Generative Adversarial Network Based on LSTM and Convolutional Block Attention Module for Industrial Smoke Image Recognition

Dahai Li1, Rui Yang1 and Su Chen2

  1. School of Electronics and Electrical Engineering, Zhengzhou University of Science and Technology
    Zhengzhou 450064, China,
  2. Department of Mechanical and Electrical Engineering, Henan Vocational College of Water Conservancy and Environment
    Zhengzhou 450002, China


The industrial smoke scene is complex and diverse, and the cost of labeling a large number of smoke data is too high. Under the existing conditions, it is very challenging to efficiently use a large number of existing scene annotation data and network models to complete the image classification and recognition task in the industrial smoke scene. Traditional deep learn-based networks can be directly and efficiently applied to normal scene classification, but there will be a large loss of accuracy in industrial smoke scene. Therefore, we propose a novel generative adversarial network based on LSTM and convolutional block attention module for industrial smoke image recognition. In this paper, a low-cost data enhancement method is used to effectively reduce the difference in the pixel field of the image. The smoke image is input into the LSTM in generator and encoded as a hidden layer vector. This hidden layer vector is then entered into the discriminator. Meanwhile, a convolutional block attention module is integrated into the discriminator to improve the feature self-extraction ability of the discriminator model, so as to improve the performance of the whole smoke image recognition network. Experiments are carried out on real diversified industrial smoke scene data, and the results show that the proposed method achieves better image classification and recognition effect. In particular, the F scores are all above 89%, which is the best among all the results.

Key words

industrial smoke image recognition, generative adversarial network, LSTM, convolutional block attention module, data enhancement

Digital Object Identifier (DOI)

Publication information

Volume 20, Issue 4 (September 2023)
Year of Publication: 2023
ISSN: 2406-1018 (Online)
Publisher: ComSIS Consortium

Full text

DownloadAvailable in PDF
Portable Document Format

How to cite

Li, D., Yang, R., Chen, S.: Generative Adversarial Network Based on LSTM and Convolutional Block Attention Module for Industrial Smoke Image Recognition. Computer Science and Information Systems, Vol. 20, No. 4. (2023),