2021 13th International Conference on Machine Learning and Computing, 2021
We apply a transformer called tensor2tensor toolkit, which is based on Tensorflow, to overcome th... more We apply a transformer called tensor2tensor toolkit, which is based on Tensorflow, to overcome the Grapheme-to-Phoneme conversion problem. This study performs conversions to produce pronunciation symbols for certain letter sequences in Indonesian particularly. The unavailability of the G2P conversion system in Indonesian is currently being faced, so research is being carried out to create a system that can solve this problem by applying the Transformer. The transformer has a simple network architecture based solely on the attention mechanism, so we took advantage of eliminating convolution and redundancies—complex recurrent and convolution neural networks including encoders and decoders as the basis for the sequence transduction model. The excellent performance of the model is obtained through the attention mechanism by connecting the encoder and decoder. By using this tool, we carry out to compare among KBBI and CMU dictionary datasets. We attained a word error rate (WER) of 6,7% on the KBBI data set after training for three days on two core CPUs, which has an accuracy of 93,3%, improving over the existing best results CMU dictionary dataset for 26% word error rate. In this study, we carried out a detailed experimental evaluation by assessing the processing time and the error rate of words and then compared it with state of the art. By demonstrating this Transformer, this tool successfully generalizes and then applies it to several Indonesian elements with limited training data and large training data. We concluded that the transformer model is suitable for dealing with the G2P problem at hand for this task.
2020 8th International Conference on Information and Communication Technology (ICoICT), 2020
Cancer is the body' s tissue cells that continue to grow beyond normal and out of control so ... more Cancer is the body' s tissue cells that continue to grow beyond normal and out of control so that cancer cells push normal cells and cause death in normal cells. One type of cancer is cancer that attacks breast tissue or is called breast cancer. The sooner breast cancer is detected, it will increase the chance the patient will survive. One of the techniques in the early detection of breast cancer is mammography screening. To minimize human error in checking the results of mammography, a CAD system is needed in checking the results of mammography. Therefore, in this research, a system that can classify breast tissue from mammogram into three classes, namely normal, benign, and malignant has been built. The performance of the system reaches F1-Score 74.02%, Recall 76.15% and Precision 74.02%. The system achieves this performance by combining the Uniform Local Binary Pattern and GLCM features and the Random Forest classification method.
Jurnal Teknologi Informasi dan Ilmu Komputer, 2021
Ras dapat digunakan untuk mengkategorikan manusia dalam populasi atau kelompok besar. Oleh karena... more Ras dapat digunakan untuk mengkategorikan manusia dalam populasi atau kelompok besar. Oleh karena itu, pengenalan ras dapat berguna untuk mempermudah dalam mengidentifikasi seseorang dan membantu dalam mempersempit lingkup pencarian. Penggunaan wajah sebagai dasar pengenalan ras mengarahkan penelitian pada identifikasi penggunaan bagian wajah yang berpengaruh signifikan terhadap kinerja pengenalan ras. Pada penelitian ini bagian wajah berupa hidung dan mulut diidentifikasi untuk digunakan sebagai dasar pengenalan ras Mongoloid, Kaukasoid, dan Negroid. Ciri Gray Level Co-occurrence Matrix (GLCM) diekstrak dari bagian hidung dan mulut untuk selanjutnya diklasifikasi menggunakan Random Forest. Hasil eksperimen menunjukkan bahwa penggunaan ciri gabungan dari hidung dan mulut mampu menghasilkan kinerja sistem yang paling baik jika dibandingkan penggunaan hidung atau mulut saja.
