Spatial Correlation and Value Prediction in Convolutional Neural Networks

Shomron, Gil; Weiser, Uri

doi:10.1109/lca.2018.2890236

Cited by 48 publications

(20 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Selected (2,2) as. The kernel size in the 2D convolutional and the Maxpooling layers produced improved performance so the (2,2) kernel size was selected instead of the commonly used (3,3) kernel [31]. The kernel size in the 1D convolutional layer is larger than that of the 2D convolutional layer to capture the basic components with the general characteristics of the previously extracted image.…”

Section: Scrlstm Model For Fault Diagnosismentioning

confidence: 99%

Supervised-Learning-Based Intelligent Fault Diagnosis for Mechanical Equipment

Hong

Suh

2021

IEEE Access

View full text Add to dashboard Cite

Recently, anomaly detection for improving the productivity of machinery in industrial environments has drawn considerable attention. As large-scale data collection and processing are becoming easier owing to technological developments, data-based deep-learning technology is being developed to detect anomalies in mechanical equipment operation. This study proposes an ensemble model that combines stacked two-dimensional and one-dimensional convolutional neural networks (CNNs), residual long shortterm memory (LSTM), and LSTM based on supervised learning. The model, which is called the SCRLSTM model, can detect abnormal data generated by mechanical equipment. The proposed model can extract the spatial features of data using a CNN model and detect anomalous states in the time-series-based vibration datasets of machinery under various environments through residual LSTM. To verify this model, data augmentation was applied to the original time-series-based mechanical vibration dataset, which had unbalanced samples that lowered the performance of the abnormal anomaly detection model. In addition, an image-based analysis was performed by converting time-series-based raw-signal data to Mel-spectrogram images, thereby achieving better performance in the fault diagnosis system to which data augmentation was applied. The proposed SCRLSTM model shows better performance than other supervised-learning-based models on datasets having different lengths under various conditions. This indicates that the proposed anomaly detection model can be expected to improve the productivity of mechanical equipment in industrial settings.

show abstract

Section: Scrlstm Model For Fault Diagnosismentioning

confidence: 99%

Supervised-Learning-Based Intelligent Fault Diagnosis for Mechanical Equipment

Hong

Suh

2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…With the rapid development of the graphics processing units (GPUs), DL technology is widely used in image processing, computer vision, medical imaging, robot control and other fields, bringing new opportunities for semantic segmentation technology. Compared with traditional hand-designed features such as Haar, local binary patterns (LBP), histogram of oriented gradient (HOG) and scale-invariant feature transform (SIFT), deep convolutional neural network (DCNN) has richer learning features and stronger expressive ability [19]. DL has been widely used in various computer vision tasks [20][21][22].…”

Section: Related Workmentioning

confidence: 99%

Image semantic segmentation method based on GAN network and ENet model

Li¹

2021

J. eng.

View full text Add to dashboard Cite

Currently, image semantic segmentation has problems such as low accuracy and long running time. This paper proposes an image semantic segmentation method based on generative adversarial network and ENet model combined with deep neural network. This method first improves the network model of generative adversarial network. Ensure the high resolution of the generated image and achieve high similarity with the real image. While ensuring the high accuracy of image semantic segmentation, it effectively improves the real-time performance of network processing. The proposed method is verified based on public data sets. The experimental results show that the segmentation accuracy of this method can reach more than 93%, and the simulation running time is less than 0.171 s, which shows good high accuracy and high real-time performance. A feasible strategy is proposed for the further productization of semantic segmentation.

show abstract

“…Each convolutional layer in CNN has a pooling layer for downsampling after feature extraction. This unique structure makes the network have high distortion tolerance to the input samples when it is recognised [24], and makes the CNN more suitable for dealing with multi-channel 2D sequences like the three-channel 2D greyscale image above.…”

Section: Cnn Modellingmentioning

confidence: 99%

Recognition method of voltage sag causes based on two‐dimensional transform and deep learning hybrid model

Zheng

Wang

et al. 2020

IET Power Electronics

View full text Add to dashboard Cite

The voltage sags' caused recognition is the basis for formulating governance plans and clarifying liabilities for accidents. The diversification of smart grid equipment, the grid-connected power generation of new energy sources and the regional differentiation of power consumption modes pose new challenges to the traditional methods. In this study, a method based on deep learning hybrid model is proposed. The convolutional neural network is used to flexibly receive the voltage after two-dimensional transformation, so as to automatically obtain the time series and spatial characteristics of the voltage sag signals. The deep belief network is used to replace the fully connected layers in convolutional neural network, thereby enhancing the multi-label classification ability of the model. The parameters obtained by the unsupervised training of the stacked sparse denoising auto-encoder are used to initialise the weight of deep belief network, thereby improving the convergence speed and the anti-noise performance of the model. Iterative training and repeated testing of the network using pre-processed simulation data and actual recorded data verify the high recognition accuracy and strong anti-noise performance of the hybrid model. Compared with the traditional methods, the hybrid model also has good generalisation ability and can be effectively applied in practical engineering.

show abstract

Spatial Correlation and Value Prediction in Convolutional Neural Networks

Cited by 48 publications

References 10 publications

Supervised-Learning-Based Intelligent Fault Diagnosis for Mechanical Equipment

Supervised-Learning-Based Intelligent Fault Diagnosis for Mechanical Equipment

Image semantic segmentation method based on GAN network and ENet model

Recognition method of voltage sag causes based on two‐dimensional transform and deep learning hybrid model

Contact Info

Product

Resources

About