基于图像分析的堆肥腐熟度判别研究  

Research on Maturity Discrimination of Compost Based on Image Analysis

作　　者：丁雯丽 裴晓芳[1,2,3] 司广字 DING Wenli;PEI Xiaofang;SI Guangzi(School of Electronic&Information Engineering,Nanjing University of Information Science&Technology,Nanjing 210044;School of Electronic Information Engineering,Wuxi University,Wuxi 214105;Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology,Nanjing University of Information Science and Technology,Nanjing 210044)

机构地区：[1]南京信息工程大学电子与信息工程学院,南京210044 [2]无锡学院电子信息工程学院,无锡214105 [3]南京信息工程大学江苏省大气环境与装备技术协同创新中心,南京210044

出　　处：《计算机与数字工程》2023年第2期462-467,共6页Computer & Digital Engineering

基　　金：教育部产学合作协同育人项目(编号:202102563011);苏高教会“高质量公共课教学改革研究”专项课题(编号:2022JDKT138);高校哲学社会科学研究一般项目(编号:2022SJYB0979);南京信息工程大学无锡校区研究生创新实践项目(编号:WXCX202010);无锡学院教改课题(编号:JGZD202107)资助。

摘　　要：针对传统堆肥腐熟采用物理学、化学、生物学指标进行评价判定,步骤复杂、测定时间较长且工作量较大,缺乏可操作性的问题,该研究提出基于图像分析的堆肥腐熟分类识别模型,利用MBConv和Fused-MBConv的组合方式对不同原料堆肥图像的颜色、纹理、轮廓等特征进行提取,提高模型训练速度,减小参数量。在CoAtNet(Convolution+Attention)模型的基础上引入ECA-NET(Efficient Channel Attention Networks)注意力机制,进一步提高模型识别准确率。试验表明,在畜禽粪便、尾菜、秸秆以及三者混合的数据集上,改进后的Fused-CoAtNet模型腐熟度识别准确率平均达到100%、99.22%、99.74%、99.47%,与RestNet50、EfficientNetV2和CoAtNet模型相比,Fused-CoAtNet模型平均准确率分别提高0.31、0.58、0.17个百分点,对堆肥图像腐熟度的识别具有较好的判别效果,可为工厂化堆肥腐熟识别提供指导。For traditional compost ripening,physical,chemical and biological indicators are used to evaluate and judge.The steps are complex,the measurement time is long,the workload is large,and the operability is not enough.In this study,a compost maturity classification recognition model based on image analysis is proposed.The combination of MBConv and Fused MBConv is used to extract the color,texture,contour and other features of compost images of different raw materials,so as to improve the training speed of the model and reduce the number of parameters,increase the speed of model training and decrease the number of parameters.Based on the CoAtNet model,ECA-NET(Efficient Channel Attention Networks)attention mechanism is introduced to further improve the model recognition accuracy.The results show that on the dataset of livestock and poultry excrement,cauliflower,straw and the mixture of them,the average recognition accuracy of the improved Fused-CoAtNet model reaches 100%,99.22%,99.74%,99.47%.Compared with RestNet50,EfficientNetV2 and CoAtNet models,the average accuracy of Fused-CoAtNet model is increased by 0.31,0.58,0.17 percentage points,respectively.The improved Fused-CoAtNet model has a better discriminating effect on the maturity of compost image.It can provide guidance for identifying the ripeness of factory compost.

关 键 词：堆肥腐熟 图像分析 CoAtNet Fused-MBConv ECA-NET 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] S141.4[自动化与计算机技术—控制科学与工程]