Attentional Modulation Research Articles

RS-MOCO: A deep learning-based topology-preserving image registration method for cardiac T1 mapping

Cardiac T1 mapping can evaluate various clinical symptoms of myocardial tissue. However, there is currently a lack of effective, robust, and efficient methods for motion correction in cardiac T1 mapping. In this paper, we propose a deep learning-based and topology-preserving image registration framework for motion correction in cardiac T1 mapping. Notably, our proposed implicit consistency constraint dubbed BLOC, to some extent preserves the image topology in registration by bidirectional consistency constraint and local anti-folding constraint. To address the contrast variation issue, we introduce a weighted image similarity metric for multimodal registration of cardiac T1-weighted images. Besides, a semi-supervised myocardium segmentation network and a dual-domain attention module are integrated into the framework to further improve the performance of the registration. Numerous comparative experiments, as well as ablation studies, demonstrated the effectiveness and high robustness of our method. The results also indicate that the proposed weighted image similarity metric, specifically crafted for our network, contributes a lot to the enhancement of the motion correction efficacy, while the bidirectional consistency constraint combined with the local anti-folding constraint ensures a more desirable topology-preserving registration mapping.

Attentional modulation of secondary somatosensory and visual thalamus of mice.

Each sensory modality has its own primary and secondary thalamic nuclei. While the primary thalamic nuclei are well understood to relay sensory information from the periphery to the cortex, the role of secondary sensory nuclei is elusive. We trained head-fixed mice to attend to one sensory modality while ignoring a second modality, namely to attend to touch and ignore vision, or vice versa. Arrays were used to record simultaneously from the secondary somatosensory thalamus (POm) and secondary visual thalamus (LP). In mice trained to respond to tactile stimuli and ignore visual stimuli, POm was robustly activated by touch and largely unresponsive to visual stimuli. A different pattern was observed when mice were trained to respond to visual stimuli and ignore touch, with POm now more robustly activated during visual trials. This POm activity was not explained by differences in movements (i.e. whisking, licking, pupil dilation) resulting from the two tasks. Post hoc histological reconstruction of array tracks through POm revealed that subregions varied in their degree of plasticity. LP exhibited similar phenomena. We conclude that behavioral training reshapes activity in secondary thalamic nuclei. Secondary nuclei respond to the same behaviorally relevant, reward-predicting stimuli regardless of stimulus modality.

Attentional modulation of secondary somatosensory and visual thalamus of mice

Each sensory modality has its own primary and secondary thalamic nuclei. While the primary thalamic nuclei are well understood to relay sensory information from the periphery to the cortex, the role of secondary sensory nuclei is elusive. We trained head-fixed mice to attend to one sensory modality while ignoring a second modality, namely to attend to touch and ignore vision, or vice versa. Arrays were used to record simultaneously from the secondary somatosensory thalamus (POm) and secondary visual thalamus (LP). In mice trained to respond to tactile stimuli and ignore visual stimuli, POm was robustly activated by touch and largely unresponsive to visual stimuli. A different pattern was observed when mice were trained to respond to visual stimuli and ignore touch, with POm now more robustly activated during visual trials. This POm activity was not explained by differences in movements (i.e. whisking, licking, pupil dilation) resulting from the two tasks. Post hoc histological reconstruction of array tracks through POm revealed that subregions varied in their degree of plasticity. LP exhibited similar phenomena. We conclude that behavioral training reshapes activity in secondary thalamic nuclei. Secondary nuclei respond to the same behaviorally relevant, reward-predicting stimuli regardless of stimulus modality.

