Website-Based Potato Leaf Disease Detection Using the EfficientNetV2L Deep Learning Architecture

Potato plants are seasonal crops, where potatoes are classified as a nutrient-rich food ingredient. The leaves of potato plants can be attacked by diseases caused by bacteria and viruses. The modern agricultural era makes farmers face many challenges related to the recognition of potato leaf diseases. Potato leaf diseases can be recognized visually, but the recognition carried out by farmers has a drawback, namely the process of identifying the type of disease that takes a long time, this can disrupt productivity and sustainability of the harvest. The purpose of this study is to produce a model that is able to recognize the type of potato leaf disease using the Convolution Neural Network (CNN) method using the EfficientNetV2L architecture. The EfficientNetV2Larchitecture has the ability to perform good feature extraction and high computational efficiency compared to other CNN architectures. The dataset used in this study is potato leaf images taken from Kaggle, there are 2 disease classes and 1 healthy class with a total of 2,300 potato leaf images. The model proposed in this study is able to identify potato leaf disease images well and has a testing accuracy of 99.00%, training accuracy of 99.35%, validation accuracy of 98.60%, precision of 98.95%, recall of 99.02% and F1-Score of 99.98%. The model was developed into a website-based system that can quickly and accurately identify potato leaf diseases. Implementing this website system offers the advantages of easy access and speed in the potato leaf disease identification process.