📄 中文摘要
建筑行业产生大量废弃物，因此对其进行有效分类对于可持续废物管理和资源回收至关重要。本研究提出一种混合视觉管道，将深度特征提取与经典机器学习分类器相结合，用于自动化建筑和拆除（C&D）垃圾分类。构建了一个包含1,800张均衡、高质量图像的新数据集，涵盖了混凝土、砖块、木材、金属、塑料和玻璃六种常见C&D垃圾类型。该数据集用于训练和评估所提出的模型。在深度特征提取阶段，利用预训练的卷积神经网络（CNN）模型（如ResNet、VGG和Inception）作为特征提取器，从图像中捕捉高维、语义丰富的特征。这些深度特征随后被输入到一系列经典机器学习分类器中，包括支持向量机（SVM）、随机森林（Random Forest）、K近邻（K-NN）和朴素贝叶斯（Naive Bayes）。通过这种混合方法，旨在结合深度学习在特征表示方面的强大能力与传统机器学习模型在分类效率和可解释性方面的优势。实验结果表明，该管道在C&D垃圾分类任务上取得了显著的性能提升。特别是，结合ResNet特征与SVM分类器的模型在准确率、精确率、召回率和F1分数等多项指标上表现最佳，达到了95%以上的分类准确率。这证明了混合方法在处理复杂多样的C&D垃圾图像时的有效性，为实现更高效的自动化垃圾分类提供了有力的技术支持，有助于推动建筑废弃物的循环利用和环境可持续发展。
📄 English Summary
Hybrid Deep Feature Extraction and ML for Construction and Demolition Debris Classification
The construction industry generates substantial volumes of debris, making effective sorting and classification crucial for sustainable waste management and resource recovery. A hybrid vision-based pipeline is presented, integrating deep feature extraction with classical machine learning (ML) classifiers for automated construction and demolition (C&D) debris classification. A novel dataset comprising 1,800 balanced, high-quality images was assembled, encompassing six common C&D debris types: concrete, bricks, wood, metal, plastic, and glass. This dataset served for training and evaluating the proposed models. In the deep feature extraction phase, pre-trained Convolutional Neural Network (CNN) models, such as ResNet, VGG, and Inception, were employed as feature extractors to capture high-dimensional, semantically rich features from the images. These deep features were subsequently fed into a range of classical machine learning classifiers, including Support Vector Machines (SVM), Random Forests, K-Nearest Neighbors (K-NN), and Naive Bayes. This hybrid approach aims to combine the powerful feature representation capabilities of deep learning with the efficiency and interpretability advantages of traditional machine learning models in classification. Experimental results demonstrate significant performance enhancements on the C&D debris classification task. Notably, the model combining ResNet features with an SVM classifier achieved the best performance across multiple metrics, including accuracy, precision, recall, and F1-score, reaching over 95% classification accuracy. This validates the effectiveness of the hybrid method in handling complex and diverse C&D debris images, offering robust technical support for more efficient automated waste classification, and contributing to the circular economy of construction waste and environmental sustainability.