Garuda - Garba Rujukan Digital

Klasifikasi citra makanan/non makanan menggunakan metode Transfer Learning dengan model Residual Network Thiodorus, Gustavo; Prasetia, Anugrah; Ardhani, Luthfi Afrizal; Yudistira, Novanto
TEKNOLOGI: Jurnal Ilmiah Sistem Informasi Vol 11, No 2 (2021): July
Publisher : Universitas Pesantren Tinggi Darul 'Ulum (Unipdu) Jombang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.26594/teknologi.v11i2.2402

People's activities that often upload photos of food before eating are common in popular social media, such as Facebook, Instagram, Twitter, and Pinterest. The food image data circulating on social media can be used for business purposes, such as analyzing customer behavior patterns. However, not all of the uploaded images are food images, so that before the analysis is carried out, it is necessary to do a classification task between food images and non-food images. Therefore, the researcher proposes automatic food image classification using transfer learning method using the Residual Network model version of 18 (ResNet-18). Residual Network model is used because it has a residual connection mechanism to solve the vanishing gradient problem. In addition, transfer learning was chosen because this method leverages the features and weights that have been generated in the previous training process on large and more general data (Imagenet) and thus reduce computation time and increase accuracy. The test was carried out by comparing the capabilities of the ResNet18 model with AlexNet. In addition, the fine tuning and freeze layer methods used to improve the quality of the model were also carried out in this study. In the experiment, the data set was divided into 3,000 images for training data and 1,000 images for test data, while the evaluation used was correctness accuracy. The results obtained in the ResNet18 model, namely the fine tuning training method, produced an accuracy value of 0.981 while the freeze layer resulted in the best accuracy value of 0.988. The AlexNet model that uses the fine tuning training method produces an accuracy value of 0.970 while the freeze layer produces the best accuracy value of 0.978. It can be concluded that the mechanism with the best accuracy is found in the RestNet18 architecture using the freeze layer 1-3 with an accuracy of 0.988.

Klasifikasi citra makanan/non makanan menggunakan metode Transfer Learning dengan model Residual Network Thiodorus, Gustavo; Prasetia, Anugrah; Ardhani, Luthfi Afrizal; Yudistira, Novanto
TEKNOLOGI: Jurnal Ilmiah Sistem Informasi Vol 11, No 2 (2021): July
Publisher : Universitas Pesantren Tinggi Darul 'Ulum (Unipdu) Jombang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.26594/teknologi.v11i2.2402

People's activities that often upload photos of food before eating are common in popular social media, such as Facebook, Instagram, Twitter, and Pinterest. The food image data circulating on social media can be used for business purposes, such as analyzing customer behavior patterns. However, not all of the uploaded images are food images, so that before the analysis is carried out, it is necessary to do a classification task between food images and non-food images. Therefore, the researcher proposes automatic food image classification using transfer learning method using the Residual Network model version of 18 (ResNet-18). Residual Network model is used because it has a residual connection mechanism to solve the vanishing gradient problem. In addition, transfer learning was chosen because this method leverages the features and weights that have been generated in the previous training process on large and more general data (Imagenet) and thus reduce computation time and increase accuracy. The test was carried out by comparing the capabilities of the ResNet18 model with AlexNet. In addition, the fine tuning and freeze layer methods used to improve the quality of the model were also carried out in this study. In the experiment, the data set was divided into 3,000 images for training data and 1,000 images for test data, while the evaluation used was correctness accuracy. The results obtained in the ResNet18 model, namely the fine tuning training method, produced an accuracy value of 0.981 while the freeze layer resulted in the best accuracy value of 0.988. The AlexNet model that uses the fine tuning training method produces an accuracy value of 0.970 while the freeze layer produces the best accuracy value of 0.978. It can be concluded that the mechanism with the best accuracy is found in the RestNet18 architecture using the freeze layer 1-3 with an accuracy of 0.988.

