Search

This paper proposes an advanced deep learning model that integrates the Diffusion-Transformer structure and parallel attention mechanism for the tasks of growth estimation and disease detection in jujube forests. Existing methods in forestry monitoring often fall short in meeting the practical needs of large-scale and highly complex forest areas due to limitations in data processing capabilities and feature extraction precision. In response to this challenge, this paper designs and conducts a series of benchmark tests and ablation experiments to systematically evaluate and verify the performance of the proposed model across key performance metrics such as precision, recall, accuracy, and F1-score. Experimental results demonstrate that compared to traditional machine learning models like Support Vector Machines and Random Forests, as well as common deep learning models such as AlexNet and ResNet, the model proposed in this paper achieves a precision of 95%, a recall of 92%, an accuracy of 93%, and an F1-score of 94% in the task of disease detection in jujube forests, showing similarly superior performance in growth estimation tasks as well. Furthermore, ablation experiments with different attention mechanisms and loss functions further validate the effectiveness of parallel attention and parallel loss function in enhancing the overall performance of the model. These research findings not only provide a new technical path for forestry disease monitoring and health assessment but also contribute rich theoretical and experimental foundations for related fields. [ABSTRACT FROM AUTHOR]

Herein, we investigate the temperature compensation for a dual-mass MEMS gyroscope. After introducing and simulating the dual-mass MEMS gyroscope's working modes, we propose a hybrid algorithm for temperature compensation relying on improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), sample entropy, time–frequency peak filtering, non-dominated sorting genetic algorithm-II (NSGA II) and extreme learning machine. Firstly, we use ICEEMDAN to decompose the gyroscope's output signal, and then we use sample entropy to classify the decomposed signals. For noise segments and mixed segments with different levels of noise, we use time–frequency peak filtering with different window lengths to achieve a trade-off between noise removal and signal retention. For the feature segment with temperature drift, we build a compensation model using extreme learning machine. To improve the compensation accuracy, NSGA II is used to optimize extreme learning machine, with the prediction error and the 2-norm of the output-layer connection weight as the optimization objectives. Enormous simulation experiments prove the excellent performance of our proposed scheme, which can achieve trade-offs in signal decomposition, classification, denoising and compensation. The improvement in the compensated gyroscope's output signal is analyzed based on Allen variance; its angle random walk is decreased from 0.531076°/h/√Hz to 6.65894 × 10−3°/h/√Hz and its bias stability is decreased from 32.7364°/h to 0.259247°/h. [ABSTRACT FROM AUTHOR]

One of the challenges in material design is to rapidly develop new materials or improve the performance of materials by utilizing the data and knowledge of existing materials. Here, a rapid and effective method of alloy material design via data transfer learning is proposed to efficiently design new alloys using existing data. A new type of aluminum alloy (E2 alloy) with ultra strength and high toughness previously developed by the authors is used as an example. An optimal three-stage solution-aging treatment process (T66R) was efficiently designed transferring 1053 pieces of process-property relationship data of existing AA7xxx commercial aluminum alloys. It realizes the substantial improvement of strength and plasticity of E2 alloy simultaneously, which is of great significance for lightweight of high-end equipment. Meanwhile, the microstructure analysis clarifies the mechanism of alloy performance improvement. This study shows that transferring the existing alloy data is an effective method to design new alloys. [ABSTRACT FROM AUTHOR]

Inkjet printing technology can make tiles with very rich and realistic patterns, so it is widely adopted in the ceramic industry. However, the frequent nozzle blockage and inconsistent inkjet volume by inkjet printing devices, usually leads to defects such as stayguy and color blocks in the tile surface. Especially, the stayguy in complex pattern is difficult to identify by naked eyes due to it is easily covered by complex patterns and becomes invisible, this brings great challenge to tile quality inspection. Nowadays, the machine learning is employed to address the issues. The existing machine learning methods based on hand‐crafted features are capable of stayguy detection of the tiles with a simple pattern, but not applicable for complex patterns due to the interference of pattern in feature extraction. The emerging deep‐learning‐based methods have the potential to be applied for stayguy detection with complex patterns, but cannot achieve real‐time detection due to high complexity. In this paper, a lightweight hand‐crafted feature enhanced convolutional neural network (named HFENet) is proposed for rapid defect detection of tile surface. First, we perform data enhancement on the original image by global histogram equalization and image addition. Second, for the special shape of stayguy which is usually vertical, we embed the extended vertical edge detection operator (Prewitt) as convolution kernel into HFENet to extract the hand‐crafted vertical edge features of the test image and eliminate the interference of complex pattern in the feature extraction. Third, the 5 × 1 asymmetric convolution kernel with a dilation rate of 2 is used to improve the utilization of convolution kernel and reduce the complexity of the model. Fourth, to reach the real‐time requirements, a memory access cost‐aware design is proposed, which can orchestrate the number of shallow convolution layers and deep convolution layers in feature extraction. The experiments were performed on the ceramic tile image data set captured by high‐resolution industrial cameras in ceramic tile production line. Experimental results show that the HFENet outperforms the state‐of‐the‐art semantic segmentation networks (i.e., UNet, FCN‐8s, SegNet, DeepLabV3+, etc.) and lightweight networks (i.e., ShuffleNet, MobileNet, and SqueezeNet). All the code and data are available at a GitHub repository (https://github.com/RobotvisionLab/HFENet). [ABSTRACT FROM AUTHOR]

