scholarly journals Why Dilated Convolutional Neural Networks: A Proof of Their Optimality

Entropy ◽  
2021 ◽  
Vol 23 (6) ◽  
pp. 767
Author(s):  
Jonatan Contreras ◽  
Martine Ceberio ◽  
Vladik Kreinovich
Keyword(s):  
Neural Networks ◽  
Image Processing ◽  
Convolutional Neural Networks ◽  
Output Signal ◽  
Optimality Criteria ◽  
Image Processing Techniques ◽  
Dilated Convolution ◽  
Input Signals ◽  
Processing Techniques ◽  
Better Than

One of the most effective image processing techniques is the use of convolutional neural networks that use convolutional layers. In each such layer, the value of the layer’s output signal at each point is a combination of the layer’s input signals corresponding to several neighboring points. To improve the accuracy, researchers have developed a version of this technique, in which only data from some of the neighboring points is processed. It turns out that the most efficient case—called dilated convolution—is when we select the neighboring points whose differences in both coordinates are divisible by some constant ℓ. In this paper, we explain this empirical efficiency by proving that for all reasonable optimality criteria, dilated convolution is indeed better than possible alternatives.

Videoscope-based inspection of turbofan engine blades using convolutional neural networks and image processing

2019 ◽  
Vol 18 (5-6) ◽  
pp. 2020-2039 ◽  
Author(s):  
Yong-Ho Kim ◽  
Jung-Ryul Lee
Keyword(s):  
Neural Networks ◽  
Image Processing ◽  
Convolutional Neural Networks ◽  
Aircraft Engine ◽  
Recognition Algorithm ◽  
Inspection Method ◽  
Image Processing Techniques ◽  
Damage Recognition ◽  
Using Data ◽  
Processing Techniques

A typical aircraft engine consists of fans, compressors, turbines, and so on, and each is made of multiple layers of blades. Discovering the site of damages among the large number of blades during aircraft engine maintenance is quite important. However, it is impossible to look directly into the engine unless it is disassembled. For this reason, optical equipment such as a videoscope is used to visually inspect the blades of an engine through inspection holes. The videoscope inspection method has some obvious drawbacks such as the long-time attention on microscopic video feed and high labor intensity. In this research, we developed a damage recognition algorithm using convolutional neural networks and some image-processing techniques related to feature point extraction and matching in order to improve the videoscope inspection method. The image-processing techniques were mainly used for the preprocessing of the videoscope images, from which a suspected damaged region is selected after the preprocessing. The suspected region is finally classified as damaged or normal by the pre-trained convolutional neural networks. We trained the convolutional neural networks 2000 times by using data from 380 images and calculated the classification accuracy using data from 40 images. After repeating the above procedure 50 times with the data randomly divided into training and test groups, an average classification accuracy of 95.2% for each image and a damage detectability of 100% in video were obtained. For verification of the proposed approach, the convolutional neural network part was compared with the traditional neural network, and the preprocessing was compared with the region proposal network of the faster region–based convolutional neural networks. In addition, we developed a platform based on the developed damage recognition algorithm and conducted field tests with a videoscope for a real engine. The damage detection AI platform was successfully applied to the inspection video probed in an in-service engine.

scholarly journals Use of Deep Learning in Verifying the Functioning of LEDs

2020 ◽  
Vol 8 (7) ◽  
pp. 94-105
Author(s):  
Raimundo C de Oliveira ◽  
Thatielen Oliveira Pereira ◽  
Antonio Claudio Kieling
Keyword(s):  
Neural Networks ◽  
Image Processing ◽  
Deep Learning ◽  
Convolutional Neural Networks ◽  
Production Line ◽  
False Positives ◽  
Small Computer ◽  
Image Processing Techniques ◽  
Processing Techniques

This article aims to bring an alternative to carrying out manual tests of devices mounted on a production line. One of the tests done by the operator is to find out if the LEDs are present on the device being turned on and working correctly. Image processing techniques were applied in the elaboration of the dataset and the use of Convolutional Neural Networks for the classification of the colors presented by the LEDs and the recognition of their operation. An accuracy of 99.25% was obtained with a low percentage of false positives and true negatives. There were no difficulties in porting the model built to a small computer.

scholarly journals Why Dilated Convolutional Neural Networks: A Proof of Their Optimality

2021 ◽  
Author(s):  
Jonatan Contreras ◽  
Martine Ceberio ◽  
Vladik Kreinovich
Keyword(s):  
Neural Networks ◽  
Convolutional Neural Networks ◽  
Optimality Criteria ◽  
Image Processing Techniques ◽  
Optimal Subset ◽  
Speed Up ◽  
Grid Points ◽  
Original Grid ◽  
Processing Techniques ◽  
Reasonable Sense

One of the most effective image processing techniques is the use of convolutional neural networks, where we combine intensity values at grid points in the vicinity of each point. To speed up computations, researchers have developed a dilated version of this technique, in which only some points are processed. It turns out that the most efficient case is when we select points from a sub-grid. In this paper, we explain this empirical efficiency proving that the sub-grid is indeed optimal – in some reasonable sense. To be more precise, we prove that all reasonable optimality criteria, the optimal subset of the original grid is either a sub-grid, or a sub-grid-like set.

The Application of Image Recognition Based on BP Neural Networks in Automatic Steel Rolling

2013 ◽  
Vol 764 ◽  
pp. 161-164
Author(s):  
Wei Jiang
Keyword(s):  
Neural Networks ◽  
Image Processing ◽  
Image Recognition ◽  
Character Recognition ◽  
Video Image ◽  
Bp Neural Networks ◽  
Steel Rolling ◽  
Image Processing Techniques ◽  
Processing Techniques ◽  
Frame Grabber

A BP neural networks is presented for billet character recognition. Firstly, by a series of image processing techniques, the character’feature in the billet character region of the video image gathered by frame grabber is abstracted. Secondly, the BP neural networks algorithm is employed for character recognition. Application results show that the image recognition based BP neural networks can performs well in billet character recognition, and the method presented is speedy, efficient and of high value in practice.

Sign in / Sign up

Export Citation Format

Share Document