View Item 
      •   IPB Repository
      • Dissertations and Theses
      • Undergraduate Theses
      • UT - School of Data Science, Mathematic and Informatics
      • UT - Statistics and Data Sciences
      • View Item
      •   IPB Repository
      • Dissertations and Theses
      • Undergraduate Theses
      • UT - School of Data Science, Mathematic and Informatics
      • UT - Statistics and Data Sciences
      • View Item
      JavaScript is disabled for your browser. Some features of this site may not work without it.

      Deteksi Tingkat Kematangan TBS Kelapa Sawit Menggunakan Metode Convolutional Neural Network dan MobileNetV2

      Thumbnail
      View/Open
      Cover (597.7Kb)
      Fulltext (1.791Mb)
      Lampiran (385.7Kb)
      Date
      2025
      Author
      Pramudya, Ravi Mahesa
      Alamudi, Aam
      Mualifah, Laily Nissa Atul
      Metadata
      Show full item record
      Abstract
      Oil palm is one of the main plantation commodities in Indonesia with high productivity and a strategic role in the national economy. High productivity needs to be accompanied by good harvest quality to achieve optimal profit. The quality of palm oil is strongly influenced by the ripeness level of the fruit, yet classification in the field still relies heavily on visual inspection, which is prone to errors. Image classification based on artificial intelligence offers a potential solution for automatic and accurate detection of fruit ripeness. This study compares the performance of Convolutional Neural Network (CNN) and MobileNetV2 based on transfer learning in detecting the ripeness level of fresh fruit bunches (FFB) of oil palm. The dataset consists of 690 FFB images categorized into three classes: unripe, ripe, and overripe. The research stages include data splitting, preprocessing, augmentation, modeling using CNN and MobileNetV2, evaluation with 4-fold cross validation, and analysis of hyperparameter factors. The results indicate that model architecture is the main factor determining performance, with MobileNetV2 showing superior and more consistent results compared to CNN. The best-performing model was obtained from MobileNetV2 with batch size 32, Adam optimizer, and without class weight, achieving an average balanced accuracy of 87.50%, precision of 93.20%, recall of 90.97%, and F1-score of 89.74%. Furthermore, the three-way ANOVA test on batch size, optimization, and class weight showed that batch size had a significant effect on CNN, while on MobileNetV2 these factors had no significant effect.
       
      Kelapa sawit merupakan salah satu komoditas perkebunan utama di Indonesia yang memiliki produktivitas tinggi dan peran strategis dalam perekonomian nasional. Tingginya produktivitas perlu diimbangi dengan kualitas hasil panen yang baik untuk memperoleh keuntungan yang optimal. Kualitas produk minyak sawit sangat dipengaruhi oleh tingkat kematangan buah sawit. Permasalahan yang terjadi ialah klasifikasi tingkat kematangan buah sawit di Indonesia masih banyak mengandalkan pendekatan visual, yang rentan terhadap kesalahan. Klasifikasi citra berbasis kecerdasan buatan menjadi pendekatan potensial untuk mendeteksi tingkat kematangan buah sawit secara otomatis dan akurat. Penelitian ini membandingkan kinerja Convolutional Neural Network (CNN) dan MobileNetV2 berbasis transfer learning dalam mendeteksi tingkat kematangan tandan buah segar kelapa sawit. Dataset terdiri dari 690 citra TBS kelapa sawit yang dikategorikan ke dalam tiga kelas, yaitu belum matang, matang, dan terlewat matang. Proses penelitian mencakup pembagian data, praproses data, augmentasi, pemodelan dengan CNN dan MobileNetV2, evaluasi menggunakan 4-fold cross validation, serta analisis faktor berpengaruh. Hasil penelitian ini menunjukkan bahwa terdapat perbedaan performa antara kedua model, di mana MobileNetV2 terbukti lebih unggul dan konsisten dibandingkan CNN. Model dengan arsitektur MobileNetV2, batch size 32, optimasi Adam, dan tanpa penggunaan class weight dipilih sebagai model terbaik yang menunjukkan performa paling stabil dengan rata-rata balanced accuracy 87,50%, precision 93,20%, recall 90,97%, dan F1-score 89,74%. Selain itu, hasil uji ANOVA tiga arah terhadap batch size, optimasi, dan penggunaan class weight menunjukkan bahwa batch size merupakan hyperparameter yang berpengaruh signifikan pada model CNN, sedangkan pada MobileNetV2 pengaruh ketiga faktor tersebut tidak signifikan.
       
      URI
      http://repository.ipb.ac.id/handle/123456789/170967
      Collections
      • UT - Statistics and Data Sciences [82]

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository
        

       

      Browse

      All of IPB RepositoryCollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

      My Account

      Login

      Application

      google store

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository