Floating Raft Aquaculture Area Automatic Extraction Based on Fully Convolutional Network

Abstract

Cui, B.-G.; Zhong, Y.; Fei, D.; Zhang, Y.-H.; Liu, R.-J.; Chu, J.-L., and Zhao, J.H., 2019. Floating raft aquaculture area automatic extraction based on fully convolutional network. In: Jung, H.-S.; Lee, S.; Ryu, J.-H., and Cui, T. (eds.), Advances in Remote Sensing and Geoscience Information Systems of Coastal Environments. Journal of Coastal Research, Special Issue No. 90, pp. 86-94. Coconut Creek (Florida), ISSN 0749-0208.In the extraction of floating raft aquaculture areas from remote sensing images, the method of visual interpretation is time-consuming and laborious, and the traditional machine learning method has poor generalization ability and fitting ability for remote sensing data. To overcome these two problems, this paper proposes a method based on using a fully convolutional neural network to automatically extract floating raft aquaculture areas. The proposed method uses multiple convolution layers, pooling layers and nonlinear ReLU functions to extract nonlinear and invariant deep features of floating raft aquaculture areas, which effectively improves the accuracy of recognizing floating raft aquaculture areas. At the same time, L2 regularization and dropout strategies are added to the neural network model to avoid overfitting. The offshore area of Lianyungang in China was selected as the research region of the floating raft aquaculture. The experimental results show that the proposed method area in this paper effectively identifies and extracts the aquaculture areas.