Assessment of intertidal seaweed biomass based on RGB imagery

Abstract

The Above Ground Biomass (AGB) of seaweeds is the most fundamental ecological parameter as the material and energy basis of intertidal ecosystems. Therefore, there is a need to develop an efficient survey method that has less impact on the environment. With the advent of technology and the availability of popular filming devices such as smartphones and cameras, intertidal seaweed wet biomass can be surveyed by remote sensing using popular RGB imaging sensors. In this paper, 143 in situ sites of seaweed in the intertidal zone of GouQi Island, ShengSi County, Zhejiang Province, were sampled and biomass inversions were performed. The hyperspectral data of seaweed at different growth stages were analyzed, and it was found that the variation range was small (visible light range < 0.1). Through Principal Component Analysis (PCA), Most of the variance is explained in the first principal component, and the load allocated to the three kinds of seaweed is more than 90%. Through Pearson correlation analysis, 24 parameters of spectral features, 9 parameters of texture features (27 in total for the three RGB bands) and parameters of combined spectral and texture features of the images were selected for screening, and regression prediction was performed using two methods: Random Forest (RF), and Gradient Boosted Decision Tree (GBDT), combined with Pearson correlation coefficients. Compared with the other two models, GBDT has better fitting accuracy in the inversion of seaweed biomass, and the highest R2 was obtained when the top 17, 17 and 11 parameters with strong correlation were selected for the regression prediction by Pearson’s correlation coefficient for Ulva australis, Sargassum thunbergii, and Sargassum fusiforme, and the R2 for Ulva australis was 0.784, RMSE 156.129, MAE 50.691 and MAPE 28.201, the R2 for Sargassum thunbergii was 0.854, RMSE 790.487, MAE 327.108 and MAPE 19.039, and the R2 for Sargassum fusiforme was 0.808, RMSE 445.067 and MAPE 28.822. MAE was 180.172 and MAPE was 28.822. The study combines in situ survey with machine learning methods, which has the advantages of being popular, efficient and environmentally friendly, and can provide technical support for intertidal seaweed surveys.

Seaweed Habitats on the Shore: Characterization through Hyperspectral UAV Imagery and Field Sampling

“…So, the use of common algorithms could be perfectible, but it does answer the problem of the study by classifying correctly macroalgal communities. Other algorithms such as random forests or support vector machines might be considered to estimate entire shores, as for coastal/terrestrial objects [93][94][95][96][97].…”

Seaweed Growth Monitoring with a Low-Cost Vision-Based System

“…Chen et al [12] propose a seaweed monitoring technique based on RGB imaging. They show that classification accuracies between 70% and 85% can be obtained for different types of seaweed.…”