« Plant diseases are a major concern for global food security and sustainable agriculture. Timely and accurate diagnosis of plant diseases is essential to prevent their spread and mitigate their impact on crop yields. The use of digital images and deep learning techniques has shown great promise in automating the diagnosis of plant diseases. However, the identification of complicated symptoms and atypical diseases remains a challenge even for experts.
This study aims to contribute to the study of plant disease diagnosis by delving deeper into the analysis of symptoms of various diseases on oilseed rape leaves using deep learning techniques.
The focus is on developing advanced neural networks that can accurately diagnose and characterize plant diseases by using multiple views of the same symptoms and detecting atypical symptoms that may be difficult to identify even by experts.
The study incorporates classical techniques to study the decision-making processes of the networks, which will enable us to gain a better understanding of how the networks work and make decisions. Those techniques make deep models interpretable without sacrificing accuracy or retraining, therefore allowing for a more transparent and trustworthy model.
The research primarily utilizes RGB images of oilseed rape leaves acquired under strictly controlled conditions, such as standardized lighting and uniform background, to ensure consistency and accuracy. Additionally, we propose to explore the relationship between network accuracy and complexity, an important feature for future deployment as a mobile application, for instance. »