Komariyah, Aprilia Nur and Prof. Yusuf Hendrawan,, STP, M.App.Life Sc., PhD and Dr. Ir. Sandra Malin Sutan,, MP (2023) Klasifikasi Kualitas Teh Hitam Menggunakan Metode Support Vector Machine (SVM) dan Convolutional Neural Network (CNN) Berbasis Citra Digital. Magister thesis, Universitas Brawijaya.
Abstract
Sebagai negara tropis, produksi teh hitam di Indonesia sangat besar. Berdasarkan kualitasnya, teh hitam di Indonesia telah diekspor ke beberapa negara. Dalam rangka memenuhi permintaan standar kualitas yang dibutuhkan di tiap negara, teh hitam diklasifikasikan menjadi tiga jenis, diantaranya grade A, grade B, dan grade C. tetapi, pada kenyataannya industri memiliki permasalahan pada pemenuhan standar quality control karena kebanyakan industri masih menggunakan metode manual. Maka dari itu tujuan dari penelitian ini ialah untuk mengklasifikan tiga jenis mutu teh secara otomatis dengan menggunakan metode support vector machine (SVM) dan convolutional neural network (CNN). Kegiatan penelitian dilaksanakan pada bulan Juli 2022 hingga Januari 2023. Sampel bahan didapatkan dari kebun teh Wonosari Kab.Malang serta teh hitam yang beredar di pasaran. Sebanyak 3000 data digunakan dimana 2100 data digunakan pada proses training-validasi dan 900 data digunakan untuk proses training. 9 jenis data input digunakan untuk metode SVM. Sedangkan empat tipe pre-trained network digunakan untuk metode CNN yakni arsitektur AlexNet, SqueezeNet, GoogleNet dan ResNet50. Serta digunakan solver Adam, RMSProp dan SGDm serta initial learning rate sebesae 0,0001 dan 0,00005. Berdasarkan analisis sensitivitas pada SVM didapatkan kesimpulan bahwa semakin banyak data inpu yang digunakan maka nilai akurasi akan semakin tinggi. Sedangkan fungsi kernel yang terbaik ialah Fine Gaussian SVM dengan rata-rata validasi sebesar 98,81%. Sedangkan dalam proses testing dengan menggunakan confussion matrix didapatkan nilai akurasi tertinggi pada fungsi kernel Linier SVM. Pada metode CNN didapatkan beberapa arsitektur dengan nilai training mencapai 100%. Pada proses testing didapatkan bahwa beberapa arsitektur terbaik CNN mampu mengklasifikasin teh hitam tepat 100% sesuai dengan kelasnya Berdasarkan hasil tersebut dapat disimpulkan bahwa SVM dan CNN mampu mengklasifikasikan teh hitam secara efektif, mudah, murah dan cepat.
English Abstract
As a tropical country, black tea production in Indonesia is very large. Based on its quality, black tea in Indonesia has been exported to several countries. In order to meet the demand for quality standards required in each country, black tea is classified into three types, including grade A, grade B, and grade C. However, in reality the industry has problems meeting quality control standards because most industries still use manual methods. Therefore the purpose of this research is to classify three types of tea quality automatically using the support vector machine (SVM) and convolutional neural network (CNN) methods. The research activities were carried out from July 2022 to January 2023. Material samples were obtained from the Wonosari tea garden, Malang Regency and black tea on the market. A total of 3000 data is used where 2100 data is used in the training-validation process and 900 data is used for the training process. 9 types of input data are used for the SVM method. Meanwhile, four types of pre-trained networks are used for the CNN method, namely the AlexNet, SqueezeNet, GoogleNet and ResNet50 architectures. It also used Adam, RMSProp and SGDm solvers and initial learning rates of 0.0001 and 0.00005. Based on the sensitivity analysis on SVM, it can be concluded that the more input data is used, the higher the accuracy value will be. While the best kernel function is Fine Gaussian SVM with an average validation of 98.81%. Meanwhile, in the testing process using the confusion matrix, the highest accuracy value was obtained for the Linear SVM kernel function. In the CNN method, several architectures with a training value of 100% are obtained. In the testing process it was found that some of the best CNN architectures were able to classify black tea exactly 100% according to its class. Based on these results it can be concluded that SVM and CNN are able to classify black tea effectively, easily, cheaply and quickly.
Item Type: | Thesis (Magister) |
---|---|
Identification Number: | 0423100002 |
Uncontrolled Keywords: | CNN, Klasifikasi, SVM, Teh Hitam,Black Tea, Classification, CNN, SVM |
Subjects: | 300 Social sciences > 338 Production > 338.1 Agriculture > 338.16 Production efficiency |
Divisions: | S2/S3 > Magister Keteknikan Pertanian, Fakultas Teknologi Pertanian |
Depositing User: | soegeng sugeng |
Date Deposited: | 24 Jul 2023 02:05 |
Last Modified: | 24 Jul 2023 02:05 |
URI: | http://repository.ub.ac.id/id/eprint/202066 |
![]() |
Text (DALAM MASA EMBARGO)
Aprilia Nur Komariyah.pdf Restricted to Registered users only until 31 December 2025. Download (3MB) |
Actions (login required)
![]() |
View Item |