CNN Architecture for Classifying Types of Mango Based on Leaf Images

Abstract

In such conditions, it is necessary to have a system that can automatically classify plant species or identify types of plant diseases using either machine learning or deep learning. The plant classification system for ordinary people who are not familiar with the field of crops is not an easy job, it requires in-depth knowledge of the field from the experts. This study proposes a system for identifying mango plant species based on leaves using the CNN method. The reason for proposing the CNN method from previous research is that the CNN method produces good accuracy. Most previous studies to classify plant species use the leaves of the plant. The purpose of this study is to propose a CNN architectural model in classifying mango species based on leaf imagery. The input image of colored mango tree leaves measuring 224x224 is trained based on the CNN architectural model that was built. There are 4 CNN architectural models proposed in the study and 1 transfer learning InceptionV4. Based on the evaluation test results of the proposed CNN architectural model, that the best architectural model is the third. The number of parameters of the third CNN architecture is 1,245,989 with loss values and accuracy during evaluation are 1,431 and 0.55. The largest number of parameters is transfer learning InceptionV3 21,802,784, but transfer learning shows the lowest accuracy value and the highest loss, namely 0.2, and 1.61.