CNN-CBAM-LSTM: Enhancing Stock Return Prediction Through Long and Short Information Mining in Stock Prediction

Deep learning, a foundational technology in artificial intelligence, facilitates the identification of complex associations between stock prices and various influential factors through comprehensive data analysis. Stock price data exhibits unique time-series characteristics; models emphasizing long-term data may miss short-term fluctuations, while those focusing solely on short-term data may not capture cyclical trends. Existing models that integrate long short-term memory (LSTM) and convolutional neural networks (CNNs) face limitations in capturing both long- and short-term dependencies due to LSTM’s gated transmission mechanism and CNNs’ limited receptive field. This study introduces an innovative deep learning model, CNN-CBAM-LSTM, which integrates the convolutional block attention module (CBAM) to enhance the extraction of both long- and short-term features. The model’s performance is assessed using the Australian Standard & Poor’s 200 Index (AS51), showing improvement over traditional models across metrics such as RMSE, MAE, R2, and RETURN. To further confirm its robustness and generalizability, Diebold–Mariano (DM) tests and model confidence set experiments are conducted, with results indicating the consistently high performance of the CNN-CBAM-LSTM model. Additional tests on six globally recognized stock indices reinforce the model’s predictive strength and adaptability, establishing it as a reliable tool for forecasting in the stock market.

Strip and boundary detection multi-task learning network for segmentation of meibomian glands.

Automatic segmentation of meibomian glands in near-infrared meibography images is basis of morphological parameter analysis, which plays a crucial role in facilitating the diagnosis of meibomian gland dysfunction (MGD). The special strip shape and the adhesion between glands make the automatic segmentation of meibomian glands very challenging. A strip and boundary detection multi-task learning network (SBD-MTLNet) based on encoder-decoder structure is proposed to realize the automatic segmentation of meibomian glands. A strip mixed attention module (SMAM) is proposed to enhance the network's ability to recognize the strip shape of glands. To alleviate the problem of adhesion between glands, a boundary detection auxiliary network (BDA-Net) is proposed, which introduces boundary features to assist gland segmentation. A self-adaptive interactive information fusion module (SIIFM) based on reverse attention mechanism is proposed to realize information complementation between meibomian gland segmentation and boundary detection tasks. The proposed SBD-MTLNet has been evaluated on an in-house dataset (453 images) and a public dataset MGD-1K (1000 images). Due to the limited number of images, a five-fold cross validation strategy is adopted. Average dice coefficient of the proposed SBD-MTLNet reaches 81.08% and 84.32% on the in-house dataset and the public one, respectively. Comprehensive experimental results demonstrate the effectiveness the proposed SBD-MTLNet, outperforming other state-of-the-art methods. The proposed SBD-MTLNet can focus more on the shape characteristics of the meibomian glands and the boundary contour information between the adjacent glands via multi-task learning strategy. The segmentation results of the proposed method can be used for the quantitative morphological characteristics analysis of meibomian glands, which has potential for the auxiliary diagnosis of MGD in clinic.

AKS-YOLOv8: a lightweight and accurate network model for infrared ship detection

Abstract Combining infrared technology and artificial intelligence technology to achieve efficient ship detection is practically important for monitoring and scheduling of ships in harbours. Due to the infrared image texture features are fewer and the background is complex, these lead to the algorithm model having the problems of complex structure, high amount of memory consumption and not being easy to deploy. To address these problems, we propose a lightweight and accurate infrared ship detection model named AKS-YOLOv8, which is based on YOLOv8 design. Firstly, the Split Asymptotic Feature Pyramid Network (SA-FPN) is used in the neck network. It improves model to detect multi-scale infrared ship in large scenes while keeping the model lightweight. Secondly, we introduced the dynamic convolution module KernelWarehouse Dynamic Convolution (KConv) in the backbone network for enhancing the recognition of small targets in infrared images. And we also used FasterBlock instead of Bottleneck to reduce feature map redundancy and improve inference speed. Finally, we adopt Simple, Parameter-free Attention Module (SimAM) to improve the feature extraction ability for key information of infrared ships. To verify the effectiveness of AKS-YOLOv8, a comprehensive test is performed on the infrared ship target recognition database issued by Yantai iraytek Co., Ltd in comparison with some widely-used detection algorithms. The model precision of AKS-YOLOv8 is improved by 1%, and the parameters and model size are reduced by 37.2% and 32.1%, respectively. The results of the experiment indicate that AKS-YOLOv8 has successfully struck a favorable equilibrium between being lightweight and maintaining accuracy. It can provide a valuable technical reference for infrared ship detection technology.

