Electronic Theses and Dissertation

Tanaman padi merupakan makanan pokok utama masyarakat Indonesia,
sehingga kestabilan produksinya sangat penting. Serangan penyakit pada tanaman
padi dapat menyebabkan penurunan hasil panen secara signifikan dan berdampak
negatif terhadap ekonomi petani. Penelitian ini berfokus pada identifikasi
kerusakan daun padi di Kecamatan Samatiga, Kabupaten Aceh Barat yang
merupakan salah satu daerah utama penghasil beras di Provinsi Aceh. Identifikasi
manual terhadap kerusakan daun padi memerlukan tenaga ahli, dan waktu yang
cukup lama. Untuk mengatasi hal ini, dikembangkan model klasifikasi jenis
kerusakan daun padi, khususnya akibat penyakit brown spot, iron toxicity, dan
serangan hama rice leaf folder. Dataset primer yang digunakan terdiri dari 600 citra
daun padi (200 untuk tiap kelas) yang dikumpulkan dari sepuluh lahan persawahan
di Desa Pucok Lueng. Kemudian citra diperbanyak dengan teknik augmentasi
yaitu rotate 90 degree, vertical flipping, dan cropping. Model dibangun
menggunakan algoritma Convolutional Neural Network (CNN) dan K-Nearest
Neighbors (k-NN). Model terbaik yang dibangun dengan mengimplementasikan
arsitektur CNN adalah model dengan data augmentasi, learning rate 0,001, dan
epoch 80 (accuracy 0,9167; precision 0,9254; recall 0,9167; F1-score 0,9175).
Sedangkan model terbaik yang dibangun dengan mengimplementasikan algoritma
k-NN adalah model dengan data augmentasi yang menggunakan gabungan
ekstraksi fitur Canny dan GLCM, dengan nilai k=8 (accuracy 0,9000; precision
0,9158; recall 0,9000; F1-score 0,9003). Penelitian ini diharapkan dapat
mendukung pengembangan teknologi pertanian berbasis kecerdasan buatan untuk
meningkatkan produktivitas dan kesejahteraan petani.

Rice is the staple food for most Indonesians, making stable production essential. Diseases affecting rice plants can significantly reduce yields and negatively impact the livelihoods of farmers. This study focuses on identifying rice leaf damage in Samatiga Subdistrict, West Aceh Regency, one of the main rice producing areas in Aceh Province. Manual identification of rice leaf damage requires expert knowledge and considerable time. To address this issue, a classification model was developed to detect types of rice leaf damage, particularly those caused by brown spot disease, iron toxicity, and rice leaf folder pest attacks. The primary dataset consisted of 600 rice leaf images (200 images for each of the three classes) collected from ten rice fields in Pucok Lueng Village. These images were then augmented to increase the dataset size using augmentation techniques such as 90-degree rotation, vertical flipping, and cropping. The models were built using the Convolutional Neural Network (CNN) and K-Nearest Neighbors (k-NN) algorithms. The best-performing CNN model was developed using augmented data, a learning rate of 0.001, and 80 epochs, achieving an accuracy of 0.9167, a precision of 0.9254, a recall of 0.9167, and an F1 score of 0.9175. On the other hand, the best performing k-NN model was trained using augmented data with a combination of Canny and GLCM feature extraction techniques and k=8, achieving an accuracy of 0.9000, precision of 0.9158, recall of 0.9000, and an F1 score of 0.9003. This study is expected to support the development of artificial intelligence-based agricultural technology to enhance productivity and improve the welfare of farmers.