Taxonomic Classification of Spiders (Araneae) Based on Image Texture Analysis Using Multifiltering

Abstract

This work presents an innovative approach to spider image classification using machine learning techniques. The main objective is to develop an automated model capable of identifying spider specimens from different anatomical perspectives, including profile, dorsal, ventral, and genital structures, contributing to taxonomic and ecological research. The dataset consisting of 2,419 images, is configured to capture the morphological diversity of different families, structures, and sexes. For analysis, texture descriptors such as sSIFT, sDenseSIFT, and sBRISK, along with their combinations, are employed. These are integrated with normalization methods (z-score, min-max, and robust), class balancing techniques (SMOTE), and varying dataset proportions (10%, 20%, 50%, and 80%). The primary model is an SVM, tested with linear, polynomial, and RBF kernels, targeting three categories: Sex, Structure, and Family. The best performance is achieved for the Sex target (F1 Score?=?0.70) using the sBRISK-sDenseSIFT combination, z-score normalization, RBF kernel, and 80% of the dataset. For Structure, the highest F1 Score was 0.76, obtained using combined descriptors and SMOTE with a reduced dataset (20%). The Family target, due to its high granularity (34 categories), posed the greatest challenge, with a maximum F1 Score of 0.39. The analysis highlighted that combined descriptors, robust normalizations, and class balancing are critical for complex targets. Compared to the literature, this study demonstrated the effectiveness of integrated pipelines for high-granularity classification tasks. Future work includes expanding the dataset and exploring deep neural networks to improve model generalization. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.