Attention Convolutional Neural Network Research Articles

Multi-Magnification Attention Convolutional Neural Networks [AI-eXplained

Multi-Magnification Attention Convolutional Neural Networks [AI-eXplained

DA-CNN-based similar terahertz signal identification for intelligent characterization of internal debonding defects of composites under high-resolution mode

With the prevalent occupation of glass fiber reinforced polymer (GFRP) composites in engineering structures, quality inspection of GFRPs is particularly urgent to evaluate their health state. As a typical damage form during the manufacturing and lifetime service of GFRP, debonding defects not only degrades the structural strength and remaining performance of composite materials, but also brings about unpredictable challenge to overall safety of the system. Recently, the combination of terahertz (THz) spectroscopy and artificial intelligence (AI) technique has emerged great potential for automatic defect identification inside composites. However, conventional AI algorithms are difficult to classify similar THz signals and may degrade THz detection accuracy of defects due to limited feature extraction capability. Here we propose a deformable attention convolutional neural network (DA-CNN) framework-based THz characterization system, in which the defect datasets are established firstly by the THz time domain spectroscopy (THz-TDS), and then the DA-CNN framework is adopted to realize the automatic defect location and imaging by accurate THz signals classification. It is worth noting that the proposed DA-CNN framework has powerful feature extraction capability to automatically identify internal GFRP defects, especially for similar THz signals at the edge of debonding defects. A series of experiments have been performed to validate the effectiveness of proposed system, which will provide a new solution for intelligent and automatic THz characterization of internal debonding defects of composites.

Automatic diagnosis of keratitis using object localization combined with cost-sensitive deep attention convolutional neural network

Keratitis is a major cause of corneal blindness worldwide. Early identification and timely treatment of keratitis can deter the disease progression, reaching a better prognosis. The diagnosis of keratitis often requires professional ophthalmologists. However, ophthalmologists are relatively scarce and unevenly distributed, especially in underserved and remote regions, making the early diagnosis of keratitis challenging. In this study, an object localization method combined with cost-sensitive deep attention convolutional neural network (OL-CDACNN) was proposed for the automated diagnosis of keratitis. First, the single shot multibox detector (SSD) algorithm was employed to automatically locate the region of conjunctiva and cornea (Conj_Cor) on the original slit-lamp image. Then, the region of Conj_Cor was classified using a cost-sensitive deep attention convolutional network (CDACNN) to identify keratitis, other cornea abnormalities, and normal cornea. A total of 12,407 slit-lamp images collected from four clinical institutions were used to develop and evaluate the OL-CDACNN. For detecting keratitis, other cornea abnormalities, and normal cornea, the OL-CDACNN model achieved area under the receiver operating characteristic curves (AUCs) of 0.998, 0.997, and 1.000, respectively, in an internal test dataset. The comparable performance (AUCs ranged from 0.981 to 0.998) was observed in three external test datasets, further verifying its effectiveness and generalizability. Due to reliable performance, our model has a high potential to provide an accurate diagnosis and prompt referral for a patient with keratitis in an automated fashion.

Text Classification Based on Neural Network Fusion

The goal of text classification is to identify the category to which the text belongs. Text categorization is widely used in email detection, sentiment analysis, topic marking and other fields. However, good text representation is the point to improve the capability of NLP tasks. Traditional text representation adopts bag-of-words model or vector space model, which loses the context information of the text and faces the problems of high latitude and high sparsity,. In recent years, with the increase of data and the improvement of computing performance, the use of deep learning technology to represent and classify texts has attracted great attention. Convolutional Neural Network (CNN), Recurrent Neural Network (RNN) and RNN with attention mechanism are used to represent the text, and then to classify the text and other NLP tasks, all of which have better performance than the traditional methods. In this paper, we design two sentence-level models based on the deep network and the details are as follows: (1) Text representation and classification model based on bidirectional RNN and CNN (BRCNN). BRCNN’s input is the word vector corresponding to each word in the sentence; after using RNN to extract word order information in sentences, CNN is used to extract higher-level features of sentences. After convolution, the maximum pool operation is used to obtain sentence vectors. At last, softmax classifier is used for classification. RNN can capture the word order information in sentences, while CNN can extract useful features. Experiments on eight text classification tasks show that BRCNN model can get better text feature representation, and the classification accuracy rate is equal to or higher than that of the prior art. (2) Attention mechanism and CNN (ACNN) model uses the RNN with attention mechanism to obtain the context vector; Then CNN is used to extract more advanced feature information. The maximum pool operation is adopted to obtain a sentence vector; At last, the softmax classifier is used to classify the text. Experiments on eight text classification benchmark data sets show that ACNN improves the stability of model convergence, and can converge to an optimal or local optimal solution better than BRCNN.

