Search Results (40,554)

Deploying modern neural networks on resource-constrained edge devices necessitates a series of optimizations to ready them for production. These optimizations typically involve pruning, quantization, and fixed-point conversion to compress the model size and enhance energy efficiency. While these optimizations are generally adequate for most edge devices, there exists potential for further improving the energy efficiency by leveraging special-purpose hardware and unconventional computing paradigms. In this study, we explore stochastic computing neural networks and their impact on quantization and overall performance concerning weight distributions. When arithmetic operations such as addition and multiplication are executed by stochastic computing hardware, the arithmetic error may significantly increase, leading to a diminished overall accuracy. To bridge the accuracy gap between a fixed-point model and its stochastic computing implementation, we propose a novel approximate arithmetic-aware training method. We validate the efficacy of our approach by implementing the LeNet-5 convolutional neural network on an FPGA. Our experimental results reveal a negligible accuracy degradation of merely 0.01% compared with the floating-point counterpart, while achieving a substantial 27× speedup and 33× enhancement in energy efficiency compared with other FPGA implementations. Additionally, the proposed method enhances the likelihood of selecting optimal LFSR seeds for stochastic computing systems. Full article

The atmosphere exhibits variability across different time scales. Currently, in the field of atmospheric science, statistical filtering is one of the most widely used methods for extracting signals on certain time scales. However, signal extraction based on traditional statistical filters may be sensitive to missing data points, which are particularly common in meteorological data. To address this issue, this study applies a new type of temporal filters based on a one-dimensional convolution neural network (1D-CNN) and examines its performance on reducing such uncertainties. As an example, we investigate the advantages of a 1D-CNN bandpass filter in extracting quasi-biweekly-to-intraseasonal signals (10–60 days) from temperature data provided by the Hong Kong Observatory. The results show that the 1D-CNN achieves accuracies similar to a 121-point Lanczos filter. In addition, the 1D-CNN filter allows a maximum of 10 missing data points within the 60-point window length, while keeping its accuracy higher than 80% (R² > 0.8). This indicates that the 1D-CNN model works well even when missing data points exist in the time series. This study highlights another potential for applying machine learning algorithms in atmospheric and climate research, which will be useful for future research involving incomplete time series and real-time filtering. Full article

The captured energy of a wind turbine (WT) can be converted into electricity by a generator. Therefore, to improve the efficiency of this system, both the structures of WTs and generators should be considered for control. But the present challenge is WT uncertainty, while the input signals to the generator should be smooth. In this paper, a permanent magnet synchronous generator (PMSG) is considered. The dynamics of the PMSG can be described using two axes, named d-q reference frameworks, with an input in each framework direction. To obtain the maximum power and to overcome the uncertainty by means of a smooth signal, the dynamic sliding mode controller (D-SMC) is implemented. In the D-SMC, an integrator is placed in the control scheme in order to suppress the chattering, because it acts like a low-pass filter. To estimate the state added by the integrator, a new observer-based neural network (ONN) is proposed. The proof of the stability of the D-SMC and ONN is based on Lyapunov theory. To prove the advantages of the D-SMC, a comparison was also carried out by traditional sliding mode control (T-SMC) with a similar ONN. From this comparison, we know that the advantages of the D-SMC are clear in terms of real implementation, concept, and chattering suppression. Full article

Considering the shortcomings of the current monitoring system for tunnel anchor support systems, a tunnel anchor monitoring system based on LSTM-ARIMA prediction is proposed in this paper to prevent the deformation and collapse accidents that may occur in the underground mine tunnels during the backfilling process, which combines the Internet of Things and a neural network deep learning algorithm to achieve the real-time monitoring and prediction of the tunnel anchor pressure. To improve the prediction accuracy, a time series analysis algorithm is used in the prediction model of this system. In particular, an LSTM-ARIMA model is constructed to predict the tunnel anchor pressure by combining the Long Short-Term Memory (LSTM) model and the Autoregressive Integrated Moving Average (ARIMA) model. And a dynamic weighted combination method is designed based on model prediction confidence to acquire the optimal weight coefficients. This combined model enables the monitoring system to predict the anchor pressure more accurately, thereby preventing possible tunnel deformation and collapse accidents in advance. Finally, the overall system is verified using the anchor pressure dataset obtained from the 21,404 section of the Hulusu Coal Mine transportation tunnel in real-world engineering, whose results show that the pressure value predicted using the combined model is basically the same as the actual value on site, and the system has high real-time performance and stability, proving the effectiveness and reliability of the system. Full article