Abstrak Parkir di tempat parkir besar itu sangat sulit. Artinya, pengemudi harus berkeliaran di d... more Abstrak Parkir di tempat parkir besar itu sangat sulit. Artinya, pengemudi harus berkeliaran di dalam area parkir untuk mencari tempat parkir kosong. Selain itu, diperlukan tenaga ekstra untuk mencari tempat parkir yang kosong, Sistem parkir konvensional saat ini menitik beratkan pengemudi untuk mencari lahan parkir sendiri yang dapat menambah pembuangan emisi gas dan waktu, Oleh karena itu dibutuhkan sebuah sistem untuk mencari ruang kosong yang terdapat pada tempat parkir. Pada penelitian ini penulis menggunakan metode Convolutional Neural Networks (CNN), dan untuk dataset yang digunakan berupa citra dari dataset publik PKLot. dalam mengenali ketersediaan lahan parkir yang kosong penulis menggunakan arsitektur AlexNet yang memiliki parameter lebih sedikit dari arsitektur lain. Hasil terbaik pada penelitian ini menunjukan rata rata akurasi 99% dengan data uji. Kata kunci: Convolutional Neural Networks (CNN), tempat parkir, PKLot. Abstract Parking in a large parking lot is very diff...
Di dunia akademik, mahasiswa tidak jarang mengalami kesulitan dalam menentukan mata kuliah yang a... more Di dunia akademik, mahasiswa tidak jarang mengalami kesulitan dalam menentukan mata kuliah yang akan diambil pada saat pendaftaran ulang. Terkadang mahasiswa berkonsultasi dengan dosen wali ataupun bertanya pada mahasiswa lain tentang mata kuliah yang sebaiknya ia ambil. Berdasar kondisi tersebut timbul gagasan untuk membangun suatu recommender system akademik yang akan menghasilkan rekomendasi mata kuliah pilihan untuk mahasiswa. Dengan sistem ini diharapkan mahasiswa dapat memilih mata kuliah pilihan yang tepat. Teknik yang diimplementasikan pada recommender system akademik adalah teknik hibrid, yaitu mengkombinasikan teknik collaborative filtering dengan content based filtering untuk menghasilkan rekomendasi tunggal. Teknik ini dipilih dengan tujuan agar dapat mengambil manfaat dan mengeliminasi kelemahan yang ada pada tiap teknik. Karena domain permasalahan yang spesifik, yaitu akademik, pada penelitian ini ditambahkan pula teknik knowledge based dengan tujuan untuk mendapatkan ...
2020 the 6th International Conference on Communication and Information Processing, 2020
In 2016 to 2017, the value of Asia's coffee market has increased rapidly with an annual rate ... more In 2016 to 2017, the value of Asia's coffee market has increased rapidly with an annual rate of 6% on average. Each region of coffee bean production has its own quality, which will affect the price and flavor. Conducting the sorting process manually will cost a lot of time and probably inaccurate. Therefore, technology will be required to make the selection process faster. We conduct a study to measure the quality of arabica green coffee beans based on their defect level and further classify them into five grades: specialty grade, premium grade, exchange grade, below grade, and off grade, by using the computer vision approach. We use the color histogram and Local Binary Pattern (LBP) to extract color and texture features of arabica green coffee beans. These feature extraction methods could well represent the quality of the beans. For the grade determination process, we compare the performance of Random Forest and K-Nearest Neighbor (KNN). From the experimental result, we successfully showed that the combination of color and texture visual feature and machine learning approach achieved promising results with an accuracy of 87.87% and 80.47%, by using Random Forest and KNN respectively.
2021 International Conference on Communication & Information Technology (ICICT), 2021
The range of voices is an essential aspect that a singer needs to know. This knowledge is necessa... more The range of voices is an essential aspect that a singer needs to know. This knowledge is necessary so that the singer can maximize their singing potential. This study discussed about how to classify someone's vocal range into four classes commonly used in choir using Mel-frequency Cepstral Coefficient (MFCC) for its feature extraction and Convolutional Neural Network (CNN) for the classification. This study emphasized how MFCC and CNN was able to solve human vocal type classification problem. It is assisted by WavAugment for augmentation to maximize the learning process. In this study, the data used were primary so that the data were collected through surveys and experiments conducted directly by the researchers. The data used also affect the classification result, where the data need to be sparse enough to avoid the model being overfitted. The experiment is giving a good result where the training accuracy reaches 91.83% and testing accuracy is 91.14%. This model (specifically the feature extractor) was able to outperform the STFT model that usually has a competitive result with 3.11% in training accuracy and 1.15% in testing accuracy. This study is a multi-disciplinary science that has a strong influence on music, especially in the choir. This study was conducted to improve choir music and computer technology continuity by combining music with computer science.
