Banana diseases detection mobile application

• Main Author: Lee, Wai Hin
• Format: Final Year Project / Dissertation / Thesis
• Published: 2022
• Subjects: Q Science (General); T Technology (General)
• Online Access: http://eprints.utar.edu.my/4656/1/fyp_CS_2022_LWH.pdf; http://eprints.utar.edu.my/4656/
• Institution: Universiti Tunku Abdul Rahman

Banana crop yields have significantly reduced due to the continuously exploitation of the banana disease in the plantation which happening in Malaysia. Farmers have to take action in order to resolve this matter. However, in a conventional way is that they will always perform trial and error to examine the symptoms of the banana disease of their banana plant. This action may require a lot of time and it is not efficient for an early-stage diagnosis mechanism. Other than that, not most of the farmers willing to pay a huge expenditure for advisory in their plantation to resolve the disease matter. An early-stage diagnosis and low budget expenditure proposed method are required to resolve problems that farmers faced. Due to the high affordability of mobile phone and advanced of computer vision task, a mobile application can be developed that deploy with the deep learning model can acts as a medium for farmers to perform classification on the banana leave to detect the banana disease within the banana plant. From literature reviews, researchers are carrying out study on techniques of using CNN architectures to perform classification of banana diseases and deploying into a mobile-based solution. There are several available systems on the market information written in literature review section of the report. A clearer understanding regarding of computer vision task on mobile application will be presented in this report based on the reviews of strengths and weaknesses of existing system on market. The final product of this project is a developed mobile application that capable to classify disease of the infected banana leaf images. The banana disease detector is based on the TensorFlow Lite model which converted from TensorFlow Keras model of EfficientNetV2-B0 by transfer learning. The EfficientNetV2-B0 model has achieved overall 95.00% on F1-score and 95.26% in testing score. F1-score is a score combined with precision and recall score in a single metric used to evaluate performance of classifiers The banana disease detector can classify infected banana leaf images under image condition such as low light intensity with flashlight on and minor obstacles covering infected banana leaf. However, it does have some limitation where unable to classify if the image was extremely dark, blurry, and taken from far distance. Other than that, it equipped with a comprehensive community feature allowed users to share their problems and thoughts and provided with different kind of banana diseases information such as disease symptoms, causal and prevention measures. The application backend is supported by the services from the Firebase platform that working with Java programming language in Android Studio.