Perbandingan Prediksi Penggunaan Listrik dengan Menggunakan Metode Long Short Term Memory (LSTM) dan Recurrent Neural Network (RNN) Selle, Nurfatima; Yudistira, Novanto; Dewi, Candra
Jurnal Teknologi Informasi dan Ilmu Komputer Vol 9, No 1: Februari 2022
Publisher : Fakultas Ilmu Komputer, Universitas Brawijaya

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.25126/jtiik.2022915585

Energi listrikÂ telah menjadi salah satu kebutuhan yang sangat penting dan membantu kehidupan manusia di era modern saat ini. Energi listrik yang tidak dapat disimpan dalam waktu yang lama dan harus dapat selalu tersalurkan menyebabkan penyedia energi listrik harus dapat mampu menyediakan energi listrik dengan tepat. Oleh karena itu, diperlukan sebuah sistem yang mampu melakukan prediksi terhadap penggunaan listrik dengan memanfaatkan data historis penggunaan listrik sebelumnya. Sehingga PT. PLN selaku penyedia energi listrik harus dapat mampu menyesuaikan jumlah listrik yang harus disediakan dengan permintaan kebutuhan pelanggan. Penelitian ini menggunakan metode Recurrent Neural Network (RNN) dan Long Short Term Memory (LSTM) yang merupakan metode pembelajaran deep learning. Kedua metode ini mampu mengolah data dan melakukan prediksi dengan format data time series. Proses implementasi yang dilakukan yaitu normalisasi data, transformasi data, pembangunan model, training, testing, denormalisasi, dan pengujian hasil prediksi menggunakan Root Mean Square Error (RMSE). Berdasarkan penerapan metode LSTM dan pengujian pada fitur data siang dan malam, didapatkan kondisi terbaik pada penggunaan untuk fitur data siang dengan panjang sequence 20, hidden size 8, 3 LSTM layer, dan 70% data training menghasilkan rata-rata RMSE 46,72, sedangkan untuk fitur data malam didapatkan panjang sequence 30, hidden size 8, 1 LSTM layer, dan 80% data training menghasilkan rata-rata RMSE 51,05. Perbandingan antar RNN dan LSTM menghasilkan LSTM mampu menghasilkan kinerja yang lebih baik pada penggunaan deret waktu yang lebih panjang.Â AbstractElectrical energy has become one of the most important needs and helped human life nowadays. The electrical energy that cannot be stored for a long time and must always be distributed leads to an obligation for electricity providers to provide appropriate electrical energy. Therefore, we need a system that can predict the use of electricity by leveraging historical data on previous electricity usage. It aims that PT. PLN as a provider of electrical energy can able to adjust the amount of electricity that must be provided with the demands of customer needs. Our method uses are Recurrent Neural Network (RNN) and Long Short Term Memory (LSTM), which is a deep learning architecture that able to capture time-series data. The process of implementing system is data normalization, data transformation, model building, training, testing, denormalization, and testing the prediction results using the Root Mean Squared Error (RMSE). Based on the experiments on day and night data features, the best conditions were obtained at the use for daylight data features with a sequence length of 20, hidden size of 8, 3 LSTM layers, and 70% data training resulted in an average RMSE of 46.72.Â For the night data feature, the best result was achieved with the sequence length of 30, hidden size of 8, 1 LSTM layer, and 80% of training data resulting in an average RMSE of 51.05. Comparison between RNNs and LSTM shows LSTM capable of producing better performance when the longer time series is incorporated.

Perangkingan Dokumen Berbahasa Arab berdasarkan Query dengan Metode Klasifikasi Naïve Bayes dan K-Nearest Neighbor Usfita Kiftiyani; Suprapto Suprapto; Novanto Yudistira
Techno.Com Vol 19, No 4 (2020): November 2020
Publisher : LPPM Universitas Dian Nuswantoro