• A novel MPC approach was proposed to optimize the operation of a district heating substation. • The proposed method combines a reduced-order physical model and a data-driven model. • A data-driven surrogate model is developed for the performance test of the proposed approach. • The MPC controller is effective in reducing the energy consumption of district heating substations. District heating (DH) substations play a crucial role in ensuring the efficient and effective distribution of thermal energy necessary to provide space heating for buildings. However, optimizing their operation for energy savings while still ensuring indoor comfort poses significant challenges due to the complex dynamics of building demand and the inertia of building envelopes. To address these challenges, this study introduces a novel model predictive control (MPC) approach that combines a reduced-order physical model with a machine learning-based data-driven model to jointly optimize the operation parameters of a DH substation. In this approach, a reduced-order physical model is first used to capture essential operational principles and energy behaviors of the DH substations and generate candidate solutions for the control of the DH substations. Then, a data-driven model is constructed by integrating a Long Short-Term Memory model and a Back-propagation Neural Network, leveraging historical operational data of the DH substation concerned. The data-driven model is further formulated into a data-driven MPC framework to identify optimal control solutions from all candidates provided by the physical model. To evaluate the proposed approach, a data-driven surrogate model is developed using real operational data. Comparative analysis against the original fuzzy rule-based control strategy and a pure data-driven strategy demonstrates a substantial reduction in heat consumption of 4.77% and 19.47%, respectively. Moreover, compared with using a reduced-order physical model alone, this approach achieves additional benefits in reducing the energy consumption of the DH substation and minimizing indoor temperature fluctuations within the end-users. [ABSTRACT FROM AUTHOR]

Abstract: Clustering is a fundamental topic in pattern recognition and machine learning research. Traditional clustering methods deal with a single clustering task on a single data set. However, in many real applications, multiple similar clustering tasks are involved simultaneously, e.g., clustering clients of different shopping websites, in which data of different subjects are collected for each task. These tasks are cross-domains but closely related. It is proved that we can improve the individual performance of each clustering task by appropriately utilizing the underling relation. In this paper, we will propose a new approach, which performs multiple related clustering tasks simultaneously through domain adaptation. A shared subspace will be learned through domain adaptation, where the gap of distributions among tasks is reduced, and the shared knowledge will be transferred through all tasks by exploiting the strengthened relation in the learned subspace. Then the object is set as the best clustering in both the original and learned spaces. An alternating optimization method is introduced and its convergence is theoretically guaranteed. Experiments on both synthetic and real data sets demonstrate the effectiveness of the proposed approach. [Copyright &y& Elsevier]

Correct sign language recognition helps deaf people communicate normally with hearing people. However, existing sign language recognition techniques have low recognition accuracy due to insufficient feature extraction. In this paper, we propose an approach to improve the accuracy of sign language recognition. Firstly, we design a one-dimensional convolutional neural network (CNN) using the skip connection approach Secondly, we propose an improved multi-head attention mechanism that incorporates bi-directional long short-term memory (BiLSTM) networks within a multi-head attention mechanism. Thirdly, we position this improved multi-head attention mechanism behind the final convolutional layer of the proposed CNN architecture and denote the resultant architecture as BMCNN. Finally, we verify the performance of our approach with BMCNN architecture through ten-fold cross-validation on the sensing dataset. Our approach achieves 99.13% accuracy in sign language recognition. The results show that our proposed approach outperforms the traditional machine learning approaches and other state-of-the-art techniques in this field. [Display omitted] • Proposing an approach based on sensing data and deep learning. • The convolutional neural network using skip connection enhances feature extraction. • The improved multi-head attention extracts temporal properties and key features. • Realizing real-time accurate sign language recognition. [ABSTRACT FROM AUTHOR]

A micro-expression is defined as an uncontrollable muscular movement shown on the face of humans when one is trying to conceal or repress his true emotions. Many researchers have applied the deep learning framework to micro-expression recognition in recent years. However, few have introduced the human visual attention mechanism to micro-expression recognition. In this study, we propose a three-dimensional (3D) spatiotemporal convolutional neural network with the convolutional block attention module (CBAM) for micro-expression recognition. First image sequences were input to a medium-sized convolutional neural network (CNN) to extract visual features. Afterwards, it learned to allocate the feature weights in an adaptive manner with the help of a convolutional block attention module. The method was testified in spontaneous micro-expression databases (Chinese Academy of Sciences Micro-expression II (CASME II), Spontaneous Micro-expression Database (SMIC)). The experimental results show that the 3D CNN with convolutional block attention module outperformed other algorithms in micro-expression recognition. [ABSTRACT FROM AUTHOR]