Public authorities and private companies have used video cameras as part of surveillance systems, and one of their objectives is the rapid detection of physically violent actions. This task is usually performed by human visual inspection, which is labor-intensive. For this reason, different deep learning models have been implemented to remove the human eye from this task, yielding positive results. One of the main problems in detecting physical violence in videos is the variety of scenarios that can exist, which leads to different models being trained on datasets, leading them to detect physical violence in only one or a few types of videos. In this work, we present an approach for physical violence detection on images obtained from video based on threshold active learning, that increases the classifier’s robustness in environments where it was not trained. The proposed approach consists of two stages: In the first stage, pre-trained neural network models are trained on initial datasets, and we use a threshold ($\mu$) to identify those images that the classifier considers ambiguous or hard to classify. Then, they are included in the training dataset, and the model is retrained to improve its classification performance. In the second stage, we test the model with video images from other environments, and we again employ ($\mu$) to detect ambiguous images that a human expert analyzes to determine the real class or delete the ambiguity on them. After that, the ambiguous images are added to the original training set and the classifier is retrained; this process is repeated while ambiguous images exist. The model is a hybrid neural network that uses transfer learning and a threshold $\mu$ to detect physical violence on images obtained from video files successfully. In this active learning process, the classifier can detect physical violence in different environments, where the main contribution is the method used to obtain a threshold $\mu$ (which is based on the neural network output) that allows human experts to contribute to the classification process to obtain more robust neural networks and high-quality datasets. The experimental results show the proposed approach’s effectiveness in detecting physical violence, where it is trained using an initial dataset, and new images are added to improve its robustness in diverse environments. Full article

Convolutional neural networks (CNNs) face challenges in capturing long-distance text correlations, and Bidirectional Long Short-Term Memory (BiLSTM) networks exhibit limited feature extraction capabilities for text classification of public service requests. To address the abovementioned problems, this work utilizes an ensemble learning approach to integrate model elements efficiently. This study presents a method for classifying public service request text using a hybrid neural network model called BERT-BiLSTM-CNN. First, BERT (Bidirectional Encoder Representations from Transformers) is used for preprocessing to obtain text vector representations. Then, context and process sequence information are captured through BiLSTM. Next, local features in the text are captured through CNN. Finally, classification results are obtained through Softmax. Through comparative analysis, the method of fusing these three models is superior to other hybrid neural network model architectures in multiple classification tasks. It has a significant effect on public service request text classification. Full article

From the various perspectives of machine learning (ML) and the multiple models used in this discipline, there is an approach aimed at training models for the early detection (ED) of anomalies. The early detection of anomalies is crucial in multiple areas of knowledge since identifying and classifying them allows for early decision making and provides a better response to mitigate the negative effects caused by late detection in any system. This article presents a literature review to examine which machine learning models (MLMs) operate with a focus on ED in a multidisciplinary manner and, specifically, how these models work in the field of fraud detection. A variety of models were found, including Logistic Regression (LR), Support Vector Machines (SVMs), decision trees (DTs), Random Forests (RFs), naive Bayesian classifier (NB), K-Nearest Neighbors (KNNs), artificial neural networks (ANNs), and Extreme Gradient Boosting (XGB), among others. It was identified that MLMs operate as isolated models, categorized in this article as Single Base Models (SBMs) and Stacking Ensemble Models (SEMs). It was identified that MLMs for ED in multiple areas under SBMs’ and SEMs’ implementation achieved accuracies greater than 80% and 90%, respectively. In fraud detection, accuracies greater than 90% were reported by the authors. The article concludes that MLMs for ED in multiple applications, including fraud, offer a viable way to identify and classify anomalies robustly, with a high degree of accuracy and precision. MLMs for ED in fraud are useful as they can quickly process large amounts of data to detect and classify suspicious transactions or activities, helping to prevent financial losses. Full article

Image denoising and super-resolution play vital roles in imaging systems, greatly reducing the preprocessing cost of many AI techniques for object detection, segmentation, and tracking. Various advancements have been accomplished in this field, but progress is still needed. In this paper, we have proposed a novel technique named the Enhanced Learning Enriched Features (ELEF) mechanism using a deep convolutional neural network, which makes significant improvements to existing techniques. ELEF consists of two major processes: (1) Denoising, which removes the noise from images; and (2) Super-resolution, which improves the clarity and details of images. Features are learned through deep CNN and not through traditional algorithms so that we can better refine and enhance images. To effectively capture features, the network architecture adopted Dual Attention Units (DUs), which align with the Multi-Scale Residual Block (MSRB) for robust feature extraction, working sidewise with the feature-matching Selective Kernel Extraction (SKF). In addition, resolution mismatching cases are processed in detail to produce high-quality images. The effectiveness of the ELEF model is highlighted by the performance metrics, achieving a Peak Signal-to-Noise Ratio (PSNR) of 42.99 and a Structural Similarity Index (SSIM) of 0.9889, which indicates the ability to carry out the desired high-quality image restoration and enhancement. Full article

