Ratna Anggraeni, Siska and Dr. Ir. Sandra Malin Sutan,, MP and Prof. Ir. Yusuf Hendrawan,, STP., M.App.Life Sc., Ph.D. (2023) Prediksi Kandungan Flavonoid Daun Sirih Merah (Piper crocatum) berbasis Computer Vision Menggunakan Citra Reflektansi dan Fluoresensi dengan Pemodelan Artificial Neural Network. Sarjana thesis, Universitas Brawijaya.
Abstract
RINGKASAN Flavonoid adalah kelompok senyawa polifenolik yang ditemukan pada daun sirih merah yang berperan penting sebagai senyawa antioksidan, perlindungan dari sinar UV, dan bermanfaat bagi kesehatan. Pengukuran kandungan flavonoid umumnya dilakukan dengan metode destruktif dengan menggunakan spektrofotometer UV-Vis. Pengukuran dengan metode destruktif memiliki kekurangan yaitu menyebabkan kerusakan pada sampel daun, membutuhkan waktu yang lama dan biaya yang relatif mahal. Sehingga metode computer vision yang digabungkan dengan Artificial Neural Network dapat melakukan prediksi kandungan flavonoid pada daun sirih merah secara non destruktif. Tujuan dari penelitian ini adalah memprediksi kandungan flavonoid pada daun sirih merah menggunakan pemodelan ANN berdasarkan fitur tekstur warna serta memperoleh arsitektur ANN terbaik menggunakan citra reflektansi dan fluoresensi dari daun sirih merah. Fitur tekstur yang digunakan terdiri dari entropy, energy, homogeinity, sum mean, variance, contrast, correlation, maximum probability, inverse difference moment, dan cluster tendency dengan model warna RGB, HSV, HSL, Gray, dan L*a*b*. Tahap seleksi fitur dilakukan dengan metode filter dengan beberapa atribut seleksi yaitu Correlation, ReliefF, Chi-Squared, dan Gain ration. Berdasarkan penelitian diketahui rata-rata kandungan flavonoid pada daun muda sirih merah memiliki nilai 14,772 mg QE/g sementara pada daun tua memiliki nilai 10,820 mg QE/g. Pada data reflektansi diperoleh 4 fitur terbaik menggunakan Gain ratio sebagai atribut seleksi sementara pada data fluoresensi diperoleh 7 fitur terbaik menggunakan ReliefF sebagai atribut seleksi. Pada data reflektansi dan fluoresensi terpilih trainlm sebagai fungsi pembelajaran dan tansig sebagai fungsi aktivasi. Arsitektur ANN terbaik pada data reflektansi memiliki struktur 4-40-10-1 dengan learning rate 0,1 dan momentum 0,9. Sementara itu, arsitektur ANN terbaik pada data fluoresensi memiliki struktur 7-20-20-1 dengan learning rate 0,5 dan momentum 0,5. Arsitektur ANN terpilih pada data reflektansi menghasilkan MSE training 0,0097; MSE validasi 0,1567; R training 0,97132; dan R validasi 0,79416. Sementara itu, topologi ANN terpilih pada data fluoresensi menghasilkan MSE training 0,0094; MSE validasi 0,1493; R training 0,97254; dan R validasi 0,75083. Berdasarkan koefisien korelasi, citra reflektansi menghasilkan pemodelan ANN terbaik untuk prediksi kandungan flavonoid daun sirih merah.
English Abstract
SUMMARY Flavonoids are a group of polyphenolic compounds found in red betel leaves which play an important role as antioxidant compounds, protect from UV rays, and are beneficial for health.Typically, the measurement of flavonoid content is conducted using a destructive method involving UV-Vis spectroscopy. However, this method has its drawbacks, such as sample damage, prolonged time consumption, and relatively high costs. Hence, the combination of computer vision techniques and Artificial Neural Network (ANN) models can be utilized for the nondestructive prediction of flavonoid content in red betel leaves. The objective of this research is to predict the flavonoid content in red betel leaves through ANN modeling based on texture and color features. This study also aims to identify the optimal ANN architecture by utilizing reflectance and fluorescence images of red betel leaves. The texture features include entropy, energy, homogeneity, sum mean, variance, contrast, correlation, maximum probability, inverse difference moment, and cluster tendency, incorporating color models such as RGB, HSV, HSL, Gray, and Lab*. Feature selection is carried out through filter methods, including Correlation, ReliefF, Chi-Squared, and Gain ratio. Based on the research, it is known that the average flavonoid content in young red betel leaves is 14.772 mg QE/g, while in mature leaves, it is 10.820 mg QE/g. In reflectance data, four best features were obtained using Gain ratio as attribute selection, whereas in fluorescence data, seven best features were selected using ReliefF as the attribute selection method. In both reflectance and fluorescence data, trainlm was chosen as the learning function, and tansig was selected as the activation function. The optimal ANN architecture for reflectance data had a structure of 4-40-10-1 with a learning rate of 0.1 and momentum of 0.9. Meanwhile, the best ANN architecture for fluorescence data had a structure of 7-20-20-1 with a learning rate of 0.5 and momentum of 0.5. The selected ANN topology for reflectance data yielded a training MSE of 0.0097, validation MSE of 0.1567, training R of 0.97132, and validation R of 0.79416. On the other hand, the chosen ANN topology for fluorescence data resulted in a training MSE of 0.0094, validation MSE of 0.1493, training R of 0.97254, and validation R of 0.75083. Based on the correlation coefficients, reflectance images produced the best ANN modeling for predicting the flavonoid content in red betel leaves.
| Item Type: | Thesis (Sarjana) |
|---|---|
| Identification Number: | 052310 |
| Uncontrolled Keywords: | Artificial Neural Network; Flavonoid; Piper crocatum |
| Divisions: | Fakultas Teknologi Pertanian > Teknologi Industri Pertanian |
| Depositing User: | Unnamed user with username saputro |
| Date Deposited: | 15 Jan 2024 07:49 |
| Last Modified: | 15 Jan 2024 07:49 |
| URI: | http://repository.ub.ac.id/id/eprint/210367 |
![[thumbnail of DALAM MASA EMBARGO]](http://repository.ub.ac.id/style/images/fileicons/text.png) |
Text (DALAM MASA EMBARGO)
SISKA RATNA ANGGRAENI.pdf Restricted to Registered users only until 31 December 2025. Download (3MB) |
Actions (login required)
 |
View Item |