A deep learning-based approach with anti-noise ability for identification of rock microcracks using distributed fibre optic sensing data

Most of the existing deep learning-based crack identification models can achieve high accuracy when being trained and tested using data split from the same dataset with minimal noise, while perform poorly on field monitoring data with certain level of noise. This research developed a hybrid attention convolutional neural network (HACNN) for rock microcrack identification with enhanced anti-noise ability for distributed fibre optic sensing data. A hybrid attention module was designed and placed next to some certain convolutional layers to enhance the nonlinear representation ability of the proposed model. Two training interference strategies, namely small mini-batch training and adding dropout in the first convolutional layer, were employed to interfere with the training of the HACNN to enhance its robustness against noise. A series of experiments are designed based on the properties of the two training interference strategies to optimize the model parameters. Results showed that the optimized HACNN achieved higher accuracy on datasets with different signal-to-noise ratios compared to other machine learning algorithms, including the support vector machine, the multilayer perceptron, and an existing one-dimensional convolutional neural network. This research demonstrates the potential of establishing a robust DL-based model for identification of rock microcracks from noisy distributed fibre sensing optic data, even when training the model only with a smoothed dataset.

Image recognition of rice leaf diseases using atrous convolutional neural network and improved transfer learning algorithm

Image recognition of rice leaf diseases using atrous convolutional neural network and improved transfer learning algorithm

A novel rolling bearing fault diagnosis method based on parameter optimization variational mode decomposition with feature weighted reconstruction and multi-target attention convolutional neural networks under small samples.

To enhance the precision of rolling bearing fault diagnosis, an intelligent hybrid approach is proposed in this paper for signal processing and fault diagnosis in small samples. This approach is based on advanced techniques, combining parameter optimization variational mode decomposition weighted by multiscale permutation entropy (MPE) with maximal information coefficient and multi-target attention convolutional neural networks (MTACNN). First, an improved variational mode decomposition (VMD) is developed to denoise the raw signal. The whale optimization algorithm was used to optimize the penalty factor and mode component number in the VMD algorithm to obtain several intrinsic mode functions (IMFs). Second, separate MPE calculations are performed for both the raw signal and each of the IMF components obtained from the VMD decomposition; the results are used to calculate the maximum information coefficient (MIC). Subsequently, each MIC is normalized and converted to a weight coefficient for signal reconstruction. Ultimately, the reconstructed signals serve as input to the MTACNN for diagnosing rolling bearing faults. Results demonstrate that the signal processing approach exhibits superior noise reduction capability through simple processing. Furthermore, compared to several similar approaches, The method proposed for fault diagnosis achieves superior performance levels in the fault pattern recognition target and the fault severity recognition target.

Modality-based attention and dual-stream multiple instance convolutional neural network for predicting microvascular invasion of hepatocellular carcinoma.