As a vital pigment for photosynthesis in rice, chlorophyll content is closely correlated with growth status and photosynthetic capacity. The estimation of chlorophyll content allows for the monitoring of rice growth and facilitates precise management in the field, such as the application of fertilizers and irrigation. The advancement of hyperspectral remote sensing technology has made it possible to estimate chlorophyll content non-destructively, quickly, and effectively, offering technical support for managing and monitoring rice growth across wide areas. Although hyperspectral data have a fine spectral resolution, they also cause a large amount of information redundancy and noise. This study focuses on the issues of unstable input variables and the estimation model’s poor applicability to various periods when predicting rice chlorophyll content. By introducing the theory of harmonic analysis and the time-frequency conversion method, a deep neural network (DNN) model framework based on wavelet packet transform-first order differential-harmonic analysis (WPT-FD-HA) was proposed, which avoids the uncertainty in the calculation of spectral parameters. The accuracy of estimating rice chlorophyll content based on WPT-FD and WPT-FD-HA variables was compared at seedling, tillering, jointing, heading, grain filling, milk, and complete periods to evaluate the validity and generalizability of the suggested framework. The results demonstrated that all of the WPT-FD-HA models’ single-period validation accuracy had coefficients of determination (R²) values greater than 0.9 and RMSE values less than 1. The multi-period validation model had a root mean square error (RMSE) of 1.664 and an R² of 0.971. Even with independent data splitting validation, the multi-period model accuracy can still achieve R² = 0.95 and RMSE = 1.4. The WPT-FD-HA-based deep learning framework exhibited strong stability. The outcome of this study deserves to be used to monitor rice growth on a broad scale using hyperspectral data. Full article

Photovoltaic (PV) power fluctuates with weather changes, and traditional forecasting methods typically decompose the power itself to study its characteristics, ignoring the impact of multidimensional weather conditions on the power decomposition. Therefore, this paper proposes a short-term PV power generation method based on MVMD (multivariate variational mode decomposition) feature extraction and the Informer model. First, MIC correlation analysis is used to extract weather features most related to PV power. Next, to more comprehensively describe the relationship between PV power and environmental conditions, MVMD is used for time–frequency synchronous analysis of the PV power time series combined with the highest MIC correlation weather data, obtaining frequency-aligned multivariate intrinsic modes. These modes incorporate multidimensional weather factors into the data-decomposition-based forecasting method. Finally, to enhance the model’s learning capability, the Informer neural network model is employed in the prediction phase. Based on the input PV IMF time series and associated weather mode components, the Informer prediction model is constructed for training and forecasting. The predicted results of different PV IMF modes are then superimposed to obtain the total PV power generation. Experiments show that this method improves PV power generation accuracy, with an MAPE value of 4.31%, demonstrating good robustness. In terms of computational efficiency, the Informer model’s ability to handle long sequences with sparse attention mechanisms reduces training and prediction times by approximately 15%, making it faster than conventional deep learning models. Full article

Loop closure detection (LCD), also known as place recognition, is a crucial component of visual simultaneous localization and mapping (vSLAM) systems, aiding in the reduction of cumulative localization errors on a global scale. However, changes in environmental appearance and differing viewpoints pose significant challenges to the accuracy of the LCD algorithm. Addressing this issue, this paper presents a novel end-to-end framework (MetricNet) for LCDs to enhance detection performance in complex scenes with distinct appearance variations. Focusing on deep features with high distinguishability, an attention-based Channel Weighting Module(CWM) is designed to adaptively detect salient regions of interest. In addition, a patch-by-patch Similarity Measurement Module (SMM) is incorporated to steer the network for handling challenging situations that tend to cause perceptual aliasing. Experiments on three typical datasets have demonstrated MetricNet’s appealing detection performance and generalization ability compared to many state-of-the-art learning-based methods, where the mean average precision is increased by up to 11.92%, 18.10%, and 5.33% respectively. Moreover, the detection results on additional open datasets with apparent viewpoint variations and the odometry dataset for localization problems have also revealed the dependability of MetricNet under different adaptation scenarios. Full article

