基于微生物图像识别的水质监测方法研究

With the rapid development of economy, the pollution of environment especially water is increasingly serious. It's vital to carry out some effective monitoring over water to get their quality information and then adopt reasonable measures for pollution control. Among them, the detection of some aquatic microorganisms which are highly sensitive to water pollution is served as an important tool for water-quality monitoring. Comparing with the traditional physical or chemical analysis, this novel method for microbial identification and quantitative analysis, which is based on image recognition with convenient process and high detection efficiency, exhibits great application value and prospect for water monitoring. Aiming to develop such an analysis technique of aquatic microorganisms, in this project some algorithms of image segmentation, feature extraction and matching will be specially designed and studied according to the feature of the micrographs of microorganisms, such as dim border, overlapping of microorganisms, and complex background, followed by pattern classification to distinguish the marker microorganisms and perform quantity statistic, based on which the water quality can be evaluated. By working on this research project, we want to propose a fast and automatic method for water-quality monitoring of nature or treated water which can provide strong support for protection of water resources. The research finding is expected to further combine with the real time micrograph taking system, which can be used for real time water monitoring and lays a foundation for water governance and further building water-quality distribution map.

随着经济的发展，我国水体污染问题日益严重。实施水质监测对掌握水资源污染程度及选取合适的治污方法至关重要，其中水体指示性微生物检测是水质监测的重要手段。在微生物检测方法中，利用图像识别技术对微生物进行识别分类和数量统计的方法，具有快捷、高效、适合大样本水体监测等优点，有很大的应用价值和发展前景。本项目立足于水资源监测的重大需求，开发基于图像识别的微生物分析技术，根据水体中微生物图像边缘模糊、微生物重叠及背景复杂等特点针对性地设计图像分割算法及有效特征的提取及匹配方法，结合模式分类手段完成水体中常见指示性微生物的识别分类和数量统计并实现对水质的评估，提出面向自然水体状况和污水处理后水体质量的快速自动监测技术，为水资源保护及治理提供依据。本项目的研究成果可进一步与实时显微图像采集系统结合，实现对水体质量的实时监测与评估，为后续水体治理及制作一定范围内的水质分布图奠定基础。

本项目立足于水资源监测的重大需求，开发了基于图像分析的微生物识别分类技术。根据水体中微生物图像的特点，分别提出了3类不同的图像分割算法，即：针对微生物图像背景杂乱的情况，提出了一种基于图像分块的局部阈值二值化分割方法；针对微生物图像边缘较弱且容易出现断裂的情况，设计了一种基于核空间和加权邻域约束的直觉模糊C均值聚类分割算法；提出了一种优化的脉冲耦合神经网络分割算法，并通过大量实验验证了以上不同分割方法的分割效果及准确率。对于分割得到的微生物感兴趣区域，开展了基于自适应阈值及加权局部二值纹理模式的特征提取研究。随后基于广义典型相关分析法将提取得到的微生物形状特征及纹理特征进行融合，并生成独特的特征向量。最后基于支持向量机分类器，研究并实现了对水体微生物图像的分类识别。上述研究成果为进一步分析水体中微生物的种类和数量，监测水质状况和评估污水处理效果等提供技术支持。