Traditional Deep Learning Methods Research Articles

Transfer-learning-based multi-wavelength laser sensor for high fidelity and real-time monitoring of ambient temperature and humidity.

Multi-wavelength laser absorption spectroscopy has the advantages of superior sensitivity, accuracy, and robustness for gas sensing applications, offering an opportunity for the development of high-performance laser-based hygrothermographs. However, accurate and fast determination of gas parameters from multiple spectral features can be quite challenging in the presence of large numbers of features, measurement noise, and increasing demands for real-time measurements. To address this challenge, we propose a transfer-learning-based multi-wavelength laser absorption sensor for the quantitative and simultaneous measurement of temperature and concentration of water vapor, with a focus on real-time monitoring of ambient temperature and relative humidity (RH). A spectral simulation based on the most-updated HITRAN database was employed as the dataset for model pre-training and transfer learning. The experimental dataset was obtained from absorption measurements using a distributed feedback laser that probed multiple water absorption features within the band of 7179-7186c m -1. To evaluate the sensor performance, mean absolute error, error distribution, and linearity were selected. In the presence of an insufficient experimental dataset for direct data training, the proposed transfer learning approach outperformed the traditional deep learning method with a lower prediction error of 0.14°C and 0.42% for temperature and RH, respectively, as compared to the values of 0.84°C and 0.66% obtained using the traditional deep learning method. Finally, the fast data post-processing performance of the proposed transfer learning approach was demonstrated in a field test against the conventional baseline fitting method.

Damage detection of bridges subjected to moving load based on domain-adversarial neural network considering measurement and model error

Bridge damage detection methods based on moving load induced response and deep learning are popular, and they are effective when the training data generated by a finite element model (FEM) and the testing data measured from the actual bridge have the same distribution. However, their distributions are normally different due to the measurement and model error. This paper proposes a Domain-adversarial Neural Network (DANN) based damage detection method for bridge structures. The moving load induced displacement responses of the FEM and the actual bridge are taken as network inputs, and the feature space with the minimum differentiation between the edge distributions of the FEM (source domain) and the actual bridge (target domain) is extracted by adversarial training between the domain discriminator and the feature extractor. This feature space is a domain invariant feature of the source domain and the target domain and contains damage information only. Thus the damage label information is transferred from the source domain to the target domain for damage detection. Numerical and experimental examples manifest that the proposed method can reduce the distribution gap between the source domain and the target domain, and achieve higher damage detection accuracy than that of the traditional deep learning method by using a small number of sensors without a large amount of damage label data.

Enhancing identification of digital rock images using super-resolution deep neural network

Identification of pore and minerals from digital rock images plays an important role in characterization of oil and gas reservoirs. Traditional methods usually require high costs of time and labor. With the rapid development of artificial intelligence, deep learning methods have been applied to geosciences. Among them, semantic segmentation neural networks are used to identify rock images, but traditional semantic segmentation networks require high-resolution input images to improve performance and can only generate masks with the same resolutions as the input images, which again requires high-performance equipment and increases computational and equipment costs. In this paper, we propose an approach to enhancing identification of digital rock images with super-resolution neural network (SSRN) which enables low-resolution input images to generate high-resolution representations. We use SSRN to train the 10-classification thin-section images, and compare it with traditional deep learning methods. It indicates that SSRN can effectively improve the performance of the generated high-resolution masks.

Fault Identification of U-Net Based on Enhanced Feature Fusion and Attention Mechanism

Accurate fault identification is essential for geological interpretation and reservoir exploitation. However, the unclear and noisy composition of seismic data makes it difficult to identify the complete fault structure using conventional methods. Thus, we have developed an attentional U-shaped network (EAResU-net) based on enhanced feature fusion for automated end-to-end fault interpretation of 3D seismic data. EAResU-net uses an enhanced feature fusion mechanism to reduce the semantic gap between the encoder and decoder and improve the representation of fault features in combination with residual structures. In addition, EAResU-net introduces an attention mechanism, which effectively suppresses seismic data noise and improves model accuracy. The experimental results on synthetic and field data demonstrate that, compared with traditional deep learning methods for fault detection, our EAResU-net can achieve more accurate and continuous fault recognition results.