UNet-Att: a self-supervised denoising and recovery model for two-photon microscopic image

Two-photon microscopy is indispensable in cell and molecular biology for its high-resolution visualization of cellular and molecular dynamics. However, the inevitable low signal-to-noise conditions significantly degrade image quality, obscuring essential details and complicating morphological analysis. While existing denoising methods such as CNNs, Noise2Noise, and DeepCAD serve broad applications in imaging, they still have limitations in preserving texture structures and fine details in two-photon microscopic images affected by complex noise, particularly in sophisticated structures like neuronal synapses. To improve two-photon microscopy image denoising effectiveness, by experimenting on real two-photon microscopy images, we propose a novel deep learning framework, the UNet-Att model, which integrates a specifically tailored UNet++ architecture with attention mechanisms. Specifically, this approach consists of a sophisticated downsampling module for extracting hierarchical features at varied scales, and an innovative attention module that strategically emphasizes salient features during the integration process. The architecture is completed by an ingenious upsampling pathway that reconstructs the image with high fidelity, ensuring the retention of textural integrity. Additionally, the model supports end-to-end training, optimizing its denoising efficacy. The UNet-Att model proves to surpass mainstream algorithms in the dual objectives of denoising and preserving the textural intricacies of original images, which is evidenced by an increase of 9.42 dB in the high Peak Signal-to-Noise Ratio (PSNR) coupled with an improvement of 0.1131 in the Structural Similarity Index Measurement (SSIM). The ablation experiments reveal the effectiveness of each module. The associated Python packages and datasets of UNet-Att are freely available at https://github.com/ZjjDh/UNet-Att.

Edge-guided and hierarchical aggregation network for robust medical image segmentation

Medical image segmentation with the convolutional neural networks (CNNs), significantly enhances clinical analysis and disease diagnosis. However, medical images inherently exhibit large intra-class variability, minimal inter-class differences, and substantial noise levels. Extracting robust contextual information and aggregating discriminative features for fine-grained segmentation remains a formidable task. Additionally, existing methods often struggle with producing accurate mask edges, leading to blurred boundaries and reduced segmentation precision. This paper introduces a novel Edge-guided and Hierarchical Aggregation Network (EHANet) which excels at capturing rich contextual information and preserving fine spatial details, addressing the critical issues of inaccurate mask edges and detail loss prevalent in current segmentation models. The Inter-layer Edge-aware Module (IEM) enhances edge prediction accuracy by fusing early encoder layers, ensuring precise edge delineation. The Efficient Fusion Attention Module (EFA) adaptively emphasizes critical spatial and channel features while filtering out redundancies, enhancing the model’s perception and representation capabilities. The Adaptive Hierarchical Feature Aggregation Module (AHFA) module optimizes feature fusion within the decoder, maintaining essential information and improving reconstruction fidelity through hierarchical processing. Quantitative and qualitative experiments on four public datasets demonstrate the effectiveness of EHANet in achieving superior mIoU, mDice, and edge accuracy against eight other state-of-the-art segmentation methods, highlighting its robustness and precision in diverse clinical scenarios.

