池塘养殖在提供水产品同时为周边水环境带来了污染,为此对养殖池塘提取算法进行研究。针对养殖池塘提取过程中雷达图像受地形影响易产生噪声,而多光谱图像易受天气等因素影响这一现象,文中提出了一种基于决策级融合的塘坝提取方法,综合多数据源成像的优势来提高识别精度。先对雷达图像进行预处理和超分辨图像重建,再用面向对象的连通分量分割算法对图像进行水陆分割。随后,利用2015 水体指数对光学影像进行水体提取,整合光学数据的光谱特性并与雷达影像进行决策级融合,全面考虑每个数据源的分类结果,得到最终的养殖池塘分布图像。实验数据表明:决策级融合利用雷达影像较高的分辨率和光学影像丰富的波段信息,提高了对小型塘坝的分辨能力,较单一数据源提取优势显著。

An aquaculture pond extraction algorithm is proposed in this paper for the reason that pond aquaculture brings pollution to the surrounding water environment when providing aquatic products. In the process of pond extraction, radar images are susceptible to noise due to topography and multispectral images are susceptible to weather and other factors. To address this problem, a decision-level fusion-based pond extraction method is proposed to improve the recognition accuracy by integrating the advantages of multiple data sources images. Radar images are preprocessed and reconstructed by super-resolution first, and then the object-oriented connected component segmentation algorithm is used to segment the image by water and land. Then, the 2015 water index is used to extract water in optical images. The spectral characteristics of optical image is integrated to the radar image for decision level fusion. The final aquaculture distribution image is obtained after comprehensively considering the classification results of each data source. The experimental data show that the decision-level fusion takes advantage of the higher resolution of the radar image and plenty waveband information of the optical image to improve the discrimination of small ponds, which has significant advantages over the extraction from a single data source.

TN957.52