基于特征谱图解析的鱼类行为模式识别

It is recognized that water resources are of vital importance to ecology as well as the growing prosperity. On-line bio-monitoring and warning systems based on biotechnology have been widely used for evaluating the water quality. However, the application of these methods was restricted because of the low sensitivity and specificity. In order to monitoring the water pollution more effectively, we design a new pattern recognition system based on the feature spectrum of fish behavior. A novel bio-sensor monitor system will be established, and the feature spectrum can be collected from eight channels containing the target fishes. Meanwhile, EMD method will be used to analyze the feature spectrum, and the information of fish behavior from eight channels will be extracted by multisensor information fusion technology. The SVM based on prior information will be used to establish the model of fish behavior. For validating the model, sufficient training and test samples are generated by virtual sample technique. The study applies the technology of spectrum analysis and pattern recognition to biological early warning, and will dramatically enhance the sensitive and specificity of biological warning technique.

水资源是人类赖以生存与发展的宝贵资源。利用生物预警技术对水质进行在线监测是目前研究的热点，然而生物预警技术的灵敏度和特异性受限，成为制约该技术推广的关键因素。本项目在特征谱图解析的基础上，对鱼类行为进行模式识别，具有较强的科学意义和现实意义。本项目拟研制新型生物传感器，并搭建实验装置，采集8个通道鱼类运动信号的特征谱图；然后，基于EMD算法对特征谱图进行解析，并运用信息融合技术对8通道谱图数据进行特征提取及综合分析；最后，研究基于先验信息SVM的鱼类行为模型建立方法，利用虚拟样本技术生成充足的训练样本和验证样本，对所建模型进行验证。该研究将特征谱图解析和模式识别应用于生物预警，丰富和完善了生物预警技术，显著提高了生物预警技术的灵敏度和特异性。

水资源是人类赖以生存与发展的宝贵资源。利用生物预警技术对水质进行在线监测是目前研究的热点，然而生物预警技术的灵敏度和特异性受限，成为制约该技术推广的关键因素。本项目在特征谱图解析的基础上，对鱼类行为进行模式识别，具有较强的科学意义和现实意义。本项目研制了新型生物传感器，并搭建实验装置，采集8个通道鱼类运动信号的特征谱图；然后，基于自适应分解算法对特征谱图进行解析，并运用信息融合技术对8 通道谱图数据进行特征提取及综合分析；最后，研究基于先验信息SVM 的鱼类行为模型建立方法，利用虚拟样本技术生成充足的训练样本和验证样本，对所建模型进行验证。研究结果表明，对特征谱图应用自适应分解算法经过4次迭代后，所获取的趋势项跟逐级行为响应模型模型吻合；利用基于分段处理的技术，能较好地去除生物钟现象，误报率降低了92%；融合八通道特征谱信息进行综合预警，比利用单一通道进行预警更鲁棒，误报率降低了40%，漏报率降低了35%。该研究将特征谱图解析和模式识别应用于生物预警，丰富和完善了生物预警技术，显著提高了生物预警技术的灵敏度和特异性。