The presence of microvascular invasion (MVI) is a crucial indicator of postoperative recurrence in patients with hepatocellular carcinoma (HCC). Detecting MVI before surgery can improve personalized surgical planning and enhance patient survival. However, existing automatic diagnosis methods for MVI have certain limitations. Some methods only analyze information from a single slice and overlook the context of the entire lesion, while others require high computational resources to process the entire tumor with a three-dimension (3D) convolutional neural network (CNN), which could be challenging to train. To address these limitations, this paper proposes a modality-based attention and dual-stream multiple instance learning(MIL) CNN. In this retrospective study, 283 patients with histologically confirmed HCC who underwent surgical resection between April 2017 and September 2019 were included. Five magnetic resonance (MR) modalities including T2-weighted, arterial phase, venous phase, delay phase and apparent diffusion coefficient images were used in image acquisition of each patient. Firstly, Each two-dimension (2D) slice of HCC magnetic resonance image (MRI) was converted into an instance embedding. Secondly, modality attention module was designed to emulates the decision-making process of doctors and helped the model to focus on the important MRI sequences. Thirdly, instance embeddings of 3D scans were aggregated into a bag embedding by a dual-stream MIL aggregator, in which the critical slices were given greater consideration. The dataset was split into a training set and a testing set in a 4:1 ratio, and model performance was evaluated using five-fold cross-validation. Using the proposed method, the prediction of MVI achieved an accuracy of 76.43% and an AUC of 74.22%, significantly surpassing the performance of the baseline methods. Our modality-based attention and dual-stream MIL CNN can achieve outstanding results for MVI prediction.

3D-STCNN: Spatiotemporal Convolutional Neural Network Based on EEG 3D Features for Detecting Driving Fatigue

Fatigue driving has become one of the main causes of traffic accidents, and driving fatigue detection based on electroencephalogram (EEG) can effectively evaluate the driver's mental state and avoid the occurrence of traffic accidents. This article evaluates a feature extraction method for extracting multiple features of EEG signals and establishes a spatiotemporal convolutional neural network (STCNN) to detect driver fatigue. Firstly, we constructed a three-dimensional feature of the EEG signal, which includes the frequency domain, time domain, and spatial features of the EEG signal. Then, we use STCNN for fatigue state classification. STCNN is composed of an attention time network based on attention mechanism and an attention convolutional neural network based on attention mechanism. In addition, we conducted fatigue driving experiments and collected EEG signals from 14 subjects in both awake and fatigued states, ultimately collecting EEG data under three different driving task loads. We conducted extensive experiments on this basis and compared the effectiveness of STCNN and six competitive methods. The results show that the classification accuracy of STCNN is 87.55%, which can effectively detect the fatigue status of drivers.

Human Activity Recognition Using Cascaded Dual Attention CNN and Bi-Directional GRU Framework.

Vision-based human activity recognition (HAR) has emerged as one of the essential research areas in video analytics. Over the last decade, numerous advanced deep learning algorithms have been introduced to recognize complex human actions from video streams. These deep learning algorithms have shown impressive performance for the video analytics task. However, these newly introduced methods either exclusively focus on model performance or the effectiveness of these models in terms of computational efficiency, resulting in a biased trade-off between robustness and computational efficiency in their proposed methods to deal with challenging HAR problem. To enhance both the accuracy and computational efficiency, this paper presents a computationally efficient yet generic spatial-temporal cascaded framework that exploits the deep discriminative spatial and temporal features for HAR. For efficient representation of human actions, we propose an efficient dual attentional convolutional neural network (DA-CNN) architecture that leverages a unified channel-spatial attention mechanism to extract human-centric salient features in video frames. The dual channel-spatial attention layers together with the convolutional layers learn to be more selective in the spatial receptive fields having objects within the feature maps. The extracted discriminative salient features are then forwarded to a stacked bi-directional gated recurrent unit (Bi-GRU) for long-term temporal modeling and recognition of human actions using both forward and backward pass gradient learning. Extensive experiments are conducted on three publicly available human action datasets, where the obtained results verify the effectiveness of our proposed framework (DA-CNN+Bi-GRU) over the state-of-the-art methods in terms of model accuracy and inference runtime across each dataset. Experimental results show that the DA-CNN+Bi-GRU framework attains an improvement in execution time up to 167× in terms of frames per second as compared to most of the contemporary action-recognition methods.

MSF-Net: A Lightweight Multi-Scale Feature Fusion Network for Skin Lesion Segmentation.

