The Early Detection of Apple Scab Using Multispectral Imagery Under Natural Illumination Conditions

Bleasdale, Alexander and Whyatt, Duncan and Blackburn, George and Carmo-Silva, Elizabete (2024) The Early Detection of Apple Scab Using Multispectral Imagery Under Natural Illumination Conditions. PhD thesis, Lancaster University.

Text (2024Bleasdalephd)
2024Bleasdalephd.pdf - Published Version
Available under License Creative Commons Attribution-NonCommercial-NoDerivs.
Download (28MB)

Abstract

Apples are a nutritious, globally significant crop, yet their production has a substantial environmental impact due to the reliance on heavy pesticide use within disease management strategies. Apple scab, caused by the fungus Venturia inaequalis, is the primary cause of yield losses among all the pathogens that affect apples globally, leading to the highest protection costs and volumes of fungicides used for control. Conventional disease management strategies rely on uniform spraying of protective fungicides, repeated frequently throughout the growing season. This expensive, chemical-intensive approach contributes to widespread environmental contamination in many apple-growing regions worldwide. The early detection of apple scab infections offers a promising alternative for disease control by enabling targeted fungicide spraying, thereby improving the effectiveness of control measures and reducing chemical usage. Remote sensing systems are powerful tools for the early detection of plant pathogens; however, practical solutions for apple scab monitoring in commercial orchards are limited. The challenges of early detection in orchards arise from diverse symptom variations, variable illumination conditions, tree physiologies, and influences from other stress factors. Demonstrating the feasibility of early apple scab detection by remote sensing systems in uncontrolled conditions represents a significant step towards applying such systems in commercial orchards. This thesis aims to develop a remote sensing strategy for the early detection of apple scab infections under natural illumination conditions. The research rationale will first be introduced, followed by a review of the requirements for early disease detection in orchards and the current strategies and technologies available, including imaging sensors, classification methods and acquisition platforms. The capabilities of several low-cost sensing systems (multispectral, thermal, and 3D cameras) for detecting early apple scab infections under natural illumination are then assessed. Results indicate that apple scab symptoms could be manually identified several days earlier from high-resolution near-infrared (NIR) imagery than equivalent RGB imagery due to major differences in NIR radiation absorption potential between healthy and infected tissue. RGB and NIR time series datasets comprising 150 individual plants were then acquired and used to train convolutional neural networks (CNNs) to enable rapid, automated classification of apple scab symptoms. These CNNs consistently classified apple scab infections earlier and more accurately from NIR imagery than RGB imagery. This research shows that NIR imagery is effective for early apple scab detection under natural illumination conditions, and automated identification of the disease can be achieved accurately and rapidly with CNN classification models. This represents a significant advancement towards applying remote sensing systems for disease monitoring in apple orchards.