Monitoring roads and canals utilization condition for land consolidation project based on UAV remote sensing image

Abstract: The infrastructure in land consolidation projects provide important guarantee for harvest and natural calamities resistance to farmers directly, it is fundamental for rural social and economic development. However, some facilities in land consolidation projects cause problems such as fractured pavement or canal silted up after construction, which can bring negative effects to agricultural production. So it is important to find an effective and reliable technical method to monitor and evaluate the effects of land consolidation projects. Unmanned aerial vehicle (UAV) remote sensing is widely used in feature recognition, roads and canals collection and crop productivity evaluation during recent years, but it is rarely used to evaluate the quality of the infrastructure of land consolidation. To objectively monitor and effectively evaluate the post-construction utilization of the infrastructure in land consolidation projects, this paper selected typical land consolidation projects, used the multi-rotor UAV for aerial photography test to obtain high-resolution aerial images, and put forward the complete technical method and operational procedures for monitoring and evaluation of land consolidation infrastructure. Route 1 covered the whole study area. Route 2a mainly took pictures on main field roads and canals for precision shooting, and route 2b focused on roads and canals due to their width were less than 2 m. After image processing, this paper gained image grids of field roads and canals which were wider than 2 m, then selected BoW (bag of words) model to build a sample feature database of surface features, including the pavement diseases and canal silted up such as fractured pavement, obstructed pavement, potholed pavement, canal mild silted up and canal severe silted up. The BoW model included speeded-up robust features (SURF) algorithm for image characteristic representation, and image visual dictionary for local feature clustering. Finally this paper used SVM (support vector machine) to classify the images. The results showed that: UAV remote sensing could monitor and locate the condition of infrastructure post-construction utilization under sunny and cloudless days. Using the method introduced in this paper and combined with the visual interpretation and field survey, the total accuracy rate of field roads reached 80%, and the total classification accuracy rate of canals was about 70%. The cross accuracy rate of field roads and canals was about 70%. The main problem of infrastructure post-construction utilization in the study area was the road obstruction and mild silted up of the canals caused by delayed management and maintenance. After monitoring, this paper analyzed the causes of the differences of monitoring ratio between field roads and canals, and especially explained the causes of the lower monitoring ratio of canals in details. They were as follows: first, the training samples may not match the actual objects in the maps, which caused the extracted information of blocked canals incomplete; second, the spectral information of vegetation and canal water shared the same characteristic in the visible-band image, which might interference the model. This paper also used higher resolution image and linear infrastructure under 2 m to validate the reliability of the model. Route 2a was used to validate the classification accuracy due to the image resolution was higher. Route 2b was used to validate the classification accuracy due to the linear infrastructure width was under 2 m. We found that the overall accuracy of linear infrastructure increased insignificantly while the image resolution higher, meanwhile the overall accuracy of linear infrastructure decreased remarkably when the road and canal width was less than 2 m. During the process of UAV remote sensing for monitoring linear infrastructure post-construction utilization such as field roads and canals of land consolidation projects, we can use the high-resolution image efficiently in sunny and cloudless condition, and at the same time there is still much room for improvement.