Segmentation of skin lesion images facilitates the early diagnosis of melanoma. However, this remains a challenging task due to the diversity of target scales, irregular segmentation shapes, low contrast, and blurred boundaries of dermatological graphics. This paper proposes a multi-scale feature fusion network (MSF-Net) based on comprehensive attention convolutional neural network (CA-Net). We introduce the spatial attention mechanism in the convolution block through the residual connection to focus on the key regions. Meanwhile, Multi-scale Dilated Convolution Modules (MDC) and Multi-scale Feature Fusion Modules (MFF) are introduced to extract context information across scales and adaptively adjust the receptive field size of the feature map. We conducted many experiments on the public data set ISIC2018 to verify the validity of MSF-Net. The ablation experiment demonstrated the effectiveness of our three modules. The comparison experiment with the existing advanced network confirms that MSF-Net can achieve better segmentation under fewer parameters.

Multispectral remote sensing and DANet model improve the precision of urban park vegetation detection: an empirical study in Jinhai Park, Shanghai

IntroductionThe precise detection of vegetation in urban parks is crucial for accurate carbon sink calculations and planting assessments, particularly in high-density cities. Unlike traditional methods designed for forest vegetation, the detection and classification of urban park vegetation face challenges such as unclear boundaries, multiple vegetation categories, low image resolution, labor-intensive manual calculations, and unreliable modeling results. However, by utilizing unmanned aerial vehicles (UAVs) equipped with high-resolution visible and multispectral (MS) remote sensing cameras, it becomes possible to label images with green normalized difference vegetation index (GNDVI) and full-spectral three-channel information.MethodsBy employing a dual attention convolutional neural network (DANet) model that incorporates image fusion, DANet, and feature decoding networks, the high-precision detection of urban park vegetation can be significantly improved.ResultsEmpirical validation carried out in Jinhai Park since 2021 has provided evidence of the effectiveness of the DANet model when utilizing early fusion and feature fusion techniques. This model achieves an accurate detection rate of 88.6% for trees, 92.0% for shrubs, 92.6% for ground cover, and 91.8% for overall vegetation. These detection rates surpass those achieved using only visible images (88.7%) or GNDVI images (86.6%).DiscussionThe enhanced performance can be attributed to the intelligent capabilities of the double-in network. This high-precision detection model provides more precise scientific and technical support for subsequent park carbon sink calculations, assessments of existing vegetation for planting designs, and evaluations of urban ecological impacts.

Memory long and short term time series network for ultra-short-term photovoltaic power forecasting

Photovoltaic (PV) power is stochastic, intermittent and volatile, which has brought huge challenges to the safe and stable operation of the power grid. Accurate PV power forecasting is becoming a significant task for PV plant grid connecting, scheduling and guaranteeing the safety of the power grid. To improve the accuracy of PV power forecasting, a memory long and short term time series network (MLSTNet) model is proposed to perform ultra-short-term PV power forecasting with a time horizon varying from 15 min to 4 h. Firstly, the input variables are screened by calculating Spearman correlation coefficients between weather data. To cluster different types of data, the classification coefficient is set. Secondly, different type data are automatically identified in the MLSTNet model. The appropriate model parameters and network structures are adjusted for the higher forecasting speed and accuracy. By using the single-step rolling input method and the temporal attention convolutional neural network, the ability of temporal and spatial feature extraction has been significantly enhanced, allowing real-time forecasting and dynamic optimization of the proposed model. Thirdly, long-term and short-term historical data are fed into the model, and different periods dependencies are captured while incorporating multiple variable dimensions, leading to more memorable in the network. Finally, the experimental data is selected from the actual operation data of a PV plant in northern China, which represents the universal applicability and practical application of the model. The performance of the proposed model is demonstrated with actual PV plant dataset, the RMSE of proposed model decreased by 32.73%, 5.44%, 12.44%, and 31.22%. Results indicated that it is feasible to use the proposed MLSTNet model for time series forecasting. The core idea of the MLSTNet model is to achieve higher accuracy in ultra-short-term forecasting by enhancing the ability to extract temporal features and linearity in the model, providing a new idea for deep learning-based methods for ultra-short-term PV power forecasting.

