基于深度学习的活性污泥丝状菌指数识别模型  

Auto-identification Model of the Filamentous Index of Activated Sludges Based on Deep Learning

作　　者：赵立杰 陈茜 王国刚 黄明忠 ZHAO Li-jie;CHEN Xi;WANG Guo-gang;HUANG Ming-zhong(School of Information,Shenyang University of Chemical Technology)

机构地区：[1]沈阳化工大学信息工程学院

出　　处：《化工自动化及仪表》2023年第6期800-806,共7页Control and Instruments in Chemical Industry

基　　金：国家重点研发计划“网络协同制造与智能工厂”重点专项(批准号:2018YFB1700200)资助的课题;2020年辽宁省高等学校创新人才支持计划项目;2021年度高等学校基本科研项目(重点项目)(批准号:LJKZ0442)资助的课题。

摘　　要：传统丝状菌数量的计量和长度的测量依赖人工显微镜下目视方式,非常耗时且困难,为此提出一种基于活性污泥相差显微图像快速量化活性污泥丝状菌与菌胶团细菌比例等级方法,实现丝状菌指数的自动识别。该方法在ResNet18模型框架内添加空间注意力机制增强丝状菌特征提取能力,采用多分类Focal Loss损失函数缓解丝状菌与菌胶团细菌样本类别失衡问题。实际污水处理厂活性污泥相差显微图像数据的仿真实验结果,验证了所提丝状菌指数识别模型的有效性。Considering the fact that traditional measurement of both number and length of filamentous bacteria relies on the visual inspection under a manual microscope,which is time-consuming and difficult.In this paper,a rapid quantification method based on phase-contrast microscopic images of activated sludge was proposed to realize auto-identification of the filamentous index.In this method,a spatial attention mechanism was added to ResNet18 model framework to enhance feature extraction ability of filamentous bacteria,and a multi-class Focal Loss function was used to alleviate the class imbalance between filamentous bacteria and micellar bacteria samples.Simulating phase-contrast microscopic image data of the activated sludge from a real sewage treatment plant verifies effectiveness of the proposed filamentous bacteria index recognition model.

关 键 词：活性污泥 丝状菌指数 ResNet FOCAL LOSS 空间注意力机制 不均衡数据 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]