In multivariate and multistep time series prediction research, we often face the problems of insufficient spatial feature extraction and insufficient time-dependent mining of historical series data, which also brings great challenges to multivariate time series analysis and prediction. Inspired by the attention mechanism and residual module, this study proposes a multivariate time series prediction method based on a convolutional-residual gated recurrent hybrid model (CNN-DA-RGRU) with a two-layer attention mechanism to solve the multivariate time series prediction problem in these two stages. Specifically, the convolution module of the proposed model is used to extract the relational features among the sequences, and the two-layer attention mechanism can pay more attention to the relevant variables and give them higher weights to eliminate the irrelevant features, while the residual gated loop module is used to extract the time-varying features of the sequences, in which the residual block is used to achieve the direct connectivity to enhance the expressive power of the model, to solve the gradient explosion and vanishing scenarios, and to facilitate gradient propagation. Experiments were conducted on two public datasets using the proposed model to determine the model hyperparameters, and ablation experiments were conducted to verify the effectiveness of the model; by comparing it with several models, the proposed model was found to achieve good results in multivariate time series-forecasting tasks. Full article

Most traditional rigid grippers can cause damage to the surface of objects in actual production processes and are susceptible to factors such as different shapes, sizes, materials, and positions of the product. This article studies a flexible finger for flexible grippers, more commonly described as PneuNet, designs the structure of the finger, discusses the processing and manufacturing methods of the flexible finger, and prepares a physical model. The influence of structural parameters such as the thickness of the flexible finger and the angle of the air chamber on the bending performance of the finger was analyzed using the Abaqus simulation tool. An RBF-PID control algorithm was used to stabilize the internal air pressure of the flexible fingers. A flexible finger stabilization experimental platform was built to test the ultimate pressure, ultimate bending angle, and end contact force of the fingers, and the simulation results were experimentally verified. The results show that when the thickness of the flexible finger is 2 mm and the air chamber angle is 0 deg, the maximum bending angle of the flexible finger can reach about 136.3°. Under the same air pressure, the bending angle is inversely correlated with the air chamber angle and finger thickness. The experimental error of the bending angle does not exceed 3%, which is consistent with the simulation results as a whole. When the thickness is 2 mm, the maximum end contact force can reach about 1.32 N, and the end contact force decreases with the increase in the air chamber angle. The RBF-PID control algorithm used has improved response speed and a better control effect compared to traditional PID control algorithms. This article provides a clear reference for the application of flexible fingers and flexible grippers, and this research method can be applied to the analysis and design optimization of other soft brakes. Full article

Meat consumption is increasing globally. The safety and quality of meat are considered important issues for human health. During evaluations of meat quality and freshness, microbiological parameters are often analyzed. Counts of indicator cells can provide important references for meat quality. In order to eliminate the error of manual operation and improve detection efficiency, this paper proposed a Convolutional Neural Network (CNN) with a backbone called Detect-Cells-Rapidly-Net (DCRNet), which can identify and count stained cells automatically. The DCRNet replaces the single channel of residual blocks with the aggregated residual blocks to learn more features with fewer parameters. The DCRNet combines the deformable convolution network to fit flexible shapes of stained animal cells. The proposed CNN with DCRNet is self-adaptive to different resolutions of images. The experimental results indicate that the proposed CNN with DCRNet achieves an Average Precision of 81.2% and is better than traditional neural networks for this task. The difference between the results of the proposed method and manual counting is less than 0.5% of the total number of cells. The results indicate that DCRNet is a promising solution for cell detection and can be equipped in future meat quality monitoring systems. Full article

The online detection of fertilizer information is pivotal for precise and intelligent variable-rate fertilizer application. However, traditional methods face challenges such as the complex quantification of multiple components and sensor-induced cross-contamination. This study investigates integrating near-infrared principles with machine learning algorithms to identify fertilizer types and concentrations. We utilized near-infrared transmission spectroscopy and applied Partial Least Squares Discriminant Analysis (PLS-DA), Support Vector Machine (SVM), and Back-Propagation Neural Network (BPNN) algorithms to analyze full spectrum data. The BPNN model, using S-G smoothing, demonstrated a superior classification performance for the nutrient ions of four fertilizer solutions: HPO₄²⁻, NH₄⁺, H₂PO₄⁻ and K⁺. Optimization using the competitive adaptive reweighted sampling (CARS) method yielded BPNN model RMSE values of 0.3201, 0.7160, 0.2036, and 0.0177 for HPO₄²⁻, NH₄⁺, H₂PO₄⁻, and K⁺, respectively. Building on this foundation, we designed a four-channel fertilizer detection device based on the Lambert–Beer law, enabling the real-time detection of fertilizer types and concentrations. The test results confirmed the device’s robust stability, achieving 93% accuracy in identifying fertilizer types and concentrations, with RMSE values ranging from 1.0034 to 2.4947, all within ±8.0% error margin. This study addresses the practical requirements for online fertilizer detection in agricultural engineering, laying the groundwork for efficient water–fertilizer integration technology aligned with sustainable development goals. Full article