Reflectance versus Fluorescence Imaging: ANN-Based Approach for Predicting Phenol Content on Red Betel Leaves

Identification systems using computer vision are important in image recognition technology and artificial intelligence in the agricultural sector. Identifying the phenol content in red betel leaves (Piper crocatum Ruiz & Pav.) using computer vision made it possible to predict the phenol content quickly and efficiently, simplifying the process of classifying product quality. This research used computer vision methods combined with artificial neural networks (ANN) using reflectance and fluorescence images to obtain predictive modeling of phenol content in red betel leaves. The research was aimed to compare the ANN model to predict the phenol content in red betel leaves using reflectance and fluorescence images. The ANN model used 75 data from reflectance images, 75 data from fluorescence images, and 75 data from total phenol content. The color features of the image were used as input data and the total phenol content was used as output data. The dataset used was 75% as cumulative data and 25% as validation data. From the reflectance data, the best architecture chosen was 20-30-1 with a learning rate of 0.1, a momentum of 0.9, tansig as the activation function in the hidden layers, purelin as the activation function in the output layer, and trainlm as the learning function, which produced an MSE value of 0.0090, validation MSE of 0.0880, training R of 0.9777, and validation R 0.7944. Meanwhile, from the fluorescence data, the best architecture chosen was 20-10-1 with a learning rate of 0.1, momentum 0.9, tansig as activation function in the hidden layers, purelin as activation function in the output layer, and trainlm as a learning function, which produced an MSE value of 0.0099, validation MSE of 0.4368, training R of 0.9796, validation R of 0.4436. The reflectance imaging produced the best ANN modeling for predicting the phenol content of red betel leaves.