Gradient Information Research Articles

In-Depth Review of YOLOv1 to YOLOv10 Variants for Enhanced Photovoltaic Defect Detection

This review presents an investigation into the incremental advancements in the YOLO (You Only Look Once) architecture and its derivatives, with a specific focus on their pivotal contributions to improving quality inspection within the photovoltaic (PV) domain. YOLO’s single-stage approach to object detection has made it a preferred option due to its efficiency. The review unearths key drivers of success in each variant, from path aggregation networks to generalised efficient layer aggregation architectures and programmable gradient information, presented in the latest variant, YOLOv10, released in May 2024. Looking ahead, the review predicts a significant trend in future research, indicating a shift toward refining YOLO variants to tackle a wider array of PV fault scenarios. While current discussions mainly centre on micro-crack detection, there is an acknowledged opportunity for expansion. Researchers are expected to delve deeper into attention mechanisms within the YOLO architecture, recognising their potential to greatly enhance detection capabilities, particularly for subtle and intricate faults.

Efficient Adversarial Attack Based on Moment Estimation and Lookahead Gradient

Adversarial example generation is a technique that involves perturbing inputs with imperceptible noise to induce misclassifications in neural networks, serving as a means to assess the robustness of such models. Among the adversarial attack algorithms, momentum iterative fast gradient sign Method (MI-FGSM) and its variants constitute a class of highly effective offensive strategies, achieving near-perfect attack success rates in white-box settings. However, these methods’ use of sign activation functions severely compromises gradient information, which leads to low success rates in black-box attacks and results in large adversarial perturbations. In this paper, we introduce a novel adversarial attack algorithm, NA-FGTM. Our method employs the Tanh activation function instead of the sign which can accurately preserve gradient information. In addition, it utilizes the Adam optimization algorithm as well as the Nesterov acceleration, which is able to stabilize gradient update directions and expedite gradient convergence. Above all, the transferability of adversarial examples can be enhanced. Through integration with data augmentation techniques such as DIM, TIM, and SIM, NA-FGTM can further improve the efficacy of black-box attacks. Extensive experiments on the ImageNet dataset demonstrate that our method outperforms the state-of-the-art approaches in terms of black-box attack success rate and generates adversarial examples with smaller perturbations.

DMF2Net: Dynamic multi-level feature fusion network for heterogeneous remote sensing image change detection

With the rapid development of remote sensing data fusion technology, heterogeneous remote sensing image (HRSI) change detection (CD) has become a frontier field. The powerful nonlinear information processing capability of deep neural network provides the possibility for image domain conversion of HRSIs. However, the majority of existing methods rely on stacking network layers to extract deep high-level semantic features to accomplish the mutual conversion of image domains. The effective extraction and utilization of shallow features have not been fully considered. At the same time, the dependence between deep and shallow features has not been deeply explored. Therefore, these algorithms severely restrict the performance of CD. A new HRSI-CD network (DMF2Net) based on multi-level feature fusion is proposed in this paper to address these issues. DMF2Net uses central difference convolution to extract fine-grained features from shallow layers of images. It is capable of capturing the intrinsic detail features of image by aggregating intensity and gradient information. The dynamic multi-level feature fusion method is used to learn the fusion weights from the features, which are then used to guide the organic fusion of shallow and deep semantic features. This can preserve more position and detail features of the image, which can prevent the loss of small data during image conversion and enhance the model’s ability to detect subtle changes. On this basis, a new DMF2Net-based method for detecting change in bi-temporal HRSIs is proposed. We conducted extensive experiments on four publicly available HRSI-CD datasets. The experimental results showed that the proposed CD framework achieves significant improvement over the most advanced methods. The project file of the proposed framework will be obtained from https://github.com/cwlnnu/DMF2Net.