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33633/tc.v19i4.3939

Penelitian tentang perangkingan dokumen pada temu kembali informasi saat ini mudah ditemukan, hal ini terkait perkembangan keilmuan dibidang penggalian informasi yang bergerak sangat cepat. Namun, Walaupun sudah penelitian yang menggunakan Bahasa Arab sebagai objek masih terbatas. Karena keterbatasan penggunaan dokumen Bahasa Arab untuk penelitian bidang penggalian informasi maka penulis mencoba melakukan pendekatan sederhana, yaitu dengan mengimplementasikan metode klasifikasi naïve bayes dan k-Nearest Neighbor (k-NN). Tujuan dari penelitian ini adalah untuk mengetahui apakah metode klasifikasi terutama naïve bayes dan k-NN dapat digunakan untuk melakukan perangkingan, dan juga membandingkan akurasi dari kedua metode tersebut. Berdasarkan penelitian yang dilakukan, didapatkan hasil bahwa perangkingan dengan metode klasifikasi dapat dilakukan dengan tingkat akurasi metode Naïve Bayes lebih baik dibandingkan dengan metode k-NN dengan rata-rata nilai F1 Measure mencapai 72%, rata-rata nilai precision mencapai 75%, dan rata-rata nilai recall mencapai 80%. Sedangkan hasil dari metode k-NN diperoleh rata-rata nilai F1 Measure mencapai 70%, rata-rata nilai precision mencapai 76%, dan rata-rata nilai recall mencapai 79%. Namun penelitian ini masih kurang dari segi teknik yang dilakukan, yaitu dengan menghilangkan proses stemming. Sehngga penulis memberikan saran untuk penelitian selanjutnya supaya bisa dilakukan proses stemming dan menggunakan metode perangkingan yang lebih baru.

Batik Classification Using Convolutional Neural Network with Data Improvements Dewa Gede Trika Meranggi; Novanto Yudistira; Yuita Arum Sari
JOIV : International Journal on Informatics Visualization Vol 6, No 1 (2022)
Publisher : Politeknik Negeri Padang

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.30630/joiv.6.1.716

Batik is one of the Indonesian cultures that UNESCO has recognized. Batik has a variety of unique and distinctive patterns that reflect the area of origin of the batik motif. Batik motifs usually have a 'core motif' printed repeatedly on the fabric. The entry of digitization makes batik motif designs more diverse and unique. However, with so many batik motifs spread on the internet, it is difficult for ordinary people to recognize the types of batik motifs. This makes an automatic classification of batik motifs must continue to be developed. Automation of batik motif classification can be assisted with artificial intelligence. Machine learning and deep learning have produced much good performance in image recognition. In this study, we use deep learning based on a Convolutional Neural Network (CNN) to automate the classification of batik motifs. There are two datasets used in this study. The old dataset comes from a public repository with 598 data with five types of motifs. Meanwhile, the new dataset updates the old dataset by replacing the anomalous data in the old dataset with 621 data with five types of motifs. The lereng motif is changed to pisanbali due to the difficulty of obtaining the lereng motif. Each dataset was divided into three ways: original, balance patch, and patch. We used ResNet-18 architecture, which used a pre-trained model to shorten the training time. The best test results were obtained in the new dataset with the patch way of 88.88 % ±0.88, and in the old dataset, the best accuracy was found in the patch way on the test data of 66.14 % ±3.7. The data augmentation in this study did not significantly affect the accuracy because the most significant increase in accuracy is only up to 1.22%.

ISSUES AND PROBLEMS IN BRAIN MAGNETIC RESONANCE IMAGING: AN OVERVIEW Novanto Yudistira; Daut Daman
TELKOMNIKA (Telecommunication Computing Electronics and Control) Vol 6, No 1: April 2008
Publisher : Universitas Ahmad Dahlan

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.12928/telkomnika.v6i1.551

There are many issues and problems in the brain magnetic resonance imaging (MRI) area that haven’t solved or reached satisfying result yet. This paper presents an overview of the various issues and problems of the segmentation, correction, optimization, description and their application in MRI. The overview is started by describing the segmentation properties that are the most important and challenging in MRI brain manipulation. Then correction for reconstructing the brain MRI cortex, classification is utilized to classify the segmented brain image, and also review the uses of description is the great prospecting issue while some neurologist need the information resulted from brain imaging process including their potential problems from application applied by each technique. In each case, it is provided some general background information.