Feature pyramid attention network for audio‐visual scene classification

AbstractAudio‐visual scene classification (AVSC) poses a formidable challenge owing to the intricate spatial‐temporal relationships exhibited by audio‐visual signals, coupled with the complex spatial patterns of objects and textures found in visual images. The focus of recent studies has predominantly revolved around extracting features from diverse neural network structures, inadvertently neglecting the acquisition of semantically meaningful regions and crucial components within audio‐visual data. The authors present a feature pyramid attention network (FPANet) for audio‐visual scene understanding, which extracts semantically significant characteristics from audio‐visual data. The authors’ approach builds multi‐scale hierarchical features of sound spectrograms and visual images using a feature pyramid representation and localises the semantically relevant regions with a feature pyramid attention module (FPAM). A dimension alignment (DA) strategy is employed to align feature maps from multiple layers, a pyramid spatial attention (PSA) to spatially locate essential regions, and a pyramid channel attention (PCA) to pinpoint significant temporal frames. Experiments on visual scene classification (VSC), audio scene classification (ASC), and AVSC tasks demonstrate that FPANet achieves performance on par with state‐of‐the‐art (SOTA) approaches, with a 95.9 F1‐score on the ADVANCE dataset and a relative improvement of 28.8%. Visualisation results show that FPANet can prioritise semantically meaningful areas in audio‐visual signals.

Channel and Spatial Attention in Chest X-Ray Radiographs: Advancing Person Identification and Verification with Self-Residual Attention Network

Background/Objectives: In contrast to traditional biometric modalities, such as facial recognition, fingerprints, and iris scans or even DNA, the research orientation towards chest X-ray recognition has been spurred by its remarkable recognition rates. Capturing the intricate anatomical nuances of an individual’s skeletal structure, the ribcage of the chest, lungs, and heart, chest X-rays have emerged as a focal point for identification and verification, especially in the forensic field, even in scenarios where the human body damaged or disfigured. Discriminative feature embedding is essential for large-scale image verification, especially in applying chest X-ray radiographs for identity identification and verification. This study introduced a self-residual attention-based convolutional neural network (SRAN) aimed at effective feature embedding, capturing long-range dependencies and emphasizing critical spatial features in chest X-rays. This method offers a novel approach to person identification and verification through chest X-ray categorization, relevant for biometric applications and patient care, particularly when traditional biometric modalities are ineffective. Method: The SRAN architecture integrated a self-channel and self-spatial attention module to minimize channel redundancy and enhance significant spatial elements. The attention modules worked by dynamically aggregating feature maps across channel and spatial dimensions to enhance feature differentiation. For the network backbone, a self-residual attention block (SRAB) was implemented within a ResNet50 framework, forming a Siamese network trained with triplet loss to improve feature embedding for identity identification and verification. Results: By leveraging the NIH ChestX-ray14 and CheXpert datasets, our method demonstrated notable improvements in accuracy for identity verification and identification based on chest X-ray images. This approach effectively captured the detailed anatomical characteristics of individuals, including skeletal structure, ribcage, lungs, and heart, highlighting chest X-rays as a viable biometric tool even in cases of body damage or disfigurement. Conclusions: The proposed SRAN with self-residual attention provided a promising solution for biometric identification through chest X-ray imaging, showcasing its potential for accurate and reliable identity verification where traditional biometric approaches may fall short, especially in postmortem cases or forensic investigations. This methodology could play a transformative role in both biometric security and healthcare applications, offering a robust alternative modality for identity verification.

Sa-SNN: spiking attention neural network for image classification

Spiking neural networks (SNNs) are known as third generation neural networks due to their energy efficient and low power consumption. SNNs have received a lot of attention due to their biological plausibility. SNNs are closer to the way biological neural systems work by simulating the transmission of information through discrete spiking signals between neurons. Influenced by the great potential shown by the attention mechanism in convolutional neural networks, Therefore, we propose a Spiking Attention Neural Network (Sa-SNN). The network includes a novel Spiking-Efficient Channel Attention (SECA) module that adopts a local cross-channel interaction strategy without dimensionality reduction, which can be achieved by one-dimensional convolution. It is implemented by convolution, which involves a small number of model parameters but provides a significant performance improvement for the network. The design of local inter-channel interactions through adaptive convolutional kernel sizes, rather than global dependencies, allows the network to focus more on the selection of important features, reduces the impact of redundant features, and improves the network’s recognition and generalisation capabilities. To investigate the effect of this structure on the network, we conducted a series of experiments. Experimental results show that Sa-SNN can perform image classification tasks more accurately. Our network achieved 99.61%, 99.61%, 94.13%, and 99.63% on the MNIST, Fashion-MNIST, N-MNIST datasets, respectively, and Sa-SNN performed well in terms of accuracy compared with mainstream SNNs.

