Scholarlink

<p>In the new green economy and sustainable development, waste recycling and management have become a central issue for reducing the environmental impacts of waste mixtures. Poor waste recycling and management directly impact climate change, air pollution, and many environmental ecosystems. Recognizing recyclable waste materials from millions of tons of waste mixtures is a huge challenge that can be solved using artificial intelligence techniques based on machine learning and deep learning technologies. This paper introduces a deep learning approach to recognizing recyclable waste materials based on novel transfer learning models. Two hybrid transfer learning models, Xception-CNN and MobileNet-CNN, are based on the classical neural network design of Xception and MobileNet models but with new hyperparameters and an invented classification method. The proposed models have been tested and validated on a benchmark dataset consisting of 25,000 images. The obtained results clarified the superiority of <i>the MobileNet-CNN</i> technique in recognizing recyclable waste Materials compared to the other six deep learning techniques, where the MobileNet-CNN achieved a testing accuracy reached to 91%. Moreover, the MobileNet-CNN has been analyzed using the SHAP analysis technique and compared to the Xception-CNN model for two samples of the given dataset. The results of SHAP value distribution for MobileNet-CNN are more balanced around zero which makes it less sensitive to small variations in features compared to Xception-CNN. These results set the basis for future research into utilizing the two proposed transfer learning models, Xception-CNN and MobileNet-CNN, to solve additional classification problems in waste recycling issues using different datasets.</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

Recognizing recyclable waste materials based on deep transfer learning models

  • 期刊论文

  • 发表日期：

    2025年10月11日

查看全文
  • Ghada Dahy,
  • Mohamed Torky,
  • Vaclav Snasel,
  • Aboul Ella Hassanein

摘要

In the new green economy and sustainable development, waste recycling and management have become a central issue for reducing the environmental impacts of waste mixtures. Poor waste recycling and management directly impact climate change, air pollution, and many environmental ecosystems. Recognizing recyclable waste materials from millions of tons of waste mixtures is a huge challenge that can be solved using artificial intelligence techniques based on machine learning and deep learning technologies. This paper introduces a deep learning approach to recognizing recyclable waste materials based on novel transfer learning models. Two hybrid transfer learning models, Xception-CNN and MobileNet-CNN, are based on the classical neural network design of Xception and MobileNet models but with new hyperparameters and an invented classification method. The proposed models have been tested and validated on a benchmark dataset consisting of 25,000 images. The obtained results clarified the superiority of the MobileNet-CNN technique in recognizing recyclable waste Materials compared to the other six deep learning techniques, where the MobileNet-CNN achieved a testing accuracy reached to 91%. Moreover, the MobileNet-CNN has been analyzed using the SHAP analysis technique and compared to the Xception-CNN model for two samples of the given dataset. The results of SHAP value distribution for MobileNet-CNN are more balanced around zero which makes it less sensitive to small variations in features compared to Xception-CNN. These results set the basis for future research into utilizing the two proposed transfer learning models, Xception-CNN and MobileNet-CNN, to solve additional classification problems in waste recycling issues using different datasets.