Coastal Sargassum Level Estimation from Smartphone Pictures

Abstract

Since 2011, significant and atypical arrival of two species of surface dwelling algae, Sargassum natans and Sargassum Fluitans, have been detected in the Mexican Caribbean. This massive accumulation of algae has had a great environmental and economic impact. Most works addressing this topic use high-resolution satellite imagery which is expensive or may be time delayed. We propose to estimate the amount of Sargassum based on ground-level smartphone photographs that, unlike previous approaches, is much less expensive and can be implemented to make predictions almost in real time. Another contribution of this work is the creation of a Sargassum images dataset with more than one thousand examples collected from public forums such as Facebook or Instagram, labeled into 5 categories of Sargassum level (none, low, mild, plenty, and excessive), a relevant difference with respect to previous works, which only detect the presence or not of Sargassum in a image. Several state-of-the-art convolutional networks: AlexNet, GoogleNet, VGG, and ResNet, were tested using this dataset. The VGG network trained under fine-tuning showed the best performance. The results of the carried out experiments show that convolutional neuronal networks are adequate for providing an estimate of the Sargassum level only from smartphone cameras images.

Keywords

• Sargassum
• classification
• deep learning
• environmental monitoring