Please use this identifier to cite or link to this item:
http://repository.ipb.ac.id/handle/123456789/166310| Title: | Perancangan Sistem Deteksi Viabilitas Bibit Cabai (Capsicum Annuum L) Berbasis Deep learning. |
| Other Titles: | Development of a Deep learning System for Detecting Viability in Chili Seedlings (Capsicum annuum L.). |
| Authors: | Supriyanto Solahudin, Mohamad Tasmara, Jasmine |
| Issue Date: | 2025 |
| Publisher: | IPB University |
| Abstract: | Persemaian merupakan tahapan penting dalam budidaya cabai yang
mempengaruhi kesehatan, ketahanan, dan performa pertumbuhan tanaman. Deteksi
dini tahap perkecambahan sangat penting untuk mengurangi risiko kegagalan
persemaian dan meningkatkan kualitas bibit. Namun, deteksi manual pada nursery
berskala besar menimbulkan tantangan efisiensi dan akurasi. Penelitian ini
bertujuan untuk mengembangkan sistem deteksi viabilitas bibit cabai berbasis deep
learning dengan menggunakan algoritma You Only Look Once (YOLO) yang
terintegrasi dengan sistem monitoring berbasis website menggunakan kamera RGB.
Proses penelitian meliputi: (1) akuisisi citra tahap awal perkecambahan, (2)
penyusunan dan pre-processing dataset, (3) anotasi dan pelabelan data, (4)
pelatihan model menggunakan YOLOv5 dan YOLOv8, (5) pengujian dan validasi
model, serta (6) implementasi sistem berbasis website. Dataset yang digunakan
terdiri dari 15.000 citra bibit cabai yang diklasifikasikan ke dalam tiga kelas:
berkecambah, tidak berkecambah, dan muncul daun kotiledon.
Hasil pelatihan menunjukkan bahwa model YOLOv8 memiliki performa
deteksi yang tinggi. Nilai presisi, recall, dan F1-score untuk masing-masing kelas
adalah sebagai berikut: berkecambah (0,938; 0,952; 0,945), tidak berkecambah
(0,953; 0,950; 0,951), dan muncul daun kotiledon (0,990; 0,982; 0,986). Rata-rata
makro (macro average) mencapai 0,961 untuk presisi, recall, dan F1-score, dengan
akurasi keseluruhan sebesar 96,25%. Model yang dikembangkan
diimplementasikan ke dalam sebuah website dashboard interaktif yang
memfasilitasi pemantauan persemaian secara real-time maupun berbasis unggahan
manual. Sistem ini menyediakan visualisasi deteksi, klasifikasi bibit, informasi
lokasi budidaya, serta integrasi hasil ke dalam basis data. Evaluasi kinerja aplikasi
sistem monitoring berbasis website menunjukkan bahwa performa deteksi
dipengaruhi oleh ukuran file citra, bandwidth jaringan, jumlah individu tanaman
dalam citra, jenis browser engine, dan spesifikasi perangkat. Proses deteksi lebih
cepat terjadi pada citra tanpa kompresi, jaringan berkecepatan tinggi, jumlah objek
sedikit, serta perangkat dan browser yang lebih efisien. Secara keseluruhan, sistem
dapat bekerja mendekati real-time dengan kondisi teknis yang optimal.
