Aquaculture Detection Based On Feature Stability And Classification Accuracy For Landsat8 Images

In the sea area use management,the aquaculture region has been rapidly expanded,which have adversely influence on the ecological environment of the coastal zone.In order to protect the ecological environment of the coastal region and maintain the sustainable development of the local aquaculture industry,it is necessary to monitor the coastal aquaculture regions rapidly and effectively.Compared with on-site investigations,satellite remote sensing images can be used to monitor offshore aquaculture regions quickly.In remote sensing images,Landsat images are widely used due to their wide coverages.However,due to its low resolution and uneven edge distribution,the edge detection of the aquaculture region are discontinuous,resulting in the edge characteristics of the aquaculture region not obviously.The spectral features have the same spectrum with different surface features,demonstrating the difficulty to detect the aquaculture regions in Landsat images.The existing algorithms does not consider feature selection stability and classification accuracy at the same time,resulting in unstable image classification performance in different regions,so the performance can still be improved.To solve this problem,This paper designed a aquaculture region detection algorithm based on feature selection and classification.The parts of Liaoning,Jiangsu and Guangdong as the chosen regions with Landsat8 images.The main work of this paper is as follows:(1)A coastal zone multi-scale segmentation algorithm is presented.Because the regional characteristics of the aquaculture region have advantages over those of the pixels,in order to reduce the computational complexity,this paper gives a multi-scale segmentation algorithm for the coastal zone based on object-oriented idea.(2)A strategy based on feature stability and classification accuracy is proposed.Because the filter algorithms cannot provide features which are more suitable for aquaculture regions.To solve this problem,we propose a strategy based on both feature stability and classification accuracy.The filter algorithm used for aquaculture detection can be obtained through this strategy.At the same time,the type of classifier and the optimal feature subset dimension can be determined.The validity of the algorithm is verified by experimental comparison and analysis and this algorithm can improve the stability and classification accuracy to a certain extent.(3)The experimental verifications of the aquaculture regions detection are implemented.Compared the algorithm of this paper with the other two embedded feature selection algorithms,the detection algorithm in this paper has similar accuracy,but the running time is greatly reduced.The performance indicators include accuracy,recall,running time and so on.The algorithm in this paper is also better than the comparison algorithm for aquaculture regions.Experiments show that the detection accuracy of the aquaculture regions in this paper is over 90%and the difference of performance in different image areas is small.