Cross-modal multiscale multi-instance learning for long-term ECG classification

At present, deep learning models have been widely used in electrocardiogram (ECG) classification. However, when processing ECG signals over a long period of time, it is difficult for existing deep learning methods to effectively diagnose and classify ECG signals based on a few abnormal heartbeats in these long-term time series data. In this paper, we present a novel cross-modal multiscale multi-instance learning approach for long-term electrocardiogram (ECG) classification. Our method leverages the foundational paradigm of multi-instance learning (MIL) to tackle the challenge of processing long-term ECG signals, while incorporating cross-modal information to guide the global integration of instances. Furthermore, to enhance model performance and bolster its robustness, we introduce multiscale data into the modeling process. The experimental results show that our model achieves excellent performance in different patterns of different datasets, outperforming a series of other traditional deep learning methods and MIL-based methods. Our model achieves impressive F1 scores of 0.8512 and 0.8293 on the St. Petersburg INCART Arrhythmia dataset and MIT-BIH Arrhythmia dataset, respectively. In the context of intrapatient patterns, our model outperforms the suboptimal models, exhibiting improvements of 1.41% and 0.48% on F1, and 2.07% and 0.93% on Recall. Similarly, for interpatient patterns, our model demonstrates superior performance, surpassing the suboptimal models with substantial gains of 4.21% and 3.07% on F1, and 4.71% and 0.55% on Recall.

Subdomain Adaptation Capsule Network for Partial Discharge Diagnosis in Gas-Insulated Switchgear.

Deep learning methods, especially convolutional neural networks (CNNs), have achieved good results in the partial discharge (PD) diagnosis of gas-insulated switchgear (GIS) in the laboratory. However, the relationship of features ignored in CNNs and the heavy dependance on the amount of sample data make it difficult for the model developed in the laboratory to achieve high-precision, robust diagnosis of PD in the field. To solve these problems, a subdomain adaptation capsule network (SACN) is adopted for PD diagnosis in GIS. First, the feature information is effectively extracted by using a capsule network, which improves feature representation. Then, subdomain adaptation transfer learning is used to accomplish high diagnosis performance on the field data, which alleviates the confusion of different subdomains and matches the local distribution at the subdomain level. Experimental results demonstrate that the accuracy of the SACN in this study reaches 93.75% on the field data. The SACN has better performance than traditional deep learning methods, indicating that the SACN has potential application value in PD diagnosis of GIS.

Tea Leaf Disease Classification using Domain Adaptation Method

Tea trees are extremely vulnerable to diseases and insect pests in the growth process, which seriously affects the yield and quality of tea leaves. This requires the identification and treatment of infected tea leaves in time. With the rapid development of artificial intelligence and computer vision, there is a new trend to identify the images by computer. However, there often exists environmental changes such as illumination intensity, sample angles and background in the leaf images collected in natural scenes, which results in the difference of data distribution. This makes the traditional deep learning method unable to solve the problem of cross-domain classification very well, thus seriously affecting the accuracy of classification. This study mainly focuses on the cross-domain task with unaligned data distribution. A domain-adaptive based method was proposed to realize the cross-domain classification of tea leaf diseases. The experimental results verify the effectiveness of the presented method and provide new thoughts for the cross-domain classification problem in agriculture.

Fine-grained Traffic Classification Based on Improved Residual Convolutional Network in Software Defined Networks

As Internet traffic becomes increasingly complex, a growing number of applications are carrying different services. To provides service-aware traffic classification, allowing network operators to better analyze network composition as well as manage and schedule efficiently network resources to facilitate sustainable development network, fine-grained traffic classification has become a crucial and challenging problem. Software defined networks has become the most promising new network architecture in the future with two major advantages of centralized control and programmability. Deep learning has also become a mainstream technology by its excellent feature extraction ability for large datasets. Thus the combination of software defined networks and deep learning can effectively solve the challenge of data set collection and feature extraction in the field of traffic classification. Inspired by research in computer vision, in this paper we propose an improved residual convolutional network approach to network traffic classification, which addresses the network degradation problem that occurs with increasing network depth in traditional deep learning methods for fine-grained network traffic identification, and can effectively learn deeper network features to achieve fine-grained network traffic classification. Experimental results show that the overall accuracy of the proposed method can reach 99.93%, which is about 4%, 13%, 7% and 2% higher than other state-of-the-art models such as classical residual networks ResNet-18, One-dimensional Convolutional Neural Networks (1D-CNN), Long Short-Term Memory (LSTM) and combined model CNN-LSTM respectively, thus verify the effectiveness of our method.

