Evaluation of Deep Learning for Image-based Black Pepper Disease and Nutrient Deficiency Classification
Choy Yuen Khew, Yi Qin Teow, Ee Tiing Lau, Siaw San Hwang, Chih How Bong, Nung Kion Lee
Abstract
Black pepper (Piper nigrum) diseases and nutrient deficiency can often be observed based on the symptoms exerted on its leaves. This paper aimed to investigate the effectiveness of employing a deep learning approach to classify black pepper disease and nutrient deficiency based on leaf images. We constructed a customized convolutionary neural network to determine how its training parameters would affect the prediction performances. Another two deep learning neural networks VGG16 and Inception V3, are also employed for comparisons. We have sampled 947 images from farms in Sarawak consisted of 8 classes in total. Image augmentation is performed on the images to produce a total of 9532 images. The result shows that the customized CNN performed slightly better than the other two deep learning approaches at a 0.98 sensitivity rate. Furthermore, image augmentation contributed to improving prediction performance for all the deep learning models. This study has demonstrated that deep learning is a feasible approach for classifying black pepper diseases and nutrient deficiency based on leaf images.