Internet street view image fusion method using convolutional neural network

The use of image fusion technology in the area of information processing is continuing to advance in depth thanks to ongoing hardware advancements and related research. An enhanced convolutional neural network approach is developed to fuse visible and infrared images, and image pre-processing is carried out utilising an image alignment method with edge detection in order to gain more accurate and trustworthy image information. The performance of the fast wavelet decomposition, convolutional neural network, and modified convolutional neural network techniques is compared and examined using four objective assessment criteria. The experimental findings demonstrated that the picture alignment was successful with an offset error of fewer than 3 pixels in the horizontal direction and an angle error of less than 0.3∘ in both directions. The revised convolutional neural network method increased the information entropy, mean gradient, standard deviation, and edge detection information by an average of 46.13%, 39.40%, 19.91%, and 3.72%. The runtime of the modified approach was lowered by 19.42% when compared to the convolutional neural network method, which enhanced the algorithm’s performance and boosted the effectiveness of picture fusion. The image fusion accuracy reached 98.61%, indicating that the method has better fusion performance and is of practical value for improving image fusion quality.

Adversarial Robustness Enhancement for Deep Learning-Based Soft Sensors: An Adversarial Training Strategy Using Historical Gradients and Domain Adaptation.

Despite their high prediction accuracy, deep learning-based soft sensor (DLSS) models face challenges related to adversarial robustness against malicious adversarial attacks, which hinder their widespread deployment and safe application. Although adversarial training is the primary method for enhancing adversarial robustness, existing adversarial-training-based defense methods often struggle with accurately estimating transfer gradients and avoiding adversarial robust overfitting. To address these issues, we propose a novel adversarial training approach, namely domain-adaptive adversarial training (DAAT). DAAT comprises two stages: historical gradient-based adversarial attack (HGAA) and domain-adaptive training. In the first stage, HGAA incorporates historical gradient information into the iterative process of generating adversarial samples. It considers gradient similarity between iterative steps to stabilize the updating direction, resulting in improved transfer gradient estimation and stronger adversarial samples. In the second stage, a soft sensor domain-adaptive training model is developed to learn common features from adversarial and original samples through domain-adaptive training, thereby avoiding excessive leaning toward either side and enhancing the adversarial robustness of DLSS without robust overfitting. To demonstrate the effectiveness of DAAT, a DLSS model for crystal quality variables in silicon single-crystal growth manufacturing processes is used as a case study. Through DAAT, the DLSS achieves a balance between defense against adversarial samples and prediction accuracy on normal samples to some extent, offering an effective approach for enhancing the adversarial robustness of DLSS.

A novel single-parameter continuously differentiable filled function for global optimization

ABSTRACT The filled function approach is an effective global optimization method. The properties of the filled function directly affect the efficiency of the algorithm and have been discussed by many scholars, resulting in various new forms of filled functions. However, most of existing filled functions are not continuously differentiable and have parameters that are difficult to adjust. To remedy this problem, a new continuously differentiable filled function with an easily adjustable parameter is constructed. According to the theoretical analysis of the filled function, a new easy-to-implement filled function algorithm is designed utilizing the efficient local search algorithm based on gradient information. Numerical experiments show that the proposed filled function algorithm is a feasible and effective global optimization algorithm with less computational cost.

Learning optimal integration of spatial and temporal information in noisy chemotaxis.

We investigate the boundary between chemotaxis driven by spatial estimation of gradients and chemotaxis driven by temporal estimation. While it is well known that spatial chemotaxis becomes disadvantageous for small organisms at high noise levels, it is unclear whether there is a discontinuous switch of optimal strategies or a continuous transition exists. Here, we employ deep reinforcement learning to study the possible integration of spatial and temporal information in an a priori unconstrained manner. We parameterize such a combined chemotactic policy by a recurrent neural network and evaluate it using a minimal theoretical model of a chemotactic cell. By comparing with constrained variants of the policy, we show that it converges to purely temporal and spatial strategies at small and large cell sizes, respectively. We find that the transition between the regimes is continuous, with the combined strategy outperforming in the transition region both the constrained variants as well as models that explicitly integrate spatial and temporal information. Finally, by utilizing the attribution method of integrated gradients, we show that the policy relies on a nontrivial combination of spatially and temporally derived gradient information in a ratio that varies dynamically during the chemotactic trajectories.

