Differentiation of River Sediments Fractions in UAV Aerial Images by Convolution Neural Network

Riverbed material has multiple functions in river ecosystems, such as habitats, feeding grounds, spawning and shelters for aquatic organisms, particle size of riverbed reflects the tractive force channel flow. Therefore, regular surveys are conducted environmental protection flood control projects. The field method is most conventional survey. However, require much labor, time, cost to collect on site. Furthermore, its spatial representativeness also a problem because limited survey area against wide riverbank. As further solution these problems, this study, we tried an automatic classification conditions using aerial photography with unmanned vehicle (UAV) image recognition artificial intelligence (AI) improve efficiency. Due AI processing, large number images can be handled regardless whether they fine or coarse particles. We that have difference characteristics convolutional neural network (CNN). GoogLeNet, Alexnet, VGG-16 ResNet, common pre-trained networks, were retrained perform new task 70 transfer learning. Among networks tested, GoogleNet showed best performance study. overall accuracy reached 95.4%. On other hand, it was supposed shadows gravels caused error classification. taken uniform temporal period gives higher classifying same training data. results suggest potential evaluating materials UAV CNN.