Implementasi Convolutional Neural Network untuk Sistem Prediksi Pigmen Fotosintesis pada Tanaman Secara Real Time

Abstract

It is common that evaluation on plant health is done by conducting measurement on photosynthetic pigments. Analysis of the presence or absence of some particular pigments could reveal any information about plant responses to the environment or climate changes. This is due to the fact that relative pigment concentrations are influenced by environmental factors such as light and nutrient availability. In this research, a non-destructive and rapid method was developed to identify the existence of photosynthetic pigments in plant leaf i.e. chlorophyll, carotenoid, and anthocyanin. The method used leaf’s RGB digital image as the color representation of the pigments contained in the plant being evaluated. The intelligence agent which is responsible to learn the data and provide information about the pigments was developed based on convolutional neural network (CNN) model. This model was chosen due to its capability to receive a digital image and automatically search for the best feature to learn it. Therefore, plant evaluation could run in real time. The result of the experiment reveals that CNN model could learn the color-pigment relationship very well. The best architecture is ShallowNet using Adam optimizer, batch size 30 and trained with 15 epoch. The MSE of the pigments prediction reaches 0.0055 (actual data range -0.2 up to 2.2) for training and 0.029 for testing.