基于卷积神经网络识别三维荧光光谱的水污染溯源研究

为快速、准确地追溯水中污染物来源,及时阻断污染,为生态环境管理部门提供科学的技术支撑,更好履行"三个说得清"中的"说得清污染来源",以昆明市安宁市8家重点企业为研究对象,采集企业内各工段废水进行三维荧光检测获得三维荧光谱图,采用目视剔除散射区域-线性归一化方法对原始荧光谱图进行预处理,基于ConvNet卷积神经网络构建水污染溯源模型并进行溯源。结果表明,目视剔除散射区域-线性归一化方法可有效剔除荧光谱图中瑞利散射区域,同时增强因水样稀释而衰弱的荧光特征;构建的溯源模型可成功识别实际废水的三维荧光谱图,识别正确率高达75%,是一种有效溯源水中污染物来源的方法。

Research on Water Pollution Traceability Based on Convolutional Neural Network Identification of Three-Dimensional Fluorescence Spectrum

In order to promptly intercept contamination through rapid and accurate tracing of the source of contaminants in water,thus providing scientific and technical support for ecological and environmental management departments,and better fulfill the "clarity of the pollution source" in the "Three Clarity",the paper took eight key enterprises in Anning City,Kunming as the research objects,and collected the wastewater from each section of the enterprises for 3D florescence detection to obtain the corresponding spectrum.Pre-processing was carried out by a visually rejecting the scattered area linear normalization method,after which the pollutants were traced by the traceability model constructed by ConvNet convolutional neural network.The results showed that visual rejection of scattered regions linear normalization method could effectively remove Rayleigh scattered regions from the fluorescence spectrum,and enhance fluorescence features weakened by dilution of the water sample.In addition,the successful identification of the 3D fluorescence spectrum of actual wastewater by the constructed traceability mode was achieved with a correct identification rate of 75%.This study provides an effective method for tracing the source of contaminants in water.