A Novel Tool for Biodiversity Studies: Earthworm Classification via NGS and Neural Networks

Earthworms are important in agriculture in the process of soil fertilization and influence its physicochemical properties. The taxonomic classification of earthworms using morphological characteristics requires experts, is difficult, and can require specimen dissection to extract detailed anatomical studies. Molecular techniques are time-consuming and expensive. The objective of this study was to distinguish earthworms belonging to different genera, Eisenia, Dendrobaena, and Lumbricus, using an innovative approach involving machine learning models built based on image texture parameters from individual color channels R, G, B, L, a, b, X, Y, Z, U, V, and S. The earthworms Eisenia fetida, Dendrobaena ssp., and Lumbricus terrestris were used as research materials. Image acquisition was performed using a flatbed scanner on a black background. In the case of each earthworm, 2172 texture parameters from images in individual color channels R, G, B, L, a, b, X, Y, Z, U, V, and S were extracted. Textures after selection were used to develop classification models using machine learning algorithms. The earthworms Eisenia fetida, Dendrobaena ssp., and Lumbricus terrestris were distinguished with the accuracy reaching 100% for models built using Logistic, Ensemble, and Narrow Neural Network. All earthworms were correctly classified. Also, in the case of other models, earthworm classes were distinguished with high accuracies, such as 99% (Naive Bayes, Random Forest, SVM, KNN), 97% (Simple Logistic), and 94% (KStar). For the most important species, E. fetida, the correctness of the species identification was confirmed by direct RNA sequencing. The application of image analysis and machine learning turned out to be a non-destructive, inexpensive, and objective approach to distinguishing earthworms belonging to different genera.