YOLOv7-Swin: Urban road trash detection based on improved YOLOv7
Huipeng Li, Tao He, Shuqin Wang, Shihang Luo, Chao Xu · 2023
In recent years, with the improvement of human awareness of environmental protection, various policies and measures for garbage classification have been put into practice. The management of road waste is crucial for maintaining the overall appearance of a city. While trash detection algorithms have been primarily focused on beach, domestic, and rural waste in the past, urban roads have received less attention due to their complex and diverse nature. Consequently, the existing algorithms for trash detection on urban roads are limited and have a low detection rate. To address this issue, this study proposes a deep learning-based urban road trash detection model called YOLOv7-Swin. This model utilizes the YOLOv7 model as the baseline and incorporates the Swin Transformer to improve feature extraction through the Attention mechanism. Additionally, a dataset named Road-Trash has been created, which consists of urban road trash samples from various road environments. This dataset serves as a target detection benchmark for urban road trash. Experimental results demonstrate that the proposed model exhibits an average accuracy mAP0.5 increase of 3% compared to the baseline model. These findings highlight the effectiveness and potential of the YOLOv7-Swin model in identifying and classifying different types of road trash on urban roads.