Remote sensing identification of coastal zone mariculture modes based on association-rules object-oriented method
-
-
Abstract
Abstract: Marine aquaculture has developed very rapidly in China and at present, China has become the largest producer of marine aquaculture in the world. While meeting the growing demand for seafood consumption, the mariculture industry also poses serious ecological and environmental problems to the coastal zone. Remote sensing recognition of mariculture modes in coastal zone is of great significance to real-time monitoring, rational planning and orderly development of mariculture, which can help to manage the coastal aquaculture mode, aquaculture structure and aquaculture capacity in coastal zone. At present, there are four main methods for remote sensing identification in aquaculture waters: 1) Extraction by visual interpretation; 2) Extraction by spectral features; 3) Analysis by spatial morphology and structure; 4) Extraction based on Object-oriented techniques. There will have mixing problems caused by "different objects with the same spectrum", "same objects with the different spectrum" and salt-and-pepper noise in image processing, if aquaculture information is extracted by spectral information or texture information alone. In order to reduce the interference of human factors of object-oriented classification rules and improve the efficiency and automation of classification rules generation, in this paper, we combined the association rules method and Object-oriented method to build a mariculture modes classifier through automatic and intelligent acquisition for different modes classification. Taking Zhelin Bay in the east of Guangdong province as an example, the GF-1 image as data source, using the spectral, geometric and texture features and their correlations of the objects of different mariculture modes as transaction data, mariculture modes strong rules were mined by Apriori algorithm. Four kinds of mariculture modes information (pond culture, cage culture, beach aquaculture, floating raft) in bay aquaculture core area were extracted. The results showed that pond culture area in Zhelin Bay was 2 228.47 hm2, cage culture area was 111.95 hm2, beach aquaculture t area was 12.95 hm2, floating raft area was 48.34 hm2. Cages in Zhelin Bay were distributed in two regions, one was in the sea area between Suizhou Island and Xuxian Island and Xi'ao Island, the other was located between the northeast corner of Haishan Island and Xizhou Island. Ponds were mainly located in the northern part of the study area in Huanggang town. Beach aquaculture was in the innermost part of Zhelin Bay, near the south side of the pond. Floating rafts were distributed around the cages and tended to be at the side of the bay center. In order to compare the association rules mining with the object-oriented combination method and the traditional methods, K-neighboring object-oriented method and association rule mining method were used respectively for the classification and extraction of marine aquaculture modes. These two classifications were implemented in Envi5.5 and Weka3.7.12 software, respectively. The specific steps were as follows, 300 sample points in Google Earth high-resolution images and field survey samples were chosen, of them, 50% were used as training samples, and the other 50% for test samples. Then, the K-adjacent object-oriented method and association rule mining method were selected respectively for classification. Finally, the classification accuracy of the three methods was compared and evaluated. The classification accuracy of mariculture model extraction based on association-rules object-oriented method was 88.65%, which was 14.38 percent point higher than that of K-adjacent object-oriented classification, and 12.16 percent point higher than that of association rules mining classification. association-rules object-oriented method broadened the access to information in the traditional logical reasoning classification method, made the classification more intelligent, and enhanced the classification speed and algorithm reliability. This method can improve the classification accuracy remarkably, which can be an effective support method for the complex mariculture modes recognition.
-
-