ML-driven semi-auto multi-scenario horizon interpretation with uncertainty quantification

ML-driven semi-auto multi-scenario horizon interpretation with uncertainty quantification

Enhanced multi-scale trademark element detection using the improved DETR

The exponential growth in the number of registered trademarks, coupled with the escalating incidents of trademark infringement, has made the automatic detection of such infractions a crucial area of study in the domain of market regulation. In light of the diverse range of elements and the pervasive presence of small targets in trademark images, we present an enhanced version of the DETR-based Multi-Scale Trademark Element Detection Network (MSTED-Net). Our primary innovation lies in incorporating a dual fusion mechanism that integrates the Spatial Attention Module (SAM) and Global Context Network (GCNet) within the backbone network, thereby providing a more robust approach to capture the essential characteristics of the trademark images under investigation. Subsequently, we develop a Multi-scale Feature Augmentation Pyramid (MFA-FPN), which aims to further fortify the model’s ability to extract features and boost the detection efficiency for small targets. The efficacy of our proposed detection network is demonstrated through experimental results, showcasing an outstanding detection accuracy of 91.12% in comparison to other state-of-the-art detection algorithms.

A Study of Potential Applications of Student Emotion Recognition in Primary and Secondary Classrooms

Emotion recognition is critical to understanding students’ emotional states. However, problems such as crowded classroom environments, changing light, and occlusion often affect the accuracy of recognition. This study proposes an emotion recognition algorithm specifically for classroom environments. Firstly, the study adds the self-made MCC module and the Wise-IoU loss function to make object detection in the YOLOv8 model more accurate and efficient. Compared with the native YOL0v8x, it reduces the parameters by 16% and accelerates the inference speed by 20%. Secondly, in order to address the intricacies of the classroom setting and the specific requirements of the emotion recognition task, a multi-channel emotion recognition network (MultiEmoNet) has been developed. This network fuses skeletal, environmental, and facial information, and introduces a central loss function and an attention module AAM to enhance the feature extraction capability. The experimental results show that MultiEmoNet achieves a classification accuracy of 91.4% on a homemade classroom student emotion dataset, which is a 10% improvement over the single-channel classification algorithm. In addition, this study also demonstrates the dynamic changes in students’ emotions in the classroom through visual analysis, which helps teachers grasp students’ emotional states in real time. This paper validates the potential of multi-channel information-fusion deep learning techniques for classroom teaching analysis and provides new ideas and tools for future improvements to emotion recognition techniques.

Underwater acoustic multi-target recognition based on channel attention mechanism

The recognition of ship-radiated noise presents a critical challenge in underwater acoustic target recognition, primarily due to interference from multiple vessels in real-world settings. This interference produces complex signals that contain multiple and overlapping targets. Despite the practical relevance of multi-target recognition, most existing research in underwater acoustic target recognition has concentrated on single-target recognition. To address the multi-target recognition problem, this paper proposes a deep learning (DL) framework that combines residual neural network (ResNet) architecture with channel attention modules. These modules excel in selectively filtering channel information within feature maps, amplifying informative features while mitigating noise interference. Thus, it enables accurate recognize both the quantity and types of targets within single-channel acoustic signals. In addition, this paper explores the effects of signal duration and energy ratios between different targets in mixed signals on recognition performance. Experimental results reveal that shorter signal durations hinder the extraction of discriminative features, reducing recognition accuracy. Notably, for signal durations of 5 s, the proposed model achieves a recognition accuracy of 97.16% on the dataset with energy ratios between −5 dB and 5 dB, closely approaching the accuracy of 98.09% obtained at a dataset with energy ratio of 0 dB.