Data-driven few-shot crop pest detection based on object pyramid for smart agriculture

Farmers are often annoyed by crop pests, which can lead to serious losses if not detected in time. Deep learning has gradually penetrated into the agricultural field due to its proven superior performance, promoting the development of smart agriculture. Its real-time performance and effectiveness can help farmers spot pests in time and reduce losses. However, with the increasing application of deep learning technology, its disadvantage of relying heavily on large amounts of data has gradually been exposed. Before training a network, people need to spend a lot of manpower, time, and money to prepare a complete dataset and make the correct labels. To solve this problem, we introduced a less-shot object detection method in crop pest detection. We provide the first few-shot pest detection dataset (FSPD), which contains 20 categories and 2000 samples for this research. We use a few-shot object detection network based on an object pyramid called multiscale positive sample refinement. Numerous experiments are conducted to prove that, with only few (1/2/3/5/10-shot) samples, the accuracy of our method on FSPD is better than that of traditional deep learning methods, such as YOLO and faster R-CNN. After embedding the object pyramid branch, the network performance improves by an average of 5.9% on three splits compared with that without embedding. With input of only 10-shot novel samples, our method can achieve 76%, 79%, and 82% for the mAP on three splits, respectively. As a research study on the detection of crop diseases and pests with less radiation, our work can give some new insight to the future scholars, which is of great significance to the development of smart agriculture.

Seismic impedance inversion based on deep learning with geophysical constraints

Seismic inversion plays an essential role in the exploration and development of oil and gas reservoirs. With the development of neural networks, deep learning has achieved a wide range of applications in seismic inversion due to its powerful feature extraction and nonlinear fitting capabilities. However, when applying the traditional deep learning methods to seismic inversion based on trace-by-trace, problems such as overfitting and poor continuity of prediction results are prone to occur. To address these problems, we propose a closed-loop Unet with geophysical constraints based on spatial background information, called SG-CUnet. In SG-CUnet, sparse reflection coefficients and seismic forward modeling process are used as geophysical constraints of the network to improve the stability and accuracy of prediction results. Meanwhile, the spatial information of seismic data is added to the training process of SG-CUnet to improve the lateral continuity of prediction results. The effectiveness of proposed SG-CUnet is verified by synthetic Marmousi2 and field data examples. In synthetic data applications, the SG-CUnet proposed in this paper can give more accurate impedance estimation results than other methods. Furthermore, in the field area application, we combine the proposed SG-CUnet with a transfer learning strategy for semi-supervised training to solve the limited labeled data problem. The predicted results show that the semi-supervised SG-CUnet has better reservoir structure details and lateral continuity.

A Study of Weather-Image Classification Combining VIT and a Dual Enhanced-Attention Module

A weather-image-classification model combining a VIT (vision transformer) and dual augmented attention module is proposed to address the problems of the insufficient feature-extraction capability of traditional deep-learning methods with the recognition accuracy still to be improved and the limited types of weather phenomena existing in the dataset. A pre-trained model vision transformer is used to acquire the basic semantic feature representation of weather images. Dual augmented attention combined with convolutional self-attention and Atrous self-attention modules are used to acquire the low-level and high-level deep-image semantic representations, respectively, and the feature vectors are spliced and fed into the linear layer to obtain the weather types. Experimental validation is performed on the publicly available standard weather-image datasets MWD (Multi-class Weather Database) and WEAPD (Weather Phenomenon Database), and the two datasets are combined to enhance the comprehensiveness of the model for weather-phenomena recognition. The results show that the model achieves the highest F1 scores of 97.47%, 87.69% and 92.73% on the MWD, WEAPD and merged datasets, respectively. These scores are higher than the scores of recent deep-learning models with excellent performance in the experimental comparison, thereby, proving the effectiveness of the model.

Locating the source of self-potential using few-shot learning

Self-potential method has the advantages of convenience, rapidity and sensitivity. It has many applications in geophysical exploration. Therefore, the inversion problem of self-potential is very important. In order to obtain the location of the source, we propose a new self-potential inversion method based on few-shot learning. Firstly, a fully-connected neural network is used to extract the features of self-potential data and the data of well-surface pole–pole device method, and then the cosine distance is used to calculate the similarity between them. So, we can transform the positioning problem on the two-dimensional (2D) profile into the classification problem in two directions. In order to verify the feasibility of this method, we carried out a simulated landfill leakage experiment and a field saline water diffusion experiment, and both achieved good results. Compared with self-potential tomography method, the positioning accuracy of our method is increased by about 10.42%. Compared with traditional deep learning methods, we combine multimodal data to make the results more interpretable. In addition, the results of the two experiments also prove that our method has strong generalization ability, which provides a new idea to solve the differences between the simulation model and the actual field.