2021 International Conference on Software Engineering & Computer Systems and 4th International Conference on Computational Science and Information Management (ICSECS-ICOCSIM), 2021
Children’s emotions can affect the learning process, especially positive emotions, making them mo... more Children’s emotions can affect the learning process, especially positive emotions, making them more focused on learning. In addition, in terms of identifying someone’s emotions, we can represent them through facial expressions by combining the local binary pattern (LBP) and the local ternary patterns (LTP) method, known as Co-ChiLeRFE. The reasons for combining the two methods are that the LBP has proven to be very good at performing feature extraction, especially in describing textures. At the same time, LTP is adept at dealing with uniform motifs such as those on the face area. Subsequently, in this study, we used the NIMH child emotional faces picture set (NIMH-ChEFS), which has five-class expressions: sad, neutral, happy, angry, and afraid. To achieve optimal results in the Co-ChiLeRFE method, we set the LBP parameter as P = 8, R = 8, and the LTP parameter threshold value of one. The results we got from this experiment achieved a system performance superior accuracy of 92.51%.
2021 9th International Conference on Information and Communication Technology (ICoICT), 2021
An employee should be competent and expertise in their respective fields. An evaluation is needed... more An employee should be competent and expertise in their respective fields. An evaluation is needed to maintain the quality of employee’s performance, one of which can be done by observing their activity during working hours. This research discusses the classification of the employee’s activity in desk work. Classification of employee’s activity is investigated using ResNet and the Cyclical Learning Rate method in a novel dataset, i.e. vision-based employee activity. Classification is done by looking at three types of employee activities: talking on the phone, using a PC, and playing smartphone. The most optimal result of this research is ResNet50 using CLR with image input of 224x224x3 which has an accuracy of 87.01% and 12.99% error rate for talking on the phone, 99.95% accuracy and 0.05% error rate for using a pc, 81.67% accuracy and 18.83% error rate for playing smartphone, and has a decreasing loss value. In addition, this research shows that cyclical learning rate significantly affects the model performance.
Jurnal Teknologi Informasi dan Ilmu Komputer, 2021
Penggunaan ponsel sudah sangat erat dengan kehidupaan anak usia dini sehingga menimbulkan beberap... more Penggunaan ponsel sudah sangat erat dengan kehidupaan anak usia dini sehingga menimbulkan beberapa dampak negatif bagi anak usia dini terutama berkurangnya interaksi dengan dunia sekitarnya. Salah satu teknologi yang dapat dikembangkan pada ponsel adalah computer vision. Salah satu penggunaan computer vision adalah object recognition yang memberikan solusi untuk membantu mengenali objek. Pada penelitian ini dibangun sistem pengenalan objek benda di dalam rumah yang diaplikasikan pada ponsel yang diharapkan membantu anak usia dini mengenali benda disekitarnya. MobileNet merupakan salah satu feature extraction yang memiliki kinerja yang baik dan ringan digunakan pada perangkat ponsel. Arsitektur MobileNet terdiri dari layer depthwise convolution dan layer pointwise convolution dalam mengekstraksi fitur. Percobaan ini juga menggunakan arsitektur Single Shot Multibox Detector (SSD) sebagai metode dalam mendeteksi objek. Pre-trained model dari dataset COCO digunakan pada eksperimen, untu...
