Scholarlink

<p>Solid waste management is crucial for environmental sustainability, and effective waste classification systems are vital due to the escalating volume of waste. This study proposes an innovative deep learning model that integrates EfficientNetB0, InceptionV3, and the HyperColumn technique for waste image classification. The model aims to learn multi-resolution features more effectively by combining the strengths of these architectures. EfficientNetB0 provides high performance with fewer parameters, while InceptionV3 captures multi-scale features using various filters. The HyperColumn technique fuses feature maps from different layers of both models, enabling the extraction of both general and detailed features. The XGBClassifier is employed in the classification phase to optimize final performance. Experimental results on the Household_Garbage and TrashNet datasets demonstrate the model’s superiority, achieving 99.87% accuracy, 99.77% precision, 99.81% recall, 99.79% F1-Score, 99.99% specificity, and 99.90% AUC on the Household_Garbage dataset, and 99.40% accuracy, 99.49% precision, 99.54% recall, 99.51% F1-Score, 99.88% specificity, and 99.71% AUC on the TrashNet dataset. These results highlight the model’s accuracy and balance, positioning it as an important step towards sustainable waste management through automated and efficient solutions. The code and data are available at <a href="https://github.com/GaffariCelik/SolidWasteClassification">https://github.com/GaffariCelik/SolidWasteClassification</a>.</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

Multi-layer feature fusion for high-accuracy solid waste classification using a hybrid deep learning model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月10日

查看全文
  • Gaffari Celik

摘要

Solid waste management is crucial for environmental sustainability, and effective waste classification systems are vital due to the escalating volume of waste. This study proposes an innovative deep learning model that integrates EfficientNetB0, InceptionV3, and the HyperColumn technique for waste image classification. The model aims to learn multi-resolution features more effectively by combining the strengths of these architectures. EfficientNetB0 provides high performance with fewer parameters, while InceptionV3 captures multi-scale features using various filters. The HyperColumn technique fuses feature maps from different layers of both models, enabling the extraction of both general and detailed features. The XGBClassifier is employed in the classification phase to optimize final performance. Experimental results on the Household_Garbage and TrashNet datasets demonstrate the model’s superiority, achieving 99.87% accuracy, 99.77% precision, 99.81% recall, 99.79% F1-Score, 99.99% specificity, and 99.90% AUC on the Household_Garbage dataset, and 99.40% accuracy, 99.49% precision, 99.54% recall, 99.51% F1-Score, 99.88% specificity, and 99.71% AUC on the TrashNet dataset. These results highlight the model’s accuracy and balance, positioning it as an important step towards sustainable waste management through automated and efficient solutions. The code and data are available at https://github.com/GaffariCelik/SolidWasteClassification.