基于改进YOLO11n-CLS的矾花图像识别分类  

Alum Flower Image Recognition and Classification Based on Improved YOLO11n-CLS

作　　者：梅豪 MEI Hao(Nanjing Institute of Technology,Nanjing Jiangsu 211167)

机构地区：[1]南京工程学院,江苏南京211167

出　　处：《软件》2025年第3期101-104,共4页Software

摘　　要：针对火力发电厂等水处理过程中因混凝剂投加量不当导致水质不达标和成本超支的问题,本文提出了一种基于改进YOLO11n-CLS的矾花图像识别分类方法。该方法引入了EIEStem模块,通过Sobel滤波器提取图像边缘特征,并结合空间特征信息,提升了特征提取能力;采用ADown下采样模块,保留多层次图像特征,减少了传统卷积下采样时的特征丢失,提升了模型训练效率与准确性;为实现模型轻量化与高效化,提出了C3k2-Faster模块,通过FasterNetBlock中的部分卷积(PConv)减少计算冗余。实验结果表明,改进后的YOLO11n-CLS网络在水下矾花图像分类任务中的准确率(Acc@1)提高了6.2%,为火力发电厂原水处理中的混凝剂投放提供了精确的指导。In response to the problem of substandard water quality and cost overruns caused by improper dosage of coagulants in water treatment processes such as thermal power plants,this paper proposes an improved YOLO11n CLS based alum image recognition and classification method.This method introduces the EIEStem module,extracts image edge features through Sobel filters,and combines spatial feature information to enhance feature extraction capability;Adopting ADown downsampling module to preserve multi-level image features,reduce feature loss during traditional convolution downsampling,and improve model training efficiency and accuracy;To achieve model lightweighting and efficiency,the C3k2 Master module is proposed,which reduces computational redundancy through partial convolution(PConv)in Faster Netblock.The experimental results show that the improved YOLO11n CLS network has high accuracy in underwater alum image classification tasks(Acc@1)Improved by 6.2%,providing precise guidance for the use of coagulants in raw water treatment for thermal power plants.

关 键 词：水处理 矾花 深度学习 图像分类 YOLO11n 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]