Khazanah Informatika: Jurnal Ilmu Komputer dan Informatika, 2020
Corn is one of the agricultural products that are essential as daily food sources or energy sourc... more Corn is one of the agricultural products that are essential as daily food sources or energy sources. Corn selection or sorting is important to produce high-quality seeds before its distribution to areas with varying conditions and agricultural characteristics. Hence, it is necessary to build corn seeds identification. In this paper, we propose a corn seed identification technique that incorporates the advantage of combining shape and color features. The identification process consists of three main stages, namely, ROI selection, feature extraction, and classification using the Artificial Neural Network (ANN) algorithm. The shape feature originates from the eccentricity value or comparison value between a distance of minor ellipse foci and major ellipse foci of an object. Meanwhile, the color features are extracted based on the HSV (Hue-Saturation-Value) channel. The experimental result shows that the proposed system achieves excellent performance for the identification of poor and good corn quality for BIMA-20 and NASA-29 species. The classification result for BIMA-20 Good vs. BIMA-20 Bad gives an accuracy of 89%, while the classification accuracy of BIMA-20 Good vs. NASA-29 Good is 97%.
Bulletin of Electrical Engineering and Informatics, 2020
In Indonesia, cancer is very burdensome financially for sufferers as well as for the country. Inc... more In Indonesia, cancer is very burdensome financially for sufferers as well as for the country. Increasing the access to early detection of cancer can be a solution to prevent the situation from worsening. Regarding the problem of cancer lesion detection, a whole-body bone scan image is the primary modality of nuclear medicine for the detection of cancer lesions on a bone. Therefore, high segmentation accuracy of the whole-body bone scan image is a crucial step in building the shape model of some predefined regions in the bone scan image where metastasis was predicted to appear frequently. In this article, we proposed an automatic whole-body bone scan image segmentation based on constrained local model (CLM). We determine 111 landmark points on the bone scan image as the input for the model building step. The resulting shape and texture model are further used in the fitting step to estimate the landmark points of predefined regions. We use the CLM-based approach using regularized land...
Advances in Intelligent Systems and Computing, 2016
Low-level image feature extraction is the basis of content based image retrieval (CBIR) systems. ... more Low-level image feature extraction is the basis of content based image retrieval (CBIR) systems. In that process, the usage of more than one descriptors has tremendous impact on the increasing of system accuracy. Based on that fact, in this paper we combined color and texture feature in the feature extraction process, namely Color Layout Descriptor (CLD) for color feature extraction and Edge Histogram Descriptor (EHD) for texture feature extraction. We measure the system performance on retrieving top-5, top-10, top-15, and top-20 relevant images. We successfully demonstrated in the experiment, that the combination of color and texture descriptor might be improved the performance of retrieval system, significantly. In our proposed system, the combination of CLD and EHD reaches 72.82% in accuracy, using adaptive weight in Late Fusion Method.
2016 8th International Conference on Information Technology and Electrical Engineering (ICITEE), 2016
Contours are one of the most commonly used shape descriptors in object recognition problem. In th... more Contours are one of the most commonly used shape descriptors in object recognition problem. In this paper, we proposed object recognition system based on shape. The shape is obtained by extracting the contour of the object in the image using common techniques in image processing domain. Further, the shape is represented by using chain coding technique and the chain coded representation is modified into the set of segments, with each segment has a particular weight in accordance with its length in its polygonal approximation of the object shape. For the purpose of similarity calculation, we modified a common algorithm used in Bioinformatics field, namely Needleman-Wunsch algorithm, in the term of scoring function. We created a new definition and implementation of the substitution matrix (for the purpose of scoring function), according to the characteristics of set of line segment. From the experiment we have conducted, we successfully shown that the weight of each segment of the object shape has positive impact in the similarity calculation, shown by the precision and recall value.
2016 4th International Conference on Information and Communication Technology (ICoICT), 2016
Object boundary has been proven as a powerful shape representation in several approaches, since t... more Object boundary has been proven as a powerful shape representation in several approaches, since they stable to illumination and variations in object color and texture. Besides that, commonly, the objects in the image are recognized based on the outline of two dimensions of the objects. However, extracting the object boundary automatically using a computer with a result that is consistent with human perception is very difficult. Regarding to that, in this paper we present a quