基于改进LeNet-5网络的污泥沉降比检测研究

Study on Sludge Sedimentation Ratio Detection Based on Improved LeNet-5 Network

在工业废水处理过程中,污泥沉降比检测对于污水处理效果起至关重要的作用,而传统的污泥沉降比检测以人工为主,工作量大且不可控,会对结果会造成不可预计的误差。提出一种基于改进的LeNet-5神经网络的污泥沉降比检测方法,训练之前先对目标进行颜色阈值判定,并在训练过程中提出一种轻量化特征重用网络模和正则分类器模式消除训练过程中Label-dropout的边缘效应,最后根据输出结果借助客户端/服务器(C/S结构)模式搭建整个系统。实验结果表明,改进后的神经网络对测试集的准确率高达96%以上,远高于传统神经网络和人工方法,而且改进后的神经网络更适用于小样本数据集的分类识别,极大提高了准确率和效率。

In the process of industrial wastewater treatment,the detection of sludge sedimentation ratio plays a vital role in the effect of wastewater treatment the traditional sludge settling ratio test is mainly manual,which is a large and uncontrollable workload and can cause unaccountable errors in the results.Therefore,propose a detection method of sludge sedimentation ratio based on improved LeNet-5 neural network.Before training,the color threshold of the target is determined.During training,a lightweight feature reuse network model and a regular classifier model are proposed to eliminate the edge effect of Label-dropout,Finally,the output results are used to build the whole system with the help of the client/server(C/S structure)mode.The experimental results show that the accuracy of the improved neural network to the test set is as high as 96%,which is far higher than the traditional neural network and artificial methods.Moreover,the improved neural network is more suitable for the classification and recognition of small sample data sets,greatly improving the accuracy and efficiency.