Scholarlink

<p>Considerable research has been directed toward pollution issues, leading to the development of various methods for gas identification. One effective technique involves the use of an electronic nose paired with deep learning (DL) algorithms, facilitating easy and accurate gas detection. A commonly used model in this domain is LeNet5, known for its simple and effective design as well as its relatively short implementation time. Typically, research in this field focuses on one-dimensional neural networks (1DNN) that use temporal inputs. In our study, we introduce an enhanced version of the LeNet5 model to classify four types of gases: no-gas, smoke, perfume, and a mixture of smoke and perfume, using infrared images. The dataset comprises 6400 infrared images. To improve classification accuracy, we applied preprocessing steps such as image augmentation, normalization, and resizing to a fixed size. Our results show that the original LeNet5 model achieves a performance rate of 93.35% for recognizing the four gas classes, whereas the modified LeNet5 model achieves an average performance rate of 96.64% , demonstrating its effectiveness.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Improved LeNet5 Model-Based Pollutant Gases Detection Using Electronic Nose and Infrared Images

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Rabeb Faleh,
  • Madallah Alruwaili

摘要

Considerable research has been directed toward pollution issues, leading to the development of various methods for gas identification. One effective technique involves the use of an electronic nose paired with deep learning (DL) algorithms, facilitating easy and accurate gas detection. A commonly used model in this domain is LeNet5, known for its simple and effective design as well as its relatively short implementation time. Typically, research in this field focuses on one-dimensional neural networks (1DNN) that use temporal inputs. In our study, we introduce an enhanced version of the LeNet5 model to classify four types of gases: no-gas, smoke, perfume, and a mixture of smoke and perfume, using infrared images. The dataset comprises 6400 infrared images. To improve classification accuracy, we applied preprocessing steps such as image augmentation, normalization, and resizing to a fixed size. Our results show that the original LeNet5 model achieves a performance rate of 93.35% for recognizing the four gas classes, whereas the modified LeNet5 model achieves an average performance rate of 96.64% , demonstrating its effectiveness.