基于数据挖掘的GF-1 遥感影像绿潮自适应阈值分区智能检测方法研究

由于受到云雾的影响，可见光影像能够高效用于绿潮检测的数据源较为有限，特别是云覆盖较为严重的可见光影像，基本无法用于检测绿潮。即使影像数据是在薄云、薄雾、无云覆盖的情况下获取的，由于其光谱反射值存在较大差异，依然很难采用同一阈值进行绿潮检测。基于此，为了提高可见光影像的利用率，实现不同云覆盖情况下，绿潮的高精度自适应阈值的自动检测，本文以GF-1影像为数据源，首先采用K-means聚类和C4.5决策树方法实现影像云覆盖情况的自动识别；其次，选取大量不同云覆盖情况下子图像样本（每个子图像样本中均包含绿潮和海水两类），分析得出不同云覆盖情况下绿潮和海水的区分阈值y与影像光谱差x=band_nir-band_red之间所具有的线性关系；然后，利用分析得出的线性关系提出一种适用于GF-1影像的绿潮分区自适应阈值自动检测方法。最后，为验证提出方法的有效性，分别采用NDVI方法、EVI方法和本文提出的自适应阈值自动检测方法进行绿潮提取实验。实验结果表明，对于GF-1卫星遥感数据，本文提出的绿潮自适应阈值分区自动检测方法明显优于传统的NDVI和EVI检测方法，不仅提高了绿潮的监测精度，而且实现了绿潮提取的全自动化。

An intelligent divisional green tide detection of adaptive threshold for GF-1 image based on data mining

Due to the influence of clouds, the visible light images that can be used effectively for green tide detection are limited, especially when the cloud coverage is serious, which can not be used to detect green tide. Even if the image data is acquired under thin cloud, mist, and cloudless coverage, it is still difficult to use the same threshold for green tide detection because of the large difference in spectral reflectance values. Based on this, in order to improve the utilization of visible light image and realize green tide high-precision automatic detection of the adaptive threshold under different cloud coverage conditions, GF-1 images are selected as data source, firstly, K-means clustering and C4.5 decision tree methods are used to automatically identify cloud coverage type; secondly, a large number of sub-image samples with different cloud coverage are selected (each sub-image sample contains two types of green tide and sea water), and the linear relationship between the classification threshold y and the image spectral difference x (x = band_nir-band_red) is analyzed under different cloud coverage, here, the classification threshold y is the value that can distinguish green tide and sea; then, green tide partition adaptive threshold automatic detection method for GF-1 image is proposed by using the linear relationship analyzed. Finally, in order to verify the effectiveness of the proposed method, NDVI、EVI methods and the adaptive threshold automatic detection method proposed in this paper are used to carry out the green tide extraction experiment. The experimental results show that for the GF-1 satellite remote sensing data, the green tide adaptive threshold partition automatic detection method is better than traditional NDVI and EVI methods, which not only improves the monitoring accuracy of green tide, but also realizes the full automation of green tide extraction.