Channel-Spatial attention convolutional neural networks trained with adaptive learning rates for surface damage detection of wind turbine blades

Damage detection of wind turbine blades can provide guidance for maintaining the wind turbine system and reduce maintenance costs. Although machine vision has made good progress in blade damage detection, the complex background information brings great challenges to blade damage detection. Existing methods treat all information or features of the image equally, which may result in insufficient attention to damage features. In this paper, a novel framework of channel-spatial attention convolutional neural networks, together with an adaptive learning rate scheme, is proposed for surface damage detection of wind turbine blades. It guides the attention of the feature extraction network to focus on the blade damage feature by embedding CBAM (Convolutional Block Attention Module) to enhance the blade damage features. To optimize the training process and make it reach saturation faster, a novel adaptive learning rate scheme is also proposed. The effectiveness of the proposed is verified on real wind turbine blade image database, containing three types of damage, manually collected from a commercial wind farm. Experimental results show that the proposed method can improve binary damage classification accuracy by 2.68% and multiple class damage classification accuracy by 5.36% in comparison with the compared state-of-the-art methods.

Vision-based fruit recognition via multi-scale attention CNN

Fruit quality assessment, grading and sorting are of vital importance to fruit processing, and all these involve fruit recognition. Vision-based fruit recognition can recognize fruit automatically and further support more applications such as fruit-picking robots, self-checkout service, and dietary guidance. Recent works resort to Convolutional Neural Networks (CNNs) for vision-based fruit recognition because of their powerful expressive capacity. However, most works simply utilize existing CNNs and ignore characteristics of fruit images, resulting in sub-optimal recognition performance. To solve this problem, we adopt a Multi-Scale Attention Network (MSANet), which explores attention from different layers of CNNs and aggregates various visual attentional features from different levels into final comprehensive representations. Extensive evaluations on four fruit benchmark datasets demonstrate that the method achieves state-of-the-art recognition performance. In addition, we report various experimental results from different deep networks as baselines. Codes and models will be released upon publication. Models and code are available at http://123.57.42.89/codes/msanet.html.

Causality Extraction Cascade Model Based on Dual Labeling

Causal relation extraction is a crucial task in natural language processing. Current extraction methods have problems, including low accuracy of causal-event division and incorrect extraction of important semantic features. This study uses the bidirectional long short-term memory (BiLSTM) and attentive convolutional neural network (ACNN) models to construct a cascaded causal relationship extraction model to improve the precision of the extraction. The model uses two kinds of labels and then divides the causal event boundary after determining the relationship between the front and rear causal events. It automatically learns semantic features from sentences, reducing the dependence on external knowledge and improving the precision of extraction. The experimental results demonstrate that the precision of causality extraction can reach 81.67% and the F1 value can reach 83.2%.

Breast cancer prediction using gated attentive multimodal deep learning

Women are prone to breast cancer, which is a major cause of death. One out of every eight women has a lifetime risk of developing this cancer. Early diagnosis of this disease is critical and enhances the success rate of cure. It is extremely important to determine which genes are associated with the disease. However, too many features make studies on gene data challenging. In this study, an attention-based multimodal deep learning model was created by combining data from clinical, copy number alteration and gene expression sources. Attention-based deep learning models can analyze mammography images and identify subtle patterns or abnormalities that may indicate the presence of cancer. These models can also integrate patient data, such as age and family history, to improve the accuracy of predictions. The objective of this study is to help breast cancer prediction tasks and improve efficiency by incorporating attention mechanisms. Our suggested methodology employs multimodal data and generates insightful characteristics to improve the prediction of the prognosis for breast cancer. It is a two-phase model; the first phase generates the stacked features using a sigmoid gated attention convolutional neural network, and the second phase uses flatten, dense and dropout processes for bi-modal attention. Based on our findings, the proposed model produced successful results and has the potential to significantly improve breast cancer detection and diagnosis, ultimately leading to better patient outcomes.