Massive MIMO secure beamforming design via manifold optimization combined with momentum

Secure beamforming with constant modulus has been regarded as a promising solution in enhancing the physical-layer security by intelligently designing the antenna phase of massive multiple-input multiple-output (MIMO) devices. The resulting problem is non-convex and NP-hard due to the constant modulus constraint (CMC). Existing methods mainly solve the problem by relaxing the constraints or the objective function, which inevitably introduces the relaxation error. We notice that the complex circle manifold (CCM), where complex numbers maintain a constant modulus, naturally satisfies the CMC. Further, considering the iteration points as high-dimensional particle points, the optimization process on the CCM can be regarded as the motion of the particle on the high-dimensional plane formed by the manifold. Based on this feature, we propose the manifold optimization method combining momentum (MOCM). Concretely, the original problem is first transformed into an unconstrained problem on the complex circle manifold. Then, the descent direction is derived to solve the problem by incorporating momentum into gradient information, which speeds up the convergence process. Simulation results show that the proposed method is 1 bps/Hz higher than existing works.

Water Flow Prediction Based on Improved Spatiotemporal Attention Mechanism of Long Short-Term Memory Network

The prediction of water plant flow should establish relationships between upstream and downstream hydrological stations, which is crucial for the early detection of flow anomalies. Long Short-Term Memory Networks (LSTMs) have been widely applied in hydrological time series forecasting. However, due to the highly nonlinear and dynamic nature of hydrological time series, as well as the intertwined coupling of data between multiple hydrological stations, the original LSTM models fail to simultaneously consider the spatiotemporal correlations among input sequences for flow prediction. To address this issue, we propose a novel flow prediction method based on the Spatiotemporal Attention LSTM (STA-LSTM) model. This model, based on an encoder–decoder architecture, integrates spatial attention mechanisms in the encoder to adaptively capture hydrological variables relevant to prediction. The decoder combines temporal attention mechanisms to better propagate gradient information and dynamically discover key encoder hidden states from all time steps within a window. Additionally, we construct an extended dataset, which preprocesses meteorological data with forward filling and rainfall encoding, and combines hydrological data from multiple neighboring pumping stations with external meteorological data to enhance the modeling capability of spatiotemporal relationships. In this paper, the actual production data of pumping stations and water plants along the East-to-West Water Diversion Project are taken as examples to verify the effectiveness of the model. Experimental results demonstrate that our STA-LSTM model can better capture spatiotemporal relationships, yielding improved prediction performance with a mean absolute error (MAE) of 3.57, a root mean square error (RMSE) of 4.61, and a mean absolute percentage error (MAPE) of 0.001. Additionally, our model achieved a 3.96% increase in R2 compared to the baseline model.

MSTG: Multi-Scale Transformer with Gradient for joint spatio-temporal enhancement

Despite the widespread availability of HDR (High Dynamic Range) display devices, the majority of video sources are still stored in SDR (Standard Dynamic Range) format, leading to an urgent need for UHD (Ultra High Definition) video reconstruction. A simple solution is to divide the task into two sub-tasks: video super-resolution (SR) and video Bit-Depth Enhancement (BDE). While some joint enhancement tasks led by SR have been explored in multiple dimensions, the quantitative dimension has often been overlooked. In this paper, we address the first joint task of BDE and SR and propose a succinct network called MSTG. We conduct a systematic analysis to explain how this effect can be achieved and overcome the core challenge arising from distortions exacerbated by the differing optimization directions of BDE and SR. This challenge is attributed to indistinguishable contours and detailed textures. To tackle this, we employ a dual-branch module that leverages both gradient and image information to discern between BDE and SR. Furthermore, we propose a novel model named Cross-scale Transformer (CTF) embedded with Cross-scale Attention (CSA). This model facilitates information aggregation while adaptively utilizing structural similarity features. It jointly extracts, aligns, and fuses frame features at multiple scales, forming the MST module. Experiments demonstrate that the proposed algorithm outperforms the cascaded video enhancement methods by large margins on both synthetic and realistic datasets.

Distributed Cooperative Quantum Learning for Discrete-Time Multiagent Source Exploration With Information Prompts.

