorithm (GA) method, named the EB-GA method, is proposed to estimate the unknown distribution parameter and hyper-parameters. The parameters of the GA are determined through using an optimal Taguchi design procedure to reduce the subjectivity of parameter determination. An algorithm is presented to implement the EB-GA method. The application of the proposed method is illustrated by an example. Monte Carlo simulation results show that the EB-GA method works well for parameter estimation in terms of small bias and mean square error.

The TLD-100H readout system performance under various radioactive I-131 exposure doses was optimized by four key factors via the revised Taguchi dynamic quality loss function. Taguchi dynamic analysis and the orthogonal array reorganizing the essential factors are crucial for the optimization of the thermoluminescent dosimeter (TLD) readout system given strict criteria of multiple irradiated environments and long-term exposure for calibrated TLDs. Accordingly, 96 TLD-100H chips were selected and randomly categorized into three batches with eight groups (four TLD chips in each group). Four factors, namely (1) initial temperature, (2) heating rate, (3) maximal temperature, and (4) TLD preheat time before reading were organized into eight combinations according to Taguchi suggestion, whereas each factor was preset at two levels. All 96 $(3\times 8\times 4=96)$ chips were put in three concentric circles with 30, 60, and 90 cm radii for 48 h, surrounding the radioactive 150mCi ($5.55\times 10^3$MBq) I-131 capsule and exposed to the cumulative doses of 88.2, 18.6, and 8.6mSv for the respective radii, accordingly. The TLD readings obtained from each group were analyzed to derive the sensitivity, coincidence, and reproducibility, then those were reorganized to draw four fish-bone-plots for the optimization. The optimal option for the TLD readout system implied the combination of A1 (a $135^{\circ }$C initial temperature), B1 (a $10^{\circ }$C/s heating rate), C1 (a $240^{\circ }$C maximal temperature), and D2 (a 15s preheat time), which was further verified by the follow-up measurements. The dominant factors were A (initial temperature) and B (heating rate), whereas C (maximal temperature) and D (preheat time) were minor and provided negligible contributions to the system performance optimization.

The dynamic recognition of human posture has very broad application prospects in fields such as human–computer interaction and virtual reality. A new method for dynamic recognition of human posture is proposed within the theoretical framework of deep learning. In our method, historical image information of human posture, current image information of human posture, and association information between each image are included as inputs in the deep learning process. Afterwards, the input information is formed into time series information and feature series information, which are then fused by the attention mechanism module. Finally, the dynamic recognition results of human posture are obtained through convolution operation. Experimental research was conducted on the AMASS dataset, and the results showed that our method can achieve better results in dynamic recognition of human posture, with both indicators superior to the other three methods. At the same time, our method has a fast convergence speed and the loss function remains low and continuously decreases during the deep learning process.

The landmark achievements of AlphaGo Zero have created great research interest into self-play in reinforcement learning. In self-play, Monte Carlo Tree Search (MCTS) is used to train a deep neural network, which is then used itself in tree searches. The training is governed by many hyper-parameters. There has been surprisingly little research on design choices for hyper-parameter values and loss functions, presumably because of the prohibitive computational cost to explore the parameter space. In this paper, we investigate 12 hyper-parameters in an AlphaZero-like self-play algorithm and evaluate how these parameters contribute to training. Through multi-objective analysis, we identify four important hyper-parameters to further assess. To start, we find surprising results where too much training can sometimes lead to lower performance. Our main result is that the number of self-play iterations subsumes MCTS-search simulations, game episodes and training epochs. As a consequence of our experiments, we provide recommendations on setting hyper-parameter values in self-play. The outer loop of self-play iterations should be emphasized, in favor of the inner loop. This means hyper-parameters for the inner loop, should be set to lower values. A secondary result of our experiments concerns the choice of optimization goals, for which we also provide recommendations.

Adhesively bonded joints showcase non-uniform stress distribution, along their length as the load is transferred through layers of dissimilar stiffness. For efficient transfer of loads, the peak interfacial shear stress is required to be engineered. In this study, inspired by electric pulses, the interphase modulus is modified according to square, sinusoidal and triangular pulses. The variation in peak stresses with increased number of pulses up to four is also investigated. The developed analytical model is solved for the interfacial shear stresses as well as the peel stresses, using energy functional approach, through MAPLE software. The abrupt changes in modulus in square pulse graded interphase are observed to create highest interfacial shear stresses among the considered grading profiles. Furthermore, the peak interfacial stresses are observed to increase with increased number of pulses. An effective elastic modulus parameter is defined to indicate the area under the modulus profile curve. The effective modulus is found to be gradually increasing with increase number of pulses in square graded interphase. Whereas, it is constant for sinusoidal- and triangular-graded interphases. A deep machine learning-based physics informed neural network model is developed to quickly solve the developed governing differential equations. Therefore, results from the machine leaning model are compared to the analytical results.

This paper introduces the concepts of off-line and on-line quality control. It goes on to explain in some detail off-line and on-line quality control techniques. It shows that, when applied properly, these techniques can generate the desired results in any manufacturing environment. The paper will also refer to statistical process control-related activities in the assembly plant of Motorola at Austin, Texas.

With the proliferation of multi-sensor remote sensing images, extraction of information has progressively concerned attention in recent years. It leads to great significant improvement in remote sensing areas for change detection techniques. Due to the advent of multi-spectral images in the change detection model, a change detection model named optimized convolutional neural network (CNN) framework is designed for monitoring land cover changes. This can assist to note the changes happening in the different periods. The proposed model is designed by integrating the normalization function and detection model by applying the deep learning (DL) algorithm. The actual experiment is carried out by taking the parameters accuracy and Kappa coefficient to show the effectiveness of the proposed model. The proposed experiment out performed well when compared to the existing mainstream techniques in multi-sensor images change detection. When compared with the existing DL techniques, the proposed design attained accuracy of 98.92% by possessing less loss function.