LFACon: Introducing Anglewise Attention to No-Reference Quality Assessment in Light Field Space.

Light field imaging can capture both the intensity information and the direction information of light rays. It naturally enables a six-degrees-of-freedom viewing experience and deep user engagement in virtual reality. Compared to 2D image assessment, light field image quality assessment (LFIQA) needs to consider not only the image quality in the spatial domain but also the quality consistency in the angular domain. However, there is a lack of metrics to effectively reflect the angular consistency and thus the angular quality of a light field image (LFI). Furthermore, the existing LFIQA metrics suffer from high computational costs due to the excessive data volume of LFIs. In this paper, we propose a novel concept of "anglewise attention" by introducing a multihead self-attention mechanism to the angular domain of an LFI. This mechanism better reflects the LFI quality. In particular, we propose three new attention kernels, including anglewise self-attention, anglewise grid attention, and anglewise central attention. These attention kernels can realize angular self-attention, extract multiangled features globally or selectively, and reduce the computational cost of feature extraction. By effectively incorporating the proposed kernels, we further propose our light field attentional convolutional neural network (LFACon) as an LFIQA metric. Our experimental results show that the proposed LFACon metric significantly outperforms the state-of-the-art LFIQA metrics. For the majority of distortion types, LFACon attains the best performance with lower complexity and less computational time.

Diagnosis of atrial fibrillation using self-complementary attentional convolutional neural network

Background and ObjectiveAutomatic recognition of wearable dynamic electrocardiographic (ECG) signals is a difficult problem in biomedical signal processing. However, with the widespread use of long-range ambulatory ECG, a large number of real-time ECG signals are generated in the clinic, and it is very difficult for clinicians to perform timely atrial fibrillation (AF) diagnosis. Therefore, developing a new AF diagnosis algorithm can relieve the pressure on the healthcare system and improve the efficiency of AF screening. MethodsIn this study, a self-complementary attentional convolutional neural network (SCCNN) was designed to accurately identify AF in wearable dynamic ECG signals. First, a 1D ECG signal was converted into a 2D ECG matrix using the proposed Z-shaped signal reconstruction method. Then, a 2D convolutional network was used to extract shallow information from adjacent sampling points at close distances and interval sampling points at distant distances in the ECG signal. The self-complementary attention mechanism (SCNet) was used to focus and fuse channel information with spatial information. Finally, fused feature sequences were used to detect AF. ResultsThe accuracies of the proposed method on the three public databases were 99.79%, 95.51%, and 98.80%. The AUC values were 99.79%, 95.51%, and 98.77%, respectively. The sensitivity on the clinical database was as high as 99.62%. ConclusionsThese results show that the proposed method can accurately identify AF and has good generalization.

RMA-CNN: A Residual Mixed-Domain Attention CNN for Bearings Fault Diagnosis and its Time-Frequency Domain Interpretability

Early fault diagnosis of bearings is crucial for ensuring safe and reliable operations. Convolutional Neural Networks (CNNs) have achieved significant breakthroughs in machinery fault diagnosis. However, complex and varying working conditions can lead to inter-class similarity and intra-class variability in datasets, making it more challenging for CNNs to learn discriminative features. Furthermore, CNNs are often considered "black boxes" and lack sufficient interpretability in the fault diagnosis field. To address these issues, this paper introduces a Residual Mixed-Domain AttentionCNN method, referred to as RMA-CNN. This method comprises multiple ResidualMixed Domain Attention Modules (RMAMs), each employing one attention mechanism to emphasize meaningful features in both time and channel domains. This significantly enhances the network's ability to learn fault-related features. Moreover, we conduct an in-depth analysis of the inherent feature learning mechanism of the attention module RMAM to improve the interpretability of CNNs in fault diagnosis applications. Experiments conducted on two datasets—a high-speed aeronautical bearing dataset and a motor bearing dataset—demonstrate that the RMA-CNN achieves remarkable results in diagnostic tasks.