In light of optimization theory and swarm evolutionary schemes, under multiple single-integrator mobile agents equipped with sensors and prompters, this article addresses a discrete-time multiagent source exploration problem with information prompts. Regarding information prompts as constraints on the unknown target, by virtue of penalty function skills (PFSs) and sequential unconstrained minimization techniques (SUMTs), the agents are driven toward the source under the guidance of the control strategy. In two cases of available and unavailable gradient information, a quantum potential well, an average optimal position estimator (AOPE), and a global optimal position estimator (GOPE) are introduced into swarm evolutionary schemes with a periodically oscillating weight, such that distributed cooperative quantum learning (DCQL) policy is proposed as a control strategy under communication restrictions, where AOPE and GOPE are developed relying on distributed consensus theory. In particular, when the gradient is unavailable, we put forth an adaptive generalized Bernstein neural network (AGBNN) to replace it based on excellent properties of Bernstein polynomials and adaptive approaches. Further, a performance analysis for the proposed policy is executed on the convergence and computational complexity, which ensures the accuracy and efficiency of the source exploration in theory. Ultimately, a simulation test is carried out, and the results validate the practicability and effectiveness of the offered method.

Stress concentration factor estimation for corroded steel plates using the surface gradient method

Pitting corrosion occurs in steel components as a stress-accumulating surface irregularity that is vulnerable to fatigue crack nucleation. In this study, stress distribution in sinusoidal surfaces was investigated to devise an estimation method for the stress concentration factor (SCF) fluctuations of corroded steel. An SCF equation for sinusoidal surfaces was developed from empirical data based on three parameters: wavelength, thickness, and position. For corroded surfaces, successive numerical differentiations can be performed on each point to determine the sinusoidal surface having a point sharing the same gradient information. Notably, the SCFs of corresponding points are comparable; SCF estimation is made feasible for irregular surfaces through sinusoidal surfaces. In two-dimensional sections, maximum error was below 5% at pits and 15% for the surface. In three-dimensional models, maximum error was below 20% at pits for wide-ranging corroded specimens. Therefore, the proposed method effectively estimates SCF for identifying areas at risk of fatigue cracking.

Reliability-Based Topology Optimization with a Proportional Topology for Reliability

This research proposes an efficient technique for reliability-based topology optimization (RBTO), which deals with uncertainty and employs proportional topology optimization (PTO) to achieve the optimal reliability structure. The recent technique, called proportional topology optimization for reliability (PTOr), uses Latin hypercube sampling (LHS) for uncertainty quantification. The difficulty of the double-loop nested problem in uncertainty quantification (UQ) with LHS can be alleviated by the power of PTO, enabling RBTO to be performed easily. The rigorous advantage of PTOr is its ability to accomplish topology optimization (TO) without gradient information, making it faster than TO with evolutionary algorithms. Particularly, for reliability-based topology design, evolutionary techniques often fail to achieve satisfactory results compared to gradient-based techniques. Unlike recent PTOr advancement, which enhances the RBTO performance, this achievement was previously unattainable. Test problems, including an aircraft pylon, reveal its performances. Furthermore, the proposed efficient framework facilitates easy integration with other uncertainty quantification techniques, increasing its performance in uncertainty quantification. Lastly, this research provides computer programs for the newcomer studying cutting-edge knowledge in engineering design, including UQ, TO, and RBTO, in a simple manner.

A global map of gullied hillslopes on Mars

The distribution of gullies on Mars has been used to support different modes of formation involving CO2, H2O or entirely dry processes. We mapped the extent of gullied hillslope using a global mosaic of ConTeXt camera (CTX) at 6 m/pixel and High Resolution Imaging Science Experiment (HiRISE) browse products at ∼3 m/pixel. Our new global catalog of gullied hillslopes allows for quantitative, area-based analysis of gullies both at global and regional scale, which was not possible with point-based or more localized previous datasets. We analyzed aspect and gradient information derived from MOLA topography for every mapped gullied hillslope. We have confirmed the trends identified in previous works, which are that the mean gradient of gullied hillslopes is lower than 30°, the angle of repose of dry material on Mars, and that the preferred aspect of gullied hillslopes is latitude-dependent, with a shift between pole-facing to equator-facing happening around 40° in both hemispheres. We established a hierarchy of factors that can explain the global distribution of gullied hillslopes: Latitude is the strictest control because gullies are constrained between 26°-83°S and 28°-76°N. The availability of steep slopes is a strong control over gully distribution, and we observed that gullied hillslopes aspect reflects the orientation of large regional reliefs, which implies that hillslope aspect is less of a factor than the hillslope gradient in explaining gully distribution. We analyzed the correlation of gully distribution with a subsurface water ice consistency map, and we found that gullies are preferentially distributed on areas with some ground ice rather than no ice at all. We identified for the first time a bias toward an east-facing trend for southern hemisphere gullies poleward of ∼40° of latitude. We also linked our gullied hillslopes dataset with a global crater morphometry dataset: 63% of the gullied area on Mars is found in craters, and we find that the area covered by gullies in a given crater is positively correlated with the depth and diameter ratio of this crater.Our observations point toward a complex interplay between local slope, insolation, and thermal properties of the substrate to explain the spatial distribution of the current population of gullies. Finally, we have assessed the very high resolution (sub-meter) imagery coverage of gullies to be 36% and that the repeated coverage reaches 21% of gullies, hence this would need expanding to better understand the spatial trends in active gullies.

