Constructing a visual detection model for floc settling velocity using machine learning

Abstract

Optimizing the dosage of coagulant is a time-consuming process, and real-time evaluation of floc settling velocity can quickly predict the coagulation effect and optimize the dosage. This study used a convolutional neural network (CNN) model to analyze the accuracy of floc image recognition of settling velocity. Python-OpenCV was employed to develop a program that segments individual flocs and detects their settling velocity to constructing a dataset of floc images and settling velocity. The results showed that the accuracy of determining the settling velocity of flocs solely based on their particle size was 88%, indicating that the floc structure is complex and a single parameter is not sufficient to accurately identify settling velocity. The results of the CNN analysis indicated that using a relatively simple Lenet5 structure can quickly achieve an accuracy of 88%, while using a Resnet18 structure can achieve recognition accuracy of over 90%. These findings suggest that machine learning techniques applied to floc images can effectively evaluate floc settling velocity, providing theoretical guidance for optimizing coagulant dosage and regulating coagulation processes.