Evaluasi pertumbuhan benih cabai yang dideteksi menggunakan model deep
learning menunjukkan dinamika pertumbuhan yang unik. Pada awal pengamatan
(HSS 1-3), mayoritas benih berada dalam kategori tidak berkecambah. Seiring
berjalannya waktu, terjadi penurunan signifikan pada kategori ini, diikuti
peningkatan jumlah benih berkecambah yang mencapai puncaknya sekitar HSS 10-
14. Selanjutnya, benih berkecambah bertransisi menjadi kategori muncul daun
kotiledon, yang jumlahnya terus meningkat dan mendominasi pada HSS 20-25
hingga akhir periode pengamatan. Penelitian ini menunjukkan bahwa model deep
learning berbasis YOLO dapat secara efektif mengidentifikasi tahap pertumbuhan
awal bibit cabai dan berpotensi diterapkan sebagai sistem monitoring otomatis di
Nursery Greenhouse. Sistem ini tidak hanya meningkatkan efisiensi dan akurasi
dalam pemantauan bibit, tetapi juga memberikan solusi praktis dan aplikatif bagi
pengelolaan persemaian tanaman hortikultura. Pengembangan lebih lanjut dapat
mencakup perluasan dan diversifikasi dataset, optimalisasi sumber daya komputasi,
dan integrasi penuh dengan sistem Internet of Things (IoT) untuk pemantauan real-time yang lebih cerdas dan berkelanjutan. Nursery management is a crucial stage in chili cultivation, impacting the health, resilience, and growth performance of plants. Early detection of germination stages is vital to reduce the risk of nursery failure and improve seedling quality. However, manual detection in large-scale nurseries presents challenges in terms of efficiency and accuracy. This research aims to develop a deep learning-based chili seedling viability detection system using the You Only Look Once (YOLO) algorithm, integrated with a website-based monitoring system using an RGB camera. The research process includes: (1) image acquisition of early germination stages, (2) dataset compilation and pre-processing, (3) data annotation and labeling, (4) model training using YOLOv5 and YOLOv8, (5) model testing and validation, and (6) website-based system implementation. The dataset used consists of 15,000 chili seedling images classified into three categories: germinated, not germinated, and cotyledon appear. The training results indicate that the YOLOv8 model demonstrates high detection performance. The precision, recall, and F1-score values for each class are as follows: germinated (0,938; 0,952; 0,945), ungerminated (0,953; 0,950; 0,951), and cotyledon-emerged (0,990; 0,982; 0.986). The macro average reached 0,961 for precision, recall, and F1-score, with an overall accuracy of 96,25%. The developed model was implemented into an interactive website dashboard that facilitates real time and manual upload-based nursery monitoring. This system provides detection visualization, seedling classification, cultivation location information, and result integration into a database. Evaluation of the website-based monitoring system application performance shows that detection performance is influenced by image file size, network bandwidth, the number of individual plants in the image, browser engine type, and device specifications. The detection process is faster for uncompressed images, high-speed networks, fewer objects, and more efficient devices and browsers. Overall, the system can operate near real-time under optimal technical conditions. Evaluation of chili seed growth detected using the deep learning model revealed unique growth dynamics. In the early observation period (DAS 1-3), most seeds were in the ungerminated category. Over time, there was a significant decrease in this category, followed by an increase in the number of germinated seeds, which peaked around DAS 10-14. Subsequently, germinated seeds transitioned into the cotyledon-emerged category, whose number continuously increased and dominated from DAS 20-25 until the end of the observation period. This research demonstrates that the YOLO-based deep learning model can effectively identify the early growth stages of chili seedlings and has the potential to be implemented as an automated monitoring system in Nursery Greenhouses. This system not only improves efficiency and accuracy in seedling monitoring but also provides a practical and applicable solution for horticultural crop nursery management. Further development may include expanding and diversifying the dataset, optimizing computational resources, and full integration with Internet of Things (IoT) systems for smarter and more sustainable real-time monitoring. |
| URI: | http://repository.ipb.ac.id/handle/123456789/166310 |
| Appears in Collections: | MT - Agriculture Technology |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| cover_F1501231010_483461716ab34465848c1518df21696d.pdf | Cover | 610.04 kB | Adobe PDF | View/Open |
| fulltext_F1501231010_1bc29ec50aac459abed764b4d00b5a6c.pdf Restricted Access | Fulltext | 2.54 MB | Adobe PDF | View/Open |
| lampiran_F1501231010_fefb759b6e5045e6b7efedd9f03e8809.pdf Restricted Access | Lampiran | 619.84 kB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.