Cross-centroid ripple pattern for facial expression recognition

AbstractIn this paper, we propose a new feature descriptor Cross-Centroid Ripple Pattern (CRIP) for facial expression recognition. CRIP encodes the transitional pattern of a facial expression by incorporating a cross-centroid relationship between two ripples located at radius r1 and r2 respectively. These ripples are generated by dividing the local neighborhood region into subregions. Thus, CRIP has the ability to preserve macro and microstructural variations in an extensive region, which enables it to deal with side views and spontaneous expressions. Furthermore, gradient information between cross centroid ripples provides strength to capture prominent edge features in active patches: eyes, nose, and mouth, that define the disparities between different facial expressions. Cross-centroid information also provides robustness to irregular illumination. Moreover, CRIP utilizes the averaging behavior of pixels at subregions that yields robustness to deal with noisy conditions. The performance of the proposed descriptor is evaluated on seven comprehensive expression datasets consisting of challenging conditions such as age, pose, ethnicity, and illumination variations. The experimental results show that our descriptor consistently achieved a better accuracy rate as compared to existing state-of-the-art approaches.

An Improved BGE-Adam Optimization Algorithm Based on Entropy Weighting and Adaptive Gradient Strategy

This paper introduces an enhanced variant of the Adam optimizer—the BGE-Adam optimization algorithm—that integrates three innovative technologies to augment the adaptability, convergence, and robustness of the original algorithm under various training conditions. Firstly, the BGE-Adam algorithm incorporates a dynamic β parameter adjustment mechanism that utilizes the rate of gradient variations to dynamically adjust the exponential decay rates of the first and second moment estimates (β1 and β2), the adjustment of β1 and β2 is symmetrical, which means that the rules that the algorithm considers when adjusting β1 and β2 are the same. This design helps to maintain the consistency and balance of the algorithm, allowing the optimization algorithm to adaptively capture the trending movements of gradients. Secondly, it estimates the direction of future gradients by a simple gradient prediction model, combining historic gradient information with the current gradient. Lastly, entropy weighting is integrated into the gradient update step. This strategy enhances the model’s exploratory nature by introducing a certain amount of noise, thereby improving its adaptability to complex loss surfaces. Experimental results on classical datasets, MNIST and CIFAR10, and gastrointestinal disease medical datasets demonstrate that the BGE-Adam algorithm has improved convergence and generalization capabilities. In particular, on the specific medical image gastrointestinal disease test dataset, the BGE-Adam optimization algorithm achieved an accuracy of 69.36%, a significant improvement over the 67.66% accuracy attained using the standard Adam algorithm; on the CIFAR10 test dataset, the accuracy of the BGE-Adam algorithm reached 71.4%, which is higher than the 70.65% accuracy of the Adam optimization algorithm; and on the MNIST dataset, the BGE-Adam algorithm’s accuracy was 99.34%, surpassing the Adam optimization algorithm’s accuracy of 99.23%. The BGE-Adam optimization algorithm exhibits better convergence and robustness. This research not only demonstrates the effectiveness of the combination of these three technologies but also provides new perspectives for the future development of deep learning optimization algorithms.

