强光照下内河溢油纹理特征提取研究

发布时间：2018-10-17 12:51

【摘要】：近年来,在航运中泄漏到海洋与内河河流中的数万吨石油对周边环境造成了极其严重的污染。在海上溢油监测技术领域,国内外已取得了瞩目的成绩。然而,内河流域因其水文环境复杂,现有溢油监测技术仍无法应对突发的溢油事故。本论文以溢油纹理特征为出发点,根据"强光照下,油膜和水面呈现不同视觉效果"这一特性,通过纹理特征提取方法,分别对油膜和水面纹理进行特征提取,得到特征量。以纹理特征量为数据源,利用支持向量机分类原理,预测油膜和水面纹理图像分类准确率。对比不同提取方法下得到的预测准确率。预测准确率越高,说明纹理特征量包含的油膜和水面纹理特征越精确,越有助于后期溢油图像的监测识别工作。基于油膜和水面纹理特征,本论文对特征提取中经典的灰度共生矩阵方法进行了详细说明。Haralick从该矩阵中提取了 14个特征量,本文从中选择了能够表征油膜和水面纹理特征的特征量:角二阶矩、对比度、相关性、熵以及逆差矩,然后在灰度共生矩阵的基础上,衍生出一维灰度共生矩阵方法。根据油膜纹理的彩色特性,将灰度共生矩阵和颜色信息相结合,获得颜色共生矩阵,并由此衍生出一维颜色共生矩阵、各分量颜色共生矩阵等方法。根据HSI空间下溢油纹理特征,本文提出一种基于色调和饱和度分量的提取方法——色调饱和度共生矩阵法。利用上述方法提取油膜和水面纹理特征量,最终获得预测准确率。对比准确率,分析各方法优劣性。实验结果表明,针对强光照下的溢油纹理特征,从颜色共生矩阵和色调饱和度共生矩阵方法中提取的纹理特征量具有更高的分类准确度。颜色信息、像素空间关系信息以及色调饱和度分量信息在表征油膜和水面纹理特征方面具有重要的研究参考价值,可用于后续强光照下内河溢油的监测、识别工作。
[Abstract]:In recent years, tens of thousands of tons of oil leaking into the ocean and inland rivers in shipping have caused extremely serious pollution to the surrounding environment. In the field of offshore oil spill monitoring technology, domestic and foreign has made remarkable achievements. However, due to its complex hydrological environment, the existing oil spill monitoring technology is still unable to cope with sudden oil spill accidents. In this paper, based on the feature of oil spill texture, according to the feature of "oil film and water surface show different visual effects under strong light", the oil film and water surface texture are extracted by the method of texture feature extraction, and the feature quantity is obtained. The classification accuracy of oil film and water surface texture images is predicted by using support vector machine (SVM). The prediction accuracy of different extraction methods is compared. The higher the prediction accuracy is, the more accurate the oil film and surface texture feature included in the texture feature quantity is, and the more helpful to monitor and identify the oil spill image. Based on oil film and surface texture features, the classical gray level co-occurrence matrix method in feature extraction is described in this paper. Haralick extracts 14 feature quantities from the matrix. In this paper, we select the characteristics of oil film and water surface texture: angular second order moment, contrast, correlation, entropy and deficit moment, and then derive one dimensional gray level co-occurrence matrix method based on gray level co-occurrence matrix. According to the color characteristics of oil film texture, the color-co-occurrence matrix is obtained by combining the gray level co-occurrence matrix with the color information, and the one-dimensional color co-occurrence matrix and each component color-co-occurrence matrix are derived. Based on the texture features of oil spill in HSI space, a new extraction method based on hue and saturation components is presented in this paper, which is called hue saturation co-occurrence matrix method. The oil film and surface texture features are extracted by the above methods, and the prediction accuracy is obtained. Compare the accuracy and analyze the advantages and disadvantages of each method. The experimental results show that the texture features extracted from the color co-occurrence matrix and the hue saturation co-occurrence matrix method have higher classification accuracy for the oil spill texture features under strong illumination. Color information, pixel spatial information and hue saturation information have important reference value in the characterization of oil film and water surface texture features, and can be used to monitor and identify the oil spill of inland rivers under the following strong illumination.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】