Bioindication of Surface Water Supported by Automatic Image Analysis Using Deep Learning Neural Network – Cyclotella Case Study

Abstract

Bioindicative methods involving the identification and counting of indicator organisms (e.g. algae) are widely used methods in the assessment of surface water quality. For this reason, the purpose of this paper was automatic image analysis using the YOLO v8 deep learning neural network, directed at the detection of freshwater algae Cyclotella. Changes in the number of these organisms can indicate changes in the water quality and the trophic status of the reservoir, which makes automating their detection an important task. Traditionally, the detection and counting of objects in microscope images was done manually, but by using machine learning and especially neural networks, the process can be automated. YOLO (You Only Look Once) is an example of a network that, after proper training and validation, is capable of performing image detection in real time. In this study, the Roboflow object tagging tool was used to create a dataset divided into training, validation and test sets. Training of the network, validation of the model and evaluation of its metrics were carried out. The paper presents the obtained metrics of the YOLO v8 network on the validation set, such as Accuracy = 0.960, Precision = 0.964, Recall = 0.995. The presented results confirm the effectiveness of the applied method in automatic analysis of microscopic images containing algae and thus the high application potential of the method in supporting bioindication studies of surface water quality.

Keywords

• surface water
• computer vision
• bioindication
• freshwater algae
• deep learning neural network
• automatic image analysis
• microscopic samples
• object annotation