Blade fouling fault detection based on shaft orbit generative adversarial network

To address the challenges of accuracy and interpretability in mechanical fault detection models, this study proposes a shaft orbit generative adversarial network (SOGAN) and applies it to detect blade fouling faults. Variational autoencoder (VAE) is used as the foundational network architecture for extracting high-dimensional latent features from the shaft orbit images. Concurrently, the invariant moments of the shaft orbit images are extracted and embedded in a bypass within the generator, thereby enhancing the accuracy of fault detection. Two sets of real-world blade fouling fault data are collected and meticulously analyzed. The proposed SOGAN model demonstrates significant performance improvements, with average increases of 18.91%, 10.20%, and 26.79% in accuracy compared to the autoencoder, VAE, and GANomaly algorithms, respectively. The F1 scores for both the groups exceed 0.98. The data generated by the proposed SOGAN model exhibit a trend-wise correspondence with the finite element modeling data. In addition, the use of gradient information for the localization and visual analysis of anomalies dynamically tracks the spatial evolution of the rotor shaft orbit throughout its lifecycle. The data generation capability and interpretability of the proposed model can effectively support digital twin modeling and health management of rotating machinery.

Deep Convolutional Network Based on Attention Mechanism for Matching Optical and SAR Images

Abstract. Complex geometric distortions and nonlinear radiation differences between optical and synthetic aperture radar (SAR) images present challenges for the matching of sufficient and evenly distributed corresponding points. To address this problem, this paper proposes a deep convolutional network based on an attention mechanism for matching optical and SAR images. In order to obtain robust feature points, we employ phase consistency instead of image intensity and gradient information for feature detection. A deep convolutional network (DCN) is designed to extract high-level semantic features between optical and SAR images, providing robustness to geometric distortion and nonlinear radiation changes. Notably, incorporating multiple inverted residual structures in the DCN facilitates efficient extraction of local and global features, promoting feature reuse, and reducing the loss of key features. Furthermore, a dense feature fusion module based on coordinate attention is designed, focusing on the spatial positional information of effective features, integrating key features into deep descriptors to enhance the robustness of deep descriptors to nonlinear radiometric differences. A coarse-to-fine strategy is then employed to enhance accuracy by eliminating mismatches. Experimental results demonstrate that the proposed network performs better than the manually designed descriptors-based methods and the stateof- the-art deep learning networks in both matching effectiveness and accuracy. Specifically, the number of matches achieved is approximately 2 times greater than that of other methods, with a 10% improvement in F-measure.

A replica exchange preconditioned Crank-Nicolson Langevin dynamic MCMC method with multi-variance strategy for Bayesian inverse problems

The preconditioned Crank-Nicolson Markov Chain Monte Carlo (pCN-MCMC) method is an efficient approach for large-scale Bayesian inference problems. It is constructed by discretizing a stochastic partial differential equation (SPDE). When the gradient information is coupled in this SPDE, the corresponding SPDE is also referred to as the Langevin diffusion. In this work, we present the replica exchange preconditioned Crank-Nicolson Langevin dynamic (repCNLD) Monte Carlo method to speed up the convergence of the pCN-MCMC and tackle the challenge of multimodal distribution. We use the Crank-Nicolson scheme to discretize the repLD and set the preconditioned matrix as the covariance matrix of the Gaussian prior. The discretization error is analyzed. Moreover, we use the multi-variance strategy (m-repCNLD) to improve computational efficiency. Perturbations are introduced into the scheme due to the likelihood estimation error, leading to the biased swapping rate. To correct the bias, we provide an unbiased estimator derived from some assumptions on the forward solvers. The accuracy and efficiency of the proposed method are demonstrated by sampling Gaussian mixture distribution and solving PDE-based nonlinear inverse problems numerically.

Evolution-guided value iteration for optimal tracking control

In this article, an evolution-guided value iteration (EGVI) algorithm is established to address optimal tracking problems for nonlinear nonaffine systems. Conventional adaptive dynamic programming algorithms rely on gradient information to improve the policy, which adheres to the first order necessity condition. Nonetheless, these methods encounter limitations when gradient information is intricate or system dynamics lack differentiability. In response to this challenge, evolutionary computation is leveraged by EGVI to search for the optimal policy without requiring an action network. The competition within the policy population serves as the driving force for policy improvement. Therefore, EGVI can effectively handle complex and non-differentiable systems. Additionally, this innovative method has the potential to enhance exploration efficiency and bolster the robustness of algorithms due to its population-based characteristics. Furthermore, the convergence of the algorithm and the stability of the policy are investigated based on the EGVI framework. Finally, the effectiveness of the established method is comprehensively demonstrated through two simulation experiments.