Bandeau Pl@ntAgroEco

Plant disease identification based on deep learning

Postdoctoral research project as part of the Pl@ntAgroEco project, supervised by Inria.

  • Title: "Plant disease identification based on deep learning" ; Inria.
  • Postdoctoral researcher: Jules Vandeputte.
  • Affiliated unit: Inria, UMR 5506 - LIRMM - ZENITH.
  • Project duration: 2023–2025.

Project summary:

As the primary source of sustenance for humans, cultivated plants play a key role in global food production. In particular, detecting and treating pests and diseases are crucial issues for ensuring food security. For example, pests and diseases cause a global yield loss of 20.5% for wheat, 20.2% for maize, and 27.2% for rice. In this context, new methods for early detection and recognition of plant diseases based on the automated identification of visual symptoms have emerged in recent years. Several scientific studies have been conducted using recent deep learning techniques to efficiently identify plant diseases. Despite very promising results, most of them have focused on data acquired in controlled laboratory conditions. Consequently, the models show a significant loss of performance when applied to field data. The goal of this postdoctoral work is to develop a robust model for recognizing pests and diseases based on field data, with the aim of being applicable in real-world conditions. Furthermore, our approach involves leveraging the data collected for this purpose to enhance the model, thereby pushing the boundaries of automated disease diagnosis.

See also