Detection Based On Semi-supervised Clustering Of Red Tide Forecasts And Seagrass

In recent years, frequent marine disasters have posed a serious threat to the marine ecological balance, human healthy, offshore tourism and aquaculture. Nowadays, the marine department has collected Terabytes or even Petabytes of marine data and the amount of data still increases at Gigabytes rate per day. The traditional supervised methods usually require a large amount of human labeling data as training set, which cost too much. Different with supervised learning, semi-supervised learning only require a little of prior knowledge, In addition, it can use a large amount of non-labeled data to tune the model, which can reduce labeling costs. In this paper, we take enteromorpha and red tide as examples and utilize semi-supervised learning mechanism to enteromorpha detection and red tide forecast. On the other hand, Marine disaster is a phenomenon arising from the interaction of multiple factors. In this paper, giving red tide as an example, we construct a network of factors based on composited complex network theory and analysis the network topology. Thus, we provide a basis for red tide forecast.The main work of this paper can be summarized as follows.(1) We propose an effective semi-supervised clustering framework for enteromorpha detection. Firstly, with a few of labels, we get the instance level knowledge of pairwise constraints. Then, we adopt metric learning to construct an optimization problem. We solve the optimization problem to obtain attribute weights and the new metric, and utilize the new metric to partition the pixel set into two clusters. Finally, we identify the corresponding categories of clusters for enteromorpha detection based on labeles. Experimental results on real enteromorpha datasets demonstrate the effectiveness of the proposed approach.(2)We propose a red tide forecast method based on semi-supervised clustering. In this paper, with unbalanced dataset, we propose avariant method for semi-supervised clustering based on metric learning with pairwise constraints. The main idea is to establish harsher conditions for learning cannot-link constraints, so as to increase distances among clusters. The effectiveness of the proposed method is verified on some balanced and unbalanced UCI datasets. Furthermore, we extend our approach and apply it to red tide forecast, and get excellent performance on real dataset, which is extracted from red tide monitoring data during2003to2009.(3)We establish a red tide monitoring network model based on compound network. A network of factors is constructed based on complex network theory. We get the relationships between different factors according to analysis the network topology. We also construct a network of red tide sites based on spatial distance and water quality. Thus, we provide a basis for red tide forecast.