An Improved 2D Pose Estimation Algorithm for Extracting Phenotypic Parameters of Tomato Plants in Complex Backgrounds

Phenotypic traits, such as plant height, internode length, and node count, are essential indicators of the growth status of tomato plants, carrying significant implications for research on genetic breeding and cultivation management. Deep learning algorithms such as object detection and segmentation have been widely utilized to extract plant phenotypic parameters. However, segmentation-based methods are labor-intensive due to their requirement for extensive annotation during training, while object detection approaches exhibit limitations in capturing intricate structural features. To achieve real-time, efficient, and precise extraction of phenotypic traits of seedling tomatoes, a novel plant phenotyping approach based on 2D pose estimation was proposed. We enhanced a novel heatmap-free method, YOLOv8s-pose, by integrating the Convolutional Block Attention Module (CBAM) and Content-Aware ReAssembly of FEatures (CARAFE), to develop an improved YOLOv8s-pose (IYOLOv8s-pose) model, which efficiently focuses on salient image features with minimal parameter overhead while achieving a superior recognition performance in complex backgrounds. IYOLOv8s-pose manifested a considerable enhancement in detecting bending points and stem nodes. Particularly for internode detection, IYOLOv8s-pose attained a Precision of 99.8%, exhibiting a significant improvement over RTMPose-s, YOLOv5s6-pose, YOLOv7s-pose, and YOLOv8s-pose by 2.9%, 5.4%, 3.5%, and 5.4%, respectively. Regarding plant height estimation, IYOLOv8s-pose achieved an RMSE of 0.48 cm and an rRMSE of 2%, and manifested a 65.1%, 68.1%, 65.6%, and 51.1% reduction in the rRMSE compared to RTMPose-s, YOLOv5s6-pose, YOLOv7s-pose, and YOLOv8s-pose, respectively. When confronted with the more intricate extraction of internode length, IYOLOv8s-pose also exhibited a 15.5%, 23.9%, 27.2%, and 12.5% reduction in the rRMSE compared to RTMPose-s, YOLOv5s6-pose, YOLOv7s-pose, and YOLOv8s-pose. IYOLOv8s-pose achieves high precision while simultaneously enhancing efficiency and convenience, rendering it particularly well suited for extracting phenotypic parameters of tomato plants grown naturally within greenhouse environments. This innovative approach provides a new means for the rapid, intelligent, and real-time acquisition of plant phenotypic parameters in complex backgrounds.

HyperGAN: A Hyperspectral Image Fusion Approach Based on Generative Adversarial Networks

The objective of hyperspectral pansharpening is to fuse low-resolution hyperspectral images (LR-HSI) with corresponding panchromatic (PAN) images to generate high-resolution hyperspectral images (HR-HSI). Despite advancements in hyperspectral (HS) pansharpening using deep learning, the rich spectral details and large data volume of HS images place higher demands on models for effective spectral extraction and processing. In this paper, we present HyperGAN, a hyperspectral image fusion approach based on Generative Adversarial Networks. Unlike previous methods that deepen the network to capture spectral information, HyperGAN widens the structure with a Wide Block for multi-scale learning, effectively capturing global and local details from upsampled HSI and PAN images. While LR-HSI provides rich spectral data, PAN images offer spatial information. We introduce the Efficient Spatial and Channel Attention Module (ESCA) to integrate these features and add an energy-based discriminator to enhance model performance by learning directly from the Ground Truth (GT), improving fused image quality. We validated our method on various scenes, including the Pavia Center, Eastern Tianshan, and Chikusei. Results show that HyperGAN outperforms state-of-the-art methods in visual and quantitative evaluations.

