OrchID: Explainable deep learning for Ophrys orchid biodiversity monitoring

Abstract

Ophrys orchids are notoriously difficult to classify due to their morphological similarities and high intraspecific variability, creating challenges for botanical identification and biodiversity monitoring. This study develops an explainable deep learning framework for automated Ophrys species identification that combines high classification accuracy with interpretable model decisions aligned to botanical expertise. We trained a ResNet-18 architecture on a curated dataset of 1,252 high-resolution images representing six putative Ophrys species, implementing biologically informed data augmentation strategies to enhance model generalization across natural variations. Our methodology compared multi-class and one-vs-all classification approaches, evaluating performance through standard metrics and expert validation of model interpretability. The framework achieved macro-averaged F1-scores of 0.92 and 0.91 for multi-class and one-vs-all strategies, respectively, demonstrating robust performance across all target species. Through integrated gradients and occlusion sensitivity analysis, we identified that the model learns taxonomically relevant morphological features, with expert botanists confirming that highlighted regions correspond to established diagnostic traits used in traditional identification keys. Our analysis revealed systematic classification challenges between closely related taxa, providing insights into the limits of morphology-based identification. The explainable framework offers botanists interpretable visual explanations while maintaining high accuracy, presenting a practical tool for field identification and biodiversity assessment. These results demonstrate the potential for transfer learning to other fine-grained plant classification tasks and support automated approaches in conservation biology applications.

Different species/taxa of orchids from the Ophrys Genus

Our approach, OrchID, is designed to recognize various species and taxa within the Ophrys genus of orchids. The figure below illustrates a selection of these species, with images extracted from the dataset utilized in this study.

Data and code availability

The data employed in this study are available upon request. The source code of the machine learning pipeline we defined, as well as the resulting trained models, are freely available in the replication package we prepared at