A River Segmentation for Flood Monitoring with Atrous Convolution via DeepLabv3
DOI:
https://doi.org/10.37934/araset.56.2.6273Keywords:
Image segmentation, River water segmentation, Fully convolutional networks, DeepLabv3, Flood managementAbstract
Flood has been identified as a common natural disaster for years. This is the evidence of the effect cause by heavy rainfall which then lead to damages of infrastructure and deaths. The presence of this natural disasters can cause a lot of problems and risk especially to human being. The prevention of flood is almost impossible as it is a natural phenomenon. In this work, we proposed a water segmentation technique to analyses the images of river in term of water at the area from the camera which will automatically detect anomalies such as sudden water increase. The Deep Learning segmentation algorithm DeepLabv3 and DeepLabv3+ are trained and tested for the task of water segmentation and the performances are compared with previous works. In our finding, the accuracy obtained by our proposed method DeepLabv3 is 97.07% thus achieved the state of art in performing the task of water segmentation. Thus, DeepLabv3 model is suit and practical in the solving the flood issue.
