IDENTIFIKASI AROMA MINT BERDASARKAN FREKUENSI SINYAL SENSOR PIEZOELEKTRIK 12 CHANNEL MENGGUNAKAN CONVOLUTIONAL NEURAL NETWORK (CNN)

Cahaya Trinala Kinanti (Departemen Teknik Elektro, Universitas Brawijaya)
Adharul Muttaqin (Departemen Teknik Elektro, Universitas Brawijaya)
Muhammad Aziz Muslim (Departemen Teknik Elektro, Universitas Brawijaya)

Jurnal Mahasiswa TEUB
Vol. 11 No. 5 (2023)

This research aims to identify mint aroma based on multisensor signal frequency using CNN-1D model. The data collected by Solih Okur at Botanical Institute of Karlsruhe Institute of Technology (KIT), Germany. There are 6 types of mint aroma and the frequency signal data is taken using piezoelectric sensors with 12 types of QCM sensors using various layering materials The data that has been obtained then goes through a pre-processing stage to remove outliers and then followed by dividing the data into training data, validation data, and test data. The training process was conducted using the Keras Tensorflow framework with a CNN-1D architecture model. There are 3 stages of testing performed, testing system modeling performance by changing architecture and parameters, testing system identification performance, and testing the system using noise data. The model with the best results was obtained with accuracy up to 98% with prediction time 2s. This finding shows that scent identification technology using multisensor and CNN-1D can identify mint aroma effectively and efficiently. Keywords: Multisensor, Convolutional Neural Network (CNN), CNN-1D, Scent Identification, Signal Identification.