Recorded recurrent deep reinforcement learning guidance laws for intercepting endoatmospheric maneuvering missiles

This work proposes a recorded recurrent twin delayed deep deterministic (RRTD3) policy gradient algorithm to solve the challenge of constructing guidance laws for intercepting endoatmospheric maneuvering missiles with uncertainties and observation noise. The attack-defense engagement scenario is modeled as a partially observable Markov decision process (POMDP). Given the benefits of recurrent neural networks (RNNs) in processing sequence information, an RNN layer is incorporated into the agent’s policy network to alleviate the bottleneck of traditional deep reinforcement learning methods while dealing with POMDPs. The measurements from the interceptor’s seeker during each guidance cycle are combined into one sequence as the input to the policy network since the detection frequency of an interceptor is usually higher than its guidance frequency. During training, the hidden states of the RNN layer in the policy network are recorded to overcome the partially observable problem that this RNN layer causes inside the agent. The training curves show that the proposed RRTD3 successfully enhances data efficiency, training speed, and training stability. The test results confirm the advantages of the RRTD3-based guidance laws over some conventional guidance laws.

Semantic and spatial‐spectral feature fusion transformer network for the classification of hyperspectral image

AbstractRecently, transformer‐based networks have been introduced for the classification of hyperspectral image (HSI). Although transformer‐based methods can well capture spectral sequence information, their ability to fuse different types of information contained in HSI is still insufficient. To exploit rich spectral, spatial and semantic information in HSI, a novel semantic and spatial‐spectral feature fusion transformer (S3FFT) network is proposed in this study. In the proposed S3FFT method, spatial attention and efficient channel attention (ECA) modules are employed for the extraction of shallow spatial‐spectral features. Then, a transformer‐based module is designed to extract advanced fused features and to produce the pseudo‐label and class probability of each pixel for semantic feature extraction. Finally, the semantic, spatial and spectral features are combined by the transformer for classification. Compared with traditional deep learning methods and recently transformer‐based methods, the proposed S3FFT shows relatively better results on three HSI datasets.

An anti-noise fault diagnosis approach for rolling bearings based on multiscale CNN-LSTM and a deep residual learning model

Bearing fault vibration signals collected in real engineering cases often contain environmental noise which can easily mask the fault type characteristics of vibration signals, making it difficult to determine the corresponding fault type when traditional deep learning methods are used for fault diagnosis. To solve the above problem, a neural network model named multiscale CNN-LSTM (convolutional neural network-long short-term memory) and a deep residual learning model was designed, which combines a multiscale wide CNN-LSTM module and a deep residual module for rolling bearing fault diagnosis. In this model, a wide convolution kernel CNN-LSTM structure with different convolution scales is used to extract a variety of different types of frequency and sequential features from vibration signals. It is worth noting that the wide convolution kernel CNN-LSTM structure not only has stronger feature extraction performance compared with the common convolution layer but can also reduce the interference of high-frequency noise. Moreover, the deep residual module with a wide convolution kernel CNN-LSTM structure is used to further improve the feature expression ability of the proposed model. The above algorithm enables the proposed model to better extract the fault features hidden in the noise signal. When compared with some state-of-the-art methods, the experimental results showed that this model has better anti-noise performance and better generalization ability for rolling bearing fault diagnosis.

An Adaptive Kernels Layer for Deep Neural Networks Based on Spectral Analysis for Image Applications.

As the pixel resolution of imaging equipment has grown larger, the images' sizes and the number of pixels used to represent objects in images have increased accordingly, exposing an issue when dealing with larger images using the traditional deep learning models and methods, as they typically employ mechanisms such as increasing the models' depth, which, while suitable for applications that have to be spatially invariant, such as image classification, causes issues for applications that relies on the location of the different features within the images such as object localization and change detection. This paper proposes an adaptive convolutional kernels layer (AKL) as an architecture that adjusts dynamically to images' sizes in order to extract comparable spectral information from images of different sizes, improving the features' spatial resolution without sacrificing the local receptive field (LRF) for various image applications, specifically those that are sensitive to objects and features locations, using the definition of Fourier transform and the relation between spectral analysis and convolution kernels. The proposed method is then tested using a Monte Carlo simulation to evaluate its performance in spectral information coverage across images of various sizes, validating its ability to maintain coverage of a ratio of the spectral domain with a variation of around 20% of the desired coverage ratio. Finally, the AKL is validated for various image applications compared to other architectures such as Inception and VGG, demonstrating its capability to match Inception v4 in image classification applications, and outperforms it as images grow larger, up to a 30% increase in accuracy in object localization for the same number of parameters.

