Pengembangan Sistem Deteksi Dini Tipburn pada Daun Tanaman Selada (Lactuca sativa L.) Berbasis Computer Vision pada Plant Factory

Mumtaz Hatta Niha'I Sipayung, Ir. Andri Prima Nugroho, S.T.P., M.Sc., Ph.D, IPU., ASEAN Eng. APEC Eng.; Dr. Mohammad Affan Fajar Falah, S.T.P., M.Agr.

2025 | Skripsi | TEKNIK PERTANIAN

Tipburn adalah gangguan fisiologis akibat defisiensi kalsium yang umum terjadi pada tanaman selada (Lactuca sativa L.) dalam sistem budidaya lingkungan terkontrol seperti pada plant factory. Masalah ini menurunkan kualitas serta nilai ekonomi panen dan memerlukan inspeksi tambahan, yang jika dilakukan secara manual bersifat subjektif dan tidak efisien untuk pencegahan dini. Penelitian ini bertujuan untuk mengembangkan sistem deteksi dini tipburn secara otomatis berbasis computer vision dengan model YOLOv11. Dataset dikumpulkan dari empat kondisi lingkungan (A–D) selama 10 hari, kemudian dianotasi ke dalam empat tingkat keparahan tipburn (0%, 25%, 75%, dan 100%) menggunakan platform Roboflow dan diperbanyak dengan augmentasi hingga menghasilkan 1038 gambar. Proses pelatihan model dilakukan pada enam konfigurasi berbeda, membandingkan pendekatan Object Detection dan Instance Segmentation dengan variasi epoch (100, 125, dan 150). Hasil evaluasi menunjukkan bahwa model Instance Segmentation dengan 150 epoch memberikan kinerja terbaik, dengan nilai precision 87%, recall 74%, mAP50 80%, mAP50-95 59%, dan F1-score 80%. Model ini diimplementasikan dalam sistem berbasis Python untuk mendeteksi dan memvisualisasikan area gejala tipburn dengan cukup akurat. Studi ini membuktikan potensi YOLOv11 dalam mendukung deteksi dini gangguan fisiologis pada tanaman secara otomatis di plant factory.

Tip burn is a physiological disorder caused by calcium deficiency, commonly observed in lettuce (Lactuca sativa L.) cultivated in controlled environments such as plant factories. This issue significantly reduces both the quality and economic value of the yield and requires additional inspection efforts. Manual inspection, however, tends to be subjective and inefficient for early-stage prevention. This study aims to develop an automatic early detection system for tip burn using a computer vision approach based on the YOLOv11 model. The dataset was collected under four different environmental conditions (A-D) over 10 days, annotated into four severity levels of tip burn (0%, 25%, 75%, and 100%) using the Roboflow platform, and augmented to produce a total of 1038 images. The training process involved six different configurations, comparing Object Detection and Instance Segmentation approaches across varying epochs (100, 125, and 150). Evaluation results indicate that the Instance Segmentation model with 150 epochs delivered the best performance, achieving a precision of 87%, recall of 74%, mAP50 of 80%, mAP50–95 of 59%, and F1-score of 80%. This model was then implemented in a Python-based system that successfully detected and visualized tip burn-affected areas with high accuracy. The results highlight YOLOv11’s strong potential for supporting automated early detection of physiological disorders in lettuce cultivated within plant factory systems.

Kata Kunci : computer vision, deteksi dini, segmentasi, tipburn, YOLOv11