Peran Big Data dan Deep Learning untuk Menyelesaikan Permasalahan Secara Komprehensif Novanto Yudistira
EXPERT: Jurnal Manajemen Sistem Informasi dan Teknologi Vol 11, No 2 (2021): December
Publisher : Universitas Bandar Lampung (UBL)

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.36448/expert.v11i2.2063

Peran sain data besar (Big Data) dan pembelajaran mesin dewasa ini tidak dapat terelakkan terutama untuk menganalisis data dan memberikan kecerdasan pada komputer agar bekerja secara otonom untuk menyelesaikan suatu pekerjaan tertentu. Perkembangan teknologi sensor dan internet membuat ketersediaan data tersebut melimpah yang selanjutnya dapat dilakukan analisis data dalam jumlah yang besar. Hal tersebut mempengaruhi bagaimana cara pandang komputasi dalam berbagai macam bidang baik ilmu alam maupun sosial. Data yang terkumpul dapat berupa beragam format dengan laju pertambahan yang cepat dan dinamis. Kita perlu algoritma atau model yang mumpuni untuk memahami dan menggali pengetahuan pada set data yang besar tersebut beserta rancangan modelnya yang secara otomatis mempunyai kemampuan memprediksi atau mendeteksi. Deep Learning dengan kapasitasnya yang besar serta hubungan korelasi antar neuron yang sangat banyak diharapkan mampu menjawab tantangan tersebut didukung oleh beberapa penelitian terkini pada penerapannnya di berbagai bidang keilmuan. Dalam paper ini akan dipaparkan contoh pemanfaatan Deep Learning pada Big Data yang telah kita lakukan pada pengenalan video aksi manusia pada Youtube, Segmentasi pada sel berskala besar, citra dada x-ray dan data time-series multi variabel hubungannya dengan pandemi COVID-19.

Facial Expression Recognition using Residual Convnet with Image Augmentations Fadhil Yusuf Rahadika; Novanto Yudistira; Yuita Arum Sari
Jurnal Ilmu Komputer dan Informasi Vol 14, No 2 (2021): Jurnal Ilmu Komputer dan Informasi (Journal of Computer Science and Information
Publisher : Faculty of Computer Science - Universitas Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.21609/jiki.v14i2.968

During the COVID-19 pandemic, many offline activities are turned into online activities via video meetings to prevent the spread of the COVID 19 virus. In the online video meeting, some micro-interactions are missing when compared to direct social interactions. The use of machines to assist facial expression recognition in online video meetings is expected to increase understanding of the interactions among users. Many studies have shown that CNN-based neural networks are quite effective and accurate in image classification. In this study, some open facial expression datasets were used to train CNN-based neural networks with a total number of training data of 342,497 images. This study gets the best results using ResNet-50 architecture with Mish activation function and Accuracy Booster Plus block. This architecture is trained using the Ranger and Gradient Centralization optimization method for 60000 steps with a batch size of 256. The best results from the training result in accuracy of AffectNet validation data of 0.5972, FERPlus validation data of 0.8636, FERPlus test data of 0.8488, and RAF-DB test data of 0.8879. From this study, the proposed method outperformed plain ResNet in all test scenarios without transfer learning, and there is a potential for better performance with the pre-training model. The code is available at https://github.com/yusufrahadika-facial-expressions-essay.

Pengenalan Aktivitas Manusia Menggunakan Sensor Akselerometer dan Giroskop pada Smatphone dengan Metode K-Nearest Neighbor Zainal Arifien; Fitra Abdurrahman Bachtiar; Novanto Yudistira
Jurnal Teknologi Informasi dan Ilmu Komputer Vol 9, No 1: Februari 2022
Publisher : Fakultas Ilmu Komputer, Universitas Brawijaya

