Automatic recognition of construction waste based on unmanned aerial vehicle images and deep learning

As one of the main components of urban waste, the appropriate disposal of construction waste is crucial for the sustainable development of cities. The recognition and classification of construction waste are crucial components of construction waste management and lay the foundation for high-value applications. This paper utilized unmanned aerial vehicle (UAV) aerial images and the YOLO model to recognize construction waste on-site. A dataset of indoor scattered, indoor dense, and demolition site conditions was established. The impact patterns of epoch number, initial learning rate, and batch-size on the model were discussed, and the optimal parameter values were determined. The influence of training dataset composition was analyzed, indicating that appropriately adding indoor dense condition images helped improve the convergence speed and recognition accuracy. In the model performance validation, the comparison with manually sorting results revealed a high accuracy in recognizing the five types of construction waste. The recall of all five types was around 0.8. The validation results showed that the model can efficiently and accurately recognize and classify construction waste in demolition site images. The proposed method can aid in the rapid assessment and dynamic monitoring of construction waste, thereby enhancing the efficiency of waste management and recycling.