Diagnosis of orange tree fruit and leaf diseases based on a new deep learning model using a graphical user interface

Arman Foroughi, Jose M. Jimenez, Jaime Lloret,
Diagnosis of orange tree fruit and leaf diseases based on a new deep learning model using a graphical user interface,
Expert Systems with Applications,
Volume 289,
2025,
128304,
ISSN 0957-4174,
https://doi.org/10.1016/j.eswa.2025.128304.
(https://www.sciencedirect.com/science/article/pii/S0957417425019232)
Abstract: Identifying the spectrum of fruit and leaf disease is one of the important pillars for farmers to correctly diagnose orange disease and quickly treat it so that it does not spread to other orange trees. For this purpose, we designed an Android and iOS application that can detect orange fruit and leaf diseases, such as melanosis, black spots, canker, and greening, as well as detect healthy oranges. It also declares everything but oranges as Not Orange’. In this case, farmers no longer need specialists to diagnose fruit disease. A new deep learning model has been proposed and built to diagnose the type of orange fruit disease. KIVY framework and KV language were used for graphical programming in Android, iOS, Windows, Linux, and Raspberry Pi operating systems. To learn the proposed model, an image dataset of orange disease and orange tree leaves, containing 5073 images, was used. We have provided data images to the proposed deep learning model algorithm so that the new model can be trained by processing these images. Powerful hardware is required for image processing operations in deep learning networks. For this purpose, we used Google Collaboratory. The output of this newly trained model can be used in Windows and Linux. However, they cannot be used in mobile hardware or Android OS. Therefore, we converted the proposed new trained model to TensorFlow lite, which can be implemented in mobile phones and Android and iOS operating systems and has been optimized for this purpose. We then created an Android package using the python-for-Android project. The Buildozer tool was used to automate the entire process. The trained model could detect canker, melanosis, greening, and black spots with 98.29% accuracy. This application is easily available to farmers, who can easily detect diseases in oranges and orange tree leaves.
Keywords: Deep learning; Kivy; Buildozer; Raspberry Pi; Orange Fruit Diseases