Plant Disease Identification Using EfficienNet

Abstract

Manually detecting plant diseases is time-consuming and prone to mistakes. Artificial intelligence (AI) and computer vision can be used to identify plant illnesses early, reducing their negative impacts while also overcoming some of the limitations of constant human monitoring. This experiment aims to utilize the Convolution Neural Network (CNN) algorithm to identify plant leaf diseases, to optimize
four CNN algorithms to identify plant leaf diseases, to make it easier for users to identify plant leaf diseases quickly and accurately. We propose using a deep learning architecture based on a recent CNN algorithm called EfficientNetB3, B4 and B5 on 66,556 plant disease leaf images sourced from Kaggle.com. The training phase with 57,067 data train images and 3,170 validation data images produces a model. For model testing, the testing phase was carried out with 3,171 image data tests, the overall test results yielded excellent accuracy and f1-scores for the three architectures, namely EfficientNetB3 0.9890%, EfficientNetB4 0.9912%, EfficientNetB5 0.9905%. Agriculture is anticipated to be the backbone of Indonesia because it is connected to both the national economy and the wellbeing of the populace.