Rancang Bangun Alat Klasifikasi Penyakit pada Daun Pohon Kapas (Gossypium) Menggunakan Metode Transfer Learning
DERBY SEPTIAN NUGRAHA, Wijayanti Dwi Astuti, S.Si., M.Sc., Ph.D.
2024 | Tugas Akhir | D4 Teknologi Rekayasa Instrumentasi dan Kontrol
Cotton trees with the Latin name Gossypium are one of the important commodity because they produce natural fibers that are used as raw materials in the textile industry. However, in cultivation, cotton trees are often infected with leaf diseases during their growth period, which causes a decrease in the quality and productivity of cotton fibers in cotton plantations. The rate of cotton loss due to cotton leaf disease can reach 24%. Therefore, if leaf diseases in cotton trees can be classified early, the quality and productivity of cotton trees can be maintained.
Based on these problems, this study aims to create a prototype of a disease classification tool for cotton tree leaves (Gossypium). This prototype uses a Raspberry PI 4 model b which will be connected to a camera so that it can take images. The model used in this study is based on a Convolutional Neural Network (CNN) which is built using transfer learning techniques, with the architecture models tested, namely MobileNetV2, MobileNetV2 + Inception, and VGG19 which have been previously trained using data from ImageNet. The dataset used in this study uses secondary data taken from the kaggle website and primary data is taken directly using a Raspberry Camera. The dataset has 3 categories of leaf classes, namely Healthy, Powdery Mildew, and Target Spot. Then the secondary dataset that will be split into training and validation data with a ratio of 80% -20%. Meanwhile, primary data is used separately as testing data. After conducting training and testing data, the highest accuracy model results obtained in this study were using the MobileNetV2 + Inception model achieving the highest Accuracy of 96.26%, Precision of 96.33%, Recall of 96.26%, and F1-Score of 96.28%.
Kata Kunci : Penyakit daun pohon kapas, Transfer Learning, CNN