Conventional water quality monitoring techniques require the collection of samples to be later analysed in a laboratory to determine their physical-chemical and biological attributes. This method has high precision but it can take several days to obtain the results, leaving no reaction time for companies where water is a key production input. Other faster methods involved the use of sensors, but these can only measure certain variables of water quality, often with less precision than manual collections. Our proposal consists of analysing water quality using high-resolution microscopic images to classify pollutants through a convolutional neural network model. Unlike conventional laboratory analysis and collection methods, we can give a much lower latency while maintaining the same efficiency and accuracy. The technology had been validated in a laboratory (TRL-4) using real samples of contaminated water with cyanobacteria and hydrocarbons provided by a Chilean water company and the Max Plank Institute. From more than 700,000 microscopic images we obtained a precision of over 93% with sensitivity over 95% for cyanobacteria classification.
Renzo Valencia, ML engineer student, Pontificia Universidad Catolica, Chile, www.uc.cl
|Is this abstract from experiment?||No|
|Name of experiment and experimental site||N/A|
|Is the speaker for that presentation defined?||Yes|