Related Experiment Video
Updated: Sep 10, 2026
Superior Auto-Identification of Trypanosome Parasites by Using a Hybrid Deep-Learning Model
Published on: October 27, 2023
A convolutional neural network-based machine learning prototype for accurate identification of five Hyalomma tick
Essia Sebai1, Nour Elhouda Habessi2, Anis Ben Aicha3
1Laboratory of Parasitology, National School of Veterinary Medicine, Institution for Agricultural Research and Higher Education, University of Manouba, 2020, Sidi Thabet, Tunisia; Research Laboratory Epidemiology of Endemic Infections in Herbivores in Tunisia: Application to Control (LR16AGR01), National School of Veterinary Medicine, 2020, Sidi Thabet, Tunisia.
Abstract:
Accurate tick species identification is essential for vector surveillance and control of tick-borne diseases, but conventional morphological diagnosis remains challenging for closely related Hyalomma species. This study developed and evaluated a deep learning-based framework for automated recognition of five medically and veterinary important Hyalomma species from Tunisia. A dataset of 1344 dorsal and ventral images of morphologically identified ticks was generated from livestock specimens, including 484 H. marginatum, H. 418 dromedarii, 79 H. excavatum, 62 H. impeltatum, and 301 H. scupense. An integrated YOLOv11-CNN cascade pipeline was designed for tick detection, genus-level filtering against non-Hyalomma arthropods, dorsal/ventral orientation recognition, sex determination, and species-level classification. Performance was assessed using standard classification metrics, YOLO-specific detection metrics, confusion matrices, and LIME-based explainability analysis. The YOLOv11 model reliably detected tick regions, while the taxonomic filtering module discriminated Hyalomma ticks from other arthropod genera, including Argas, Ixodes, Rhipicephalus, and Melophagus. CNN classifiers achieved high performance for sex determination and species recognition, with species-level accuracy ranging from 97% to 100% under controlled laboratory imaging conditions. LIME analysis indicated that model predictions were associated with biologically relevant morphological regions used in classical tick taxonomy. This study provides a first YOLOv11-CNN cascade prototype for automated identification of closely related Hyalomma species and offers a promising decision-support tool for standardized tick identification, veterinary diagnostics, and future One Health surveillance after integreting future studies involving larger and geographically diverse collections will be necessary to further improve model robustness and transferability.