DCUFormer: Enhancing pavement crack segmentation in complex environments with dual-cross/upsampling attention

Efficient road inspection and maintenance are essential to extend pavement lifespan and enhance safety. However, automated crack detection remains challenging due to varied environmental conditions and differences in image collection equipment, making robust algorithm development a critical need. Vision Transformers, with their capacity to capture long-range dependencies, offer significant advantages for crack detection in complex scenarios by effectively extracting global features. Nevertheless, existing Transformer-based methods encounter difficulties in boundary delineation due to decoder design limitations, which lead to suboptimal fusion of low-level and high-level features. To address this issue, we propose a comprehensive approach that integrates semantic preservation, detail refinement, and detail delineation. These concepts are realized through our novel Dual-Cross Attention Module (DCA) and Upsampling Attention Module (UA). The DCA module progressively filters redundant details from low-level feature layers using high-level semantic information, while preserving boundary details to refine high-level feature boundaries. In addition, the UA module employs progressive local cross-attention in upsampling, facilitating more precise boundary definitions and surpassing conventional dynamic upsampling methods. Our approach, utilizing both lightweight (MiT-B0, LVT) and middle-weight (Swin-T) backbones, demonstrates state-of-the-art performance on three diverse datasets—Crack500, CrackSC, and UAV-Crack500—highlighting its robustness across varied conditions. This work contributes to advancing Transformer-based architectures for defect segmentation in complex engineering contexts, underscoring the critical role of improved feature fusion in crack detection. The code is available at: https://github.com/SHAN-JH/DCUFormer.

Research on Irrigation Grade Discrimination Method Based on Semantic Segmentation

As one of China’s major grain crops, wheat has a high demand for water resources, making it susceptible to drought stress. Traditional irrigation evaluation methods are often based on experience and rule-based calculations, which struggle to cope with complex environmental factors and dynamic changes in crop needs. With technological advancements, deep learning-based research methods, characterized by their strong data-driven analytical capabilities, are expected to improve the accuracy of evaluation results. This paper focuses on the irrigation demand assessment of winter wheat farmland, aiming to explore a new regional-scale irrigation demand assessment method based on deep learning. By establishing samples of different irrigation evaluation levels, this study seeks to better meet the requirements of irrigation demand assessment. For the problem of regional-scale irrigation-level discrimination, the Convolutional Network Attention(CONAT) module was proposed to optimize the backbone network structure of the Mask2Former model. To tackle issues related to data imbalance and underfitting across certain categories, a loss function tailored for imbalanced sample distributions was implemented, accompanied by enhancements to the training scheme. By contrasting this refined model with alternative methods for discriminating irrigation levels, the viability of this approach was showcased.

Deep learning based heat transfer simulation of the casting process

To avoid the necessity of constitutional models, computational intensity, and the time-consuming nature inherent in numerical simulations, a pioneering approach utilizing deep learning techniques has been adopted to swiftly predict temperature fields during the solidification phase of casting processes. This methodology involves the development of rapid prediction models based on modified U-net network architectures, augmented by the integration of Inception and CBAM (Convolutional Block Attention Module) modules. The construction of the training set involved utilizing 200 diverse geometric models with each containing three kinds of components (casting, mold, and chill), where the temperature fields at a specific time, ti, were input data, while that of the subsequent time point, ti+1, served as the corresponding labels. The geometric models were generated by the erosion of 2D arbitrary shapes through an erosion method, and then their associated temperature fields were obtained via FDM-based numerical simulation. The trained deep learning models exhibit proficiency in promptly forecasting temperature fields during the solidification process for arbitrarily shaped castings at different times. The average accuracy of the predicted outcomes reaches 94.5% as the absolute temperature error set as 7 ℃ and the prediction just takes one second for a time step. Notably, these models are adept at handling multi-component with multi-materials within a geometry model, such as casting, chill, and mold corresponding to the intricate casting process.