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.25126/jtiik.2022915593

Pengenalan aktivitas manusia atau Human Activity Recognition (HAR) merupakan salah satu topik yang populer karena besarnya peluang untuk diterapkan di kehidupan sehari-hari. Tujuan dari pengenalan ini adalah untuk mengenali, mendeteksi, dan mengklasifikasikan aktivitas yang dilakukan manusia. Pengenalan aktivitas manusia adalah salah satu teknologi penting untuk memantau dinamisme seseorang sehingga dapat bermanfaat di berbagai hal. Selain untuk menjaga kesehatan, pencegahan penyakit, dan membantu menentukan jenis olah raga, HAR dapat dimanfaatkan juga untuk diterapkan pada bidang keamanan dan pengembangan teknologi. Penelitian ini menggunakan smartphone sebagai teknologi utama dalam memperoleh data dengan memanfaatkan sensor akselerometer dan giroskop yang telah tertanam di dalamnya. Terdapat 8 macam aktivitas yang diteliti dengan kombinasi lama waktu eksperimen 5, 10, dan 15 detik serta posisi smartphone dipegang bebas maupun di dalam saku celana kanan. Data yang diperoleh terdiri dari 3 sumbu (x, y, dan z) pada setiap sensor yang digunakan. Data tersebut kemudian melalui proses pengolahan dan klasifikasi menggunakan algoritme k-Nearest Neighbor (k-NN). Hasil akurasi yang didapat dalam penelitian ini mencapai 79,56%. Hasil yang diperoleh melalui penelitian ini menunjukkan bahwa perbedaan peletakan smartphone mempengaruhi hasil pengenalan aktivitas manusia secara stabil. Selain itu, perbedaan jumlah data akibat perbedaan lamanya waktu eksperimen dapat mengakibatkan perbedaan lamanya waktu komputasi. Penelitian ini menjadi penting karena hasilnya dapat menjadi batu loncatan bagi penelitian selanjutnya. Beberapa peluang pengembangan juga dilampirkan pada bagian akhir. AbstractHuman activity recognition (HAR) is one of the most popular topics because of the large opportunities for its application in life. The purpose of HAR is to recognize, detect and classify human activities. Human activity recognition is one of the important technologies for monitoring a person's dynamism so that it can be utilized in various ways. Apart from maintaining health, preventing disease, and helping determine the type of exercise, HAR can also be used to be applied in the field of security and technological developments. This study uses smartphones as the main technology in obtaining data by utilizing the built-in accelerometer and gyroscope sensors. There are 8 types of activities studied with a combination of 5, 10, and 15 seconds of experimental time and the position of the smartphone is carried freely or in the right trouser pocket. The data obtained consists of 3 axes (x, y, and z) on each sensor used. The data then processed and classified using the k-Nearest Neighbor (k-NN) algorithm. The accuracy results obtained in this study reaches 79.56%. The results obtained through this study indicate that differences in smartphone placement affect the results of human activity recognition stably. In addition, differences in the amount of data due to differences in the length of the experiment period can result in differences in the length of computation time. This research is important because the results can be used as material for further research assistance. Some development opportunities are also attached at the end.

Deteksi Covid-19 pada Citra Sinar-X Dada Menggunakan Pre-Training Deep Autoencoder Fadhil Yusuf Rahadika; Karina Amadea; Adhi Setiawan; Griselda Anjeli Sirait; Novanto Yudistira
Jurnal Ilmu Komputer & Agri-Informatika Vol. 8 No. 2 (2021)
Publisher : Departemen Ilmu Komputer - IPB University

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.29244/jika.8.2.95-104

Deteksi Covid-19 umumnya menggunakan tes laboratorium dengan metode RT-PCR untuk mendapatkan hasil yang akurat. Sayangnya, tes ini membutuhkan waktu yang cukup lama yaitu sekitar 24 jam untuk mendapatkan hasil. Selain menggunakan RT-PCR, beberapa penelitian menunjukkan bahwa deteksi menggunakan citra sinar-X menunjukkan hasil yang cukup akurat dengan waktu prediksi yang lebih cepat. Citra sinar-X yang didominasi warna dalam jangkauan grayscale dapat dikatakan memiliki karakteristik yang berbeda jika dibandingkan dengan citra secara umum, sehingga dalam penelitian ini eksperimen dilakukan terhadap pelatihan untuk kasus klasifikasi citra sinar-X dengan melatih model dari awal (scratch). Namun seringkali model yang dilatih tanpa adanya pretraining menyebabkan model tidak dapat mencapai performa yang cukup baik. Salah satu bentuk metode pretraining yang dapat digunakan adalah penggunaan autoencoder sebagai model untuk rekonstruksi citra. Dalam penelitian ini pelatihan menggunakan pretraining autoencoder menghasilkan akurasi terbaik sebesar 81.78% dengan tambahan metode CutMix, color manipulation, dan rotation sebagai augmentasi. Kami juga menunjukkan bahwa penambahan pretraining autoencoder secara konsisten dapat meningkatkan akurasi hingga 2.58% pada model yang dilatih dari awal (scratch).

Title

Found 51 Documents
Search

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Title Search

Found 51 Documents Search

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Abstract

Title

Found 51 Documents
Search