Feature Extraction and Classification of Cataluminescence Images Based on Sparse Coding Convolutional Neural Networks

The atmosphere of human existence is increasingly complex, and various harmful gases seriously endanger human health. Therefore, it is necessary to quickly and accurately detect trace toxic gases. With the application progress of cataluminescence (CTL) in the detection of harmful gases, this article...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on instrumentation and measurement Vol. 70; pp. 1 - 11
Main Authors: Shi, Guolong, He, Yigang, Zhang, Chaolong
Format: Journal Article
Language:English
Published: New York IEEE 2021
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects:
Algorithms
Artificial neural networks
Back propagation
Back propagation networks
Cataluminescence (CTL) sensor
Classification
Coding
Convolution
deep learning
Eigenvectors
Feature extraction
Gases
Image classification
Internet of Things
Kernel
Luminescence
Neural networks
Sensor arrays
sparse coding convolutional neural networks (SCNNs)
Temperature sensors
Visual fields
ISSN:0018-9456, 1557-9662
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The atmosphere of human existence is increasingly complex, and various harmful gases seriously endanger human health. Therefore, it is necessary to quickly and accurately detect trace toxic gases. With the application progress of cataluminescence (CTL) in the detection of harmful gases, this article proposed a feature extraction and classification algorithm for CTL images based on sparse coding convolutional neural networks (SCNN). First, the CTL images were obtained by the portable CTL sensor system, and the CTL images were encoded by simulating the characteristics of the visual cell receptive field, so that the sparse and internal features of the image were obtained, and the feature vectors were sorted. Then, the eigenvector with a large grayscale average gradient was selected to initialize the convolutional neural network convolution kernel. Finally, the complementarity of the feature differences between networks was measured according to the complementary measurement function, so as to optimize the weight of the back-propagation fine-tuning model of the loss function, and the accuracy of images classification was improved. The results showed that the SCNN algorithm can accurately realize the CTL images classification, further complete detection and identification of trace harmful gases.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0018-9456
1557-9662
DOI:10.1109/TIM.2020.3023508