Strength, toughness, and stress corrosion resistance are critical properties of aluminum alloys for high-end equipment manufacturing. Unfortunately, the situation of complex alloy composition, diverse aging systems, and conflicting property relationships hinder the synchronous enhancement of three properties. Here, we proposed an interpretable machine learning design strategy for high-end aluminum alloy. The critical intrinsic factors and explicit laws of elements affecting the ultimate tensile strength (UTS), fracture toughness (K IC), and stress corrosion sensitivity factor (ISSRT) of alloys were excavated: The elements with large number of electrons in d-valence electron orbitals, high boiling point, and low nuclear electron distance help enhance the UTS; The elements with low density and minimized difference in first ionization energy with aluminum help improve the K IC ; The elements with high diffusion activation energy in aluminum and high corrosion potential in seawater help reduce the ISSRT. Based on the above findings, three microalloying elements of Ti, Cr, and Zr, which have the remarkable combined effect of enhancing synchronously the three properties, were selected, and a new advanced aluminum alloy Al-10.50Zn-2.31Mg-1.56Cu-0.09Ti-0.15Cr-0.10Zr was designed. The UTS, K IC, and ISSRT were 760 ± 4 MPa, 34.9 ± 0.3 MPa·m1/2, and 13.3 % ± 1.7 %, respectively, after RRA treatment. Microstructure analysis revealed that the new alloy had almost no micron secondary phase after RRA treatment, reducing the sites for pitting and cavity formation. The addition of Ti, Cr, and Zr formed dispersoids Al 18 (Cr,Ti) 2 Mg 3 and Al 3 Zr, which contributed to the synchronous improvement of strength, toughness, and stress corrosion resistance. The high-volume fraction of precipitates significantly enhanced the strength of the alloy. [Display omitted] [ABSTRACT FROM AUTHOR]

Machine learning has attracted extensive interest in the process engineering field, due to the capability of modeling complex nonlinear process behavior. This work presents a method for combining neural network models with first-principles models in real-time optimization (RTO) and model predictive control (MPC) and demonstrates the application to two chemical process examples. First, the proposed methodology that integrates a neural network model and a first-principles model in the optimization problems of RTO and MPC is discussed. Then, two chemical process examples are presented. In the first example, a continuous stirred tank reactor (CSTR) with a reversible exothermic reaction is studied. A feed-forward neural network model is used to approximate the nonlinear reaction rate and is combined with a first-principles model in RTO and MPC. An RTO is designed to find the optimal reactor operating condition balancing energy cost and reactant conversion, and an MPC is designed to drive the process to the optimal operating condition. A variation in energy price is introduced to demonstrate that the developed RTO scheme is able to minimize operation cost and yields a closed-loop performance that is very close to the one attained by RTO/MPC using the first-principles model. In the second example, a distillation column is used to demonstrate an industrial application of the use of machine learning to model nonlinearities in RTO. A feed-forward neural network is first built to obtain the phase equilibrium properties and then combined with a first-principles model in RTO, which is designed to maximize the operation profit and calculate optimal set-points for the controllers. A variation in feed concentration is introduced to demonstrate that the developed RTO scheme can increase operation profit for all considered conditions. [ABSTRACT FROM AUTHOR]

[Display omitted] • Strength and magnetic properties of silicon steel were optimized simultaneously. • Adversarial model was applied in silicon steel composition design. • The effect of alloying elements on properties was analyzed based on Association rule mining. Limitations on data volume and quality are key bottlenecks in using machine learning for property prediction and property-oriented composition design of non-oriented silicon steel. In response, this study employed a Gaussian Mixture Model (GMM) to generate virtual samples for enhancing the in-house experimental data, by which the generated virtual data well captured the distribution of the raw experimental data. As a result, compared with the model without data augmentation, the enhanced prediction model (composition → property) fitted by virtual samples improved its R 2 value from 0.52 to 0.86. Based on this model, a multi-properties-oriented (yield strength σ s , magnetic induction B 50 , and iron loss P 10/400) composition prediction model (property → composition) was established to rapidly discover high-performance non-oriented silicon steel. Experimental characterization and theory calculation of the explored candidate alloys exhibited that their yield strength was above 750 MPa, which primarily results from solid solution strengthening (50 %) and precipitation strengthening (36 %). Moreover, the alloys possessed magnetic induction B 50 of over 1.72 T and low core losses with P 10/400 of 12.1 W/kg (0.2 mm) and 17.0 W/kg (0.35 mm). Further microstructural characterization exhibited that such satisfactory performance is associated with copper (Cu)-rich nanoprecipitates that dramatically improved the yield strength without deteriorating the magnetic properties. [ABSTRACT FROM AUTHOR]