Improving Road Surface Classification Accuracy and Real-time Waste Monitoring: A Solution and Deployment Approach

This study conducted testing and usage in actual environments after training, and discovered a significant problem. As this study classified the entire road surface situation, the condition of each road surface is highly variable and there are different greening situations and soil elements at each edge due to various factors such as greening. Therefore, this study may cause misjudgment of classification results. To address this issue, a solution is proposed, which involves pre-processing each image before classification by center-cropping and removing the edge parts with different greening situations, and focusing on detecting the central part. This method significantly improves the accuracy in actual environments and has great significance for the study's usage and further development, as shown in Figure 8.

The ultimate goal of this study is to achieve real-time monitoring of whether construction waste falls on the road. To achieve this, the study uses flask+nginx+uwsgi to become a network interface deployed on a cloud server, providing a channel for users to use. In actual applications, cameras and recorders provided by the waste truck operators and the government are installed at the front and rear of the vehicle to monitor the situation. By processing the recorded video into ten images per minute for monitoring and comparing the results from the front and rear cameras, it is possible to determine whether the spillage of construction waste is caused by the vehicle and take appropriate measures.