[论文解读] Hybrid Deep Feature Extraction and ML for Construction and Demolition Debris Classification
本论文提出一种混合管线,使用基于 Xception 的深特征与经典机器学习分类器来对四类建筑与拆除(C&D) debris 进行分类,在平衡的现场数据集上实现约 99.5% 的准确率。
The construction industry produces significant volumes of debris, making effective sorting and classification critical for sustainable waste management and resource recovery. This study presents a hybrid vision-based pipeline that integrates 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 representing four material categories, Ceramic/Tile, Concrete, Trash/Waste, and Wood was collected from real construction sites in the UAE, capturing diverse real-world conditions. Deep features were extracted using a pre-trained Xception network, and multiple ML classifiers, including SVM, kNN, Bagged Trees, LDA, and Logistic Regression, were systematically evaluated. The results demonstrate that hybrid pipelines using Xception features with simple classifiers such as Linear SVM, kNN, and Bagged Trees achieve state-of-the-art performance, with up to 99.5\% accuracy and macro-F1 scores, surpassing more complex or end-to-end deep learning approaches. The analysis highlights the operational benefits of this approach for robust, field-deployable debris identification and provides pathways for future integration with robotics and onsite automation systems.
研究动机与目标
- Motivate robust, field-deployable debris classification to improve on-site waste sorting and resource recovery.
- Develop a practical, hybrid vision pipeline combining deep feature extraction with traditional ML classifiers.
- Create a balanced, real-world dataset of C&D debris captured on UAE sites for benchmarking.
- Evaluate multiple classical classifiers on deep embeddings to identify efficient, high-performance models.
提出的方法
- Collect and curate a balanced onsite image dataset with four debris classes (Ceramic/Tile, Concrete, Trash/Waste, Wood).
- Preprocess images by grayscale conversion, resizing to 299x299, center-cropping and ImageNet-based normalization; apply augmentation only to training data.
- Extract deep features using a pre-trained Xception network (encoder only) and apply global average pooling to obtain 2048-dim feature vectors.
- Standardize features via z-score normalization using training data statistics.
- Train and evaluate multiple ML classifiers (Linear SVM, Bagged Trees, kNN, Logistic Regression ECOC, RBF SVM, LDA) on the Xception embeddings; select best-performing models.
实验结果
研究问题
- RQ1How well can a hybrid pipeline combining Xception-derived features with classical ML classifiers classify four C&D debris categories in real-world site conditions?
- RQ2Which ML classifier on deep embeddings yields the highest accuracy and macro-F1 on onsite debris images?
- RQ3What are the confusions and misclassification patterns among debris classes, and how robust are the models to field conditions?
主要发现
- The hybrid pipeline with Xception features and simple classifiers achieved 99.45% accuracy and 99.47% macro-F1 on the test set for the four-debris task.
- Linear SVM, Bagged Trees, and kNN achieved identical top performance (99.45% accuracy, 99.47% macro-F1).
- Logistic Regression (ECOC) reached 98.36% accuracy and 98.41% macro-F1, while RBF SVM and LDA scored 96.72% and 95.63% respectively.
- Confusion analysis showed near-perfect performance for Ceramic/Tile and Concrete; Wood and Trash-Waste had minor misclassifications, with some overlap where Wood was confused with Trash-Waste.
- The dataset is perfectly balanced (450 images per class) and collected from active UAE construction sites, supporting fair comparison and practical deployment insights.
- The results support deploying the hybrid pipeline for field robotics and onsite monitoring with potential extensions to robotics integration and multimodal sensing.
阅读本论文
在 Nubint 阅读器中打开原文,边读边就任意段落提问。
在 Nubint 中阅读本论文本解读由 AI 根据论文原文生成,引用前请核对原文。