Apple leaf disease detection: A comprehensive analysis of pre-trained models and platform development
Loading...
Supplementary material
Other Title
Authors
Singh, Shivinder Pal
Author ORCID Profiles (clickable)
Degree
Master of Applied Technologies (Computing)
Grantor
Unitec, Te Pūkenga – New Zealand Institute of Skills and Technology
Date
2025
Supervisors
Shakiba, Masoud
Varastehpour, Soheil
Varastehpour, Soheil
Type
Masters Thesis
Ngā Upoko Tukutuku (Māori subject headings)
Keyword
apple trees
leaf spots
plant disease
modelling
neural networks
image processing
deep learning
machine learning
pattern recognition systems in agriculture
leaf spots
plant disease
modelling
neural networks
image processing
deep learning
machine learning
pattern recognition systems in agriculture
ANZSRC Field of Research Code (2020)
Citation
Singh, S.P. (2025). Apple leaf disease detection: A comprehensive analysis of pre-trained models and platform development (Unpublished document submitted in partial fulfilment of the requirements for the degree of Master of Applied Technologies (Computing)). Unitec, Te Pūkenga - New Zealand Institute of Skills and Technology
https://hdl.handle.net/10652/6804
Abstract
Apples are one of the most popular and valuable fruits in the world, but they are also susceptible to many diseases affecting their leaves, for which it is important to identify and control them at an early stage. Conventional methods of disease classification are restrictive in the way that they are tedious, time-consuming, and require the expertise of professionals. This study explores the feasibility of applying deep learning models for apple leaf disease classification and how transfer learning can enhance the model accuracy across different datasets. It is important to determine how various architectures deal with real-world issues such as lighting variations, shadows, similar-looking disease symptoms, and intraclass variability.
For this purpose, different pre-trained Convolutional Neural Networks (CNNs), including InceptionV3, ResNet50V2, Xception, MobileNetV2, and InceptionResNetV2 were evaluated. These models were trained and then tested on various datasets with a set of different characteristics in terms of class imbalances, image quality, and augmentation techniques. Accuracy, precision, recall, and F1-score were used to measure the performance of the models and the study explored how the models perform with data augmentation, environmental noise, and class imbalances. Some models were efficient in distinguishing between different diseases to a certain level of detail, while others were more stable with respect to augmentation-induced distortions.
In addition, the research explores the practical implications through the development of an apple leaf disease detection platform. The platform enables users to upload an image and then this image can be processed to achieve an automated classification result. The platform integrates outputs from multiple models using a composite scoring approach to improve the reliability of predictions. This application shows how deep learning can be effectively applied in farming, plant disease control, and other areas of agricultural production. The study contributes to precision agriculture by providing insights into model robustness, dataset diversity, and deep learning-based disease classification. This research also reveals the strengths and weaknesses of different CNN architectures. The research demonstrates the effectiveness of using advanced deep-learning approaches for identifying apple diseases thus improving the scalability of disease management in agriculture.
Publisher
Permanent link
Link to ePress publication
DOI
Copyright holder
Author
Copyright notice
All rights reserved