Investigation of deep transfer learning for cross-turbine diagnosis of wind turbine faults

Data-driven fault diagnosis of wind turbines has gained popularity, and various deep learning models have been developed accordingly with massive amounts of data and achieved an excellent diagnosis performance. However, most existing deep learning models require a similar distribution of both training and testing data, thus the trained model cannot generalize new wind turbines with different data distributions. In addition, there are insufficient fault data in practice, and therefore the cost of training a new model from scratch is extremely high. To solve these problems, a cross-turbine fault diagnosis method based on deep transfer learning is proposed for wind turbines with the available supervisory control and data acquisition (SCADA) data. To better capture the spatial features of SCADA data, a deep multi-scale residual attention convolutional neural network (DMRACNN) is first designed. Then, the distribution differences between the source and target domain data are aligned at feature level. Specifically, we investigate the transfer performance of four different domain adaptation metrics. We evaluate our proposed method using SCADA data from two wind turbines to compare the diagnostic performance of four basic networks combined with four transfer metrics. Compared with traditional deep learning methods, our proposed DMRACNN achieved significant performance improvements. A cross-validation experiment using two turbines demonstrates the strong generalization ability of the proposed method.

Classification of COVID-19 with Belief Functions and Deep Neural Network.

At present, the entire world has suffered a lot due to the spike of COVID disease. Despite the world has been developed with so much of technology in the domain of medicine, this is a very huge challenge in all over the world. Though, there is a rapid development in medical field, those are not even sufficient to diagnose the symptoms of this COVID in earlier stage. Since the spread of this disease in all over the world, it affects the livelihood of the human. Computed Tomography (CT) images have given necessary data for the radio diagnostics to detect the COVID cases. Therefore, this paper addressed about the classification techniques to diagnose about the symptoms of this virus with the help of belief function with the support of convolution neural networks. This method initially extracts the features and correlates the features with the belief maps to decide about the classification. This research work would provide classification of more accuracy than the earlier research. Therefore, compared with the traditional deep learning method, this proposed procedure would be more efficient with desirable results achieved for accuracy as 0.87, an F1 of 0.88, and 0.95 as AUC.

Direction Consistency‐Guided Lightweight Power Line Detection Network for Aerial Images

Accurate detection of power lines in aerial images is of great significance in ensuring grid security. However, complex power line scenarios and the thin and light structure of power lines both make it difficult to detect power lines accurately. Most of the existing approaches use traditional deep learning methods, using networks with a large number of parameters, computation, and memory occupation, thus making them not lightweight enough to perform on mobile devices. Based on this, a lightweight power line detection network based on direction consistency and location attention is proposed. The network is designed with a coordinate‐aware feature extraction layer, which performs feature extraction by four‐layer stacking to achieve faster inference speed while ensuring the network has fewer parameters. This layer is also able to sense the coordinates of the center pixel of the convolution in the image during the convolution process, thus preserving the location information of the power lines. In order to enhance the power line representation, a two‐stage context‐guided module is later utilized to simultaneously learn local features, surrounding context, and global context. Then, the features are input into a Gaussian kernel estimation module and features are aggregated in the corresponding directions through Gaussian kernels of eight different directions. The main directions of the power lines in the image and the corresponding Gaussian convolution kernels are obtained by filtering the feature responses. In addition, a kernel‐guided decoder module is proposed to take advantage of the estimated power line features in the main direction of Gaussian kernel aggregation. This module can effectively enhance the power line representation and maintain the continuity of power lines. Meanwhile, low‐level features are introduced to recover the edge details to realize high performance in distinguishing dense power lines. Both ablation experiments and comparison experiments on the transmission towers and power lines aerial‐image and Power Line Aerial Image Dataset show that the proposed power line detection network has a good segmentation performance in complex scenarios. The proposed method performs the best in the comparison experiments, improving over the suboptimal method by 3.51% on average for the max F‐measure metric.

SF-CNN: Deep Text Classification and Retrieval for Text Documents

Researchers and scientists need rapid access to text documents such as research papers, source code and dissertations. Many research documents are available on the Internet and need more time to retrieve exact documents based on keywords. An efficient classification algorithm for retrieving documents based on keyword words is required. The traditional algorithm performs less because it never considers words’ polysemy and the relationship between bag-of-words in keywords. To solve the above problem, Semantic Featured Convolution Neural Networks (SF-CNN) is proposed to obtain the key relationships among the searching keywords and build a structure for matching the words for retrieving correct text documents. The proposed SF-CNN is based on deep semantic-based bag-of-word representation for document retrieval. Traditional deep learning methods such as Convolutional Neural Network and Recurrent Neural Network never use semantic representation for bag-of-words. The experiment is performed with different document datasets for evaluating the performance of the proposed SF-CNN method. SF-CNN classifies the documents with an accuracy of 94% than the traditional algorithms.