Automated red tide algae recognition by the color microscopic image

Red tide occurs frequently these years and have become a great threat to marine ecology and human health. Monitoring the abundance of red tide algae is very crucial for forecasting and responding potential red tide outbreak. Now there are lots of imaging techniques can rapidly collect algae images which can be used to estimate the algae concentration by classification and counting, but few technologies are specific to red tide algae. In this study, we construct a high-solution color microscopic image dataset contain nine common species of red tide algae. Based on the dataset, we develop a computer vision- based automated red tide recognition and classification system involving image segmentation, artificial feature extraction and classification based on machine learning algorithm. Image segmentation detect the single algae’s boundaries and acquire its bounding rectangular areas as the subimage from the original images, even where several objects stick together. Feature extraction process is applied to extract specific feature vectors in terms of own artificial features including shape, color and texture features. Considering the uncertainty of the rotation of the red tide algae and the possible influence of environmental light, the features both have rotation and brightness invariance. we use three different algorithms including Logistic Regression (LR), Support Vector Machine (SVM) and Extreme Gradient Boosting (XGBoost) to construct classifiers to classify algae images based on extracted features. We also adopt the idea of ensemble learning to achieve better performance than a single algorithm.¬ The system achieves over 95% segmentation efficiency in the and 96% classification accuracy in about 200 test images, making it comparable with a trained biologist can achieve by manual method. The study proves the potential of identifying and classifying red tide algae by color microscopic images, which may provide new ideas for monitoring red tide by imaging techniques.