Efficient multiscale targeting and evidence-based response to crop diseases

Created on June 27, 2025

2025

Objective

Recent studies[1][2] demonstrate the potential of UAV imagery combined with computer vision models for pest and disease detection. However, CNNs typically struggle to capture long-distance dependencies, while ViTs typically underperform when faced with limited training data or insufficient training time. Moreover, existing approaches often rely on computationally intensive architectures that are not energy-inefficient. You will work on addressing these limitations through a hybrid [3] CNN-ViT architecture for precise local feature extraction and global multiscale context modelling.

How

You will integrate state-of-the-art DL models with UAV-acquired orthomosaic maps, alongside RGB and multispectral imagery, to build robust and efficient pipelines for disease detection. The project will focus on Alternaria alternata in potatoes, as well as Olive Leaf Spot in olive trees. To maximize data utility, transfer learning will be applied using advanced Convolutional Neural Network (CNN) architectures and Vision Transformers (ViTs) to extract rich, latent representations of data inputs, which will be integrated through a neural data fusion approach. You will perform network compression methods, including Quantization-Aware Training and structured prunning [4].

Outputs

A new open-source framework of compressed hybrid Computer Vision models for the detection of Alternaria alternata in potatoes and Olive Leaf Spot in olive trees from UAV-acquired orthomosaic maps. Validation of the framework for different datasets [5], with an emphasis on accuracy, memory and processing requirements. If the validation yields state of the art results and if time permits, the publication and presentation of the results in an international conference.

Prerequisites

  1. Solid understanding of Machine Learning and Deep Learning fundamentals
  2. High-level coding skills in Python
  3. Nice to have or willing to learn: Coding skills in C
  4. Nice to have or willing to learn: Understanding of modern DL compression techniques
  5. Willingness to contribute to the state-of-the-art Deep Learning computer vision models
  • [1] Wieme, J. et. al., (2024). “Ultra-high-resolution UAV-imaging and supervised deep learning for accurate detection of Alternaria solani in potato fields”.
  • [2] Ksibi, A. et. al., (2022). “MobiRes-net: a hybrid deep learning model for detecting and classifying olive leaf diseases”.
  • [3] Lu, W., et. al., (2023). “A CNN-transformer hybrid model based on CSWin transformer for UAV image object detection”.
  • [4] A. Papaioannou, C.S. Kouzinopoulos, D. Ioannidis, and D. Tzovaras. "An Ultra-Low-Power Embedded AI Fire Detection and Crowd Counting System for Indoor Areas"
  • [5] Wieme, J. et. al., (2024). “Ultra-high-resolution UAV-imaging and supervised deep learning for accurate detection of Alternaria solani in potato fields”.