L'IA peut-elle identifier les espèces végétales à partir de photographies de feuilles ?

PlantNet, Seek, iNaturalist — applications qui transforment toute promenade en guide de terrain.

Background

PlantNet, Seek, and iNaturalist are mobile applications that allow users to upload photographs of plants and receive automated suggestions for species identification. These tools leverage advances in artificial intelligence and computer vision to analyze leaf images and suggest potential matches from a vast database of plant species.

AI-based plant identification relies on deep learning models, particularly convolutional neural networks (CNNs), which are trained on large datasets comprising labeled images of leaves. These models process images by extracting key morphological features such as leaf shape, venation patterns, margin structure, texture, and sometimes even color. Through training on thousands of annotated examples, the networks learn to map visual patterns to specific plant species. This capability enables rapid classification even for users with limited botanical knowledge.

Several studies have evaluated the accuracy of AI-driven plant identification systems. Research from PlantVillage, reported in May 2026, indicates that such systems can achieve classification accuracy exceeding 90% when trained on diverse and well-curated datasets. Accuracy may vary depending on image quality, species similarity, and the comprehensiveness of the training data. In some cases, these tools are used to support citizen science initiatives, agricultural monitoring, and ecological research.

However, challenges remain, including the need for extensive labeled datasets, handling of closely related species, and robustness to variations in lighting, angle, and background noise. Despite these limitations, AI-powered plant identification continues to improve and is increasingly integrated into both scientific and public platforms.