Durian Tree Type Identification Based on Durian Leaves

Abstract

Durian tree can be identified by examining its leaves. Each type of durian has a different leaf pattern. Usually, farmers or fruit cultivators are able to differentiate durian tree-type saplings (young trees) from the leaf patterns. However, this task may be difficult to common people who want to buy durian saplings from a nursery or to identify the durian tree type plant if not labeled. Therefore, it is difficult to identify the types of durian trees from manual inspection. This project aims to develop an Android application for the durian tree types identification and assist as a second opinion reference for farmers and durian tree buyers in identifying durian tree types. In this project, the proposed method to classify different types of durian is the Convolutional Neural Network (CNN). TensorFlow Lite is used to developing the model and it is implemented by using Google Collaboration. Meanwhile, for the mobile application development, the Android Studio software is used for this project. The performance of the classification technique and mobile application are analyzed in terms of accuracy and functionality. The performance of the model is evaluated using five different classes which are the Black Thorn, IOI, Kim Hong, Musang King, and Red Prawn. This study discovered that the model has an accuracy of 86.00% for training, 77.78% for validation, and 66.67% for testing. Then, it has been exported into the Android Studio software and used in the recognition system for the developed mobile application. The developed mobile application has also functioned well in terms of image selection, image capture, image display, and recognition execution.