Electronic Theses and Dissertation

Tanaman cabai merupakan komoditas hortikultura utama di indonesia, dengan produksi yang dominan dalam sektor pertanian. data badan pusat statistik (bps) dari 2021 hingga 2023 menunjukkan cabai konsisten berada dalam lima besar produksi sayuran. namun, produksi sering terhambat oleh penyakit yang disebabkan oleh mikroorganisme patogen, sehingga identifikasi penyakit sangat penting. penelitian ini menghasilkan data primer berupa gambar daun cabai dalam lima kelas yaitu daun sehat, daun keriting, daun virus kuning, daun bercak, dan daun whitefly yang telah dipublikasikan di roboflow. melalui enam kali pelatihan model deep learning menggunakan efficientdet, diperoleh model terbaik dengan nilai average precision (ap) pada intersection over union 0.5, 0.75, dan 0.5-0.95 masing-masing sebesar 70,3%, 60,5%, dan 49,2% dalam waktu pelatihan selama 126 menit 40 detik serta telah menandakan kinerja model yang baik saat melakukan deteksi objek pada data uji, dapat dilihat pada hasil nilai confidence prediction box dimulai dari nilai 50% hingga 100%. sehingga model ini optimal untuk aplikasi yang membutuhkan hasil cepat dan akurat.

Chili plants are the main horticultural commodity in Indonesia, with dominant production in the agricultural sector. Data from the Central Statistics Agency (BPS) from 2021 to 2023 shows that chili is consistently in the top five vegetable production. However, production is often hampered by diseases caused by pathogenic microorganisms, so disease identification is very important. This study produces primary data in the form of images of chili leaves in five classes, namely healthy leaves, curly leaves, yellow virus leaves, spotted leaves, and whitefly leaves which have been published in Roboflow. Through six trainings of the deep learning model using EfficientDet, the best model was obtained with an Average Precision (AP) value at Intersection over Union 0.5, 0.75, and 0.5-0.95 of 70.3%, 60.5%, and 49.2% respectively in a training time of 126 minutes 40 seconds and has indicated good model performance when detecting objects on test data, as can be seen in the results of the confidence prediction box value starting from 50% to 100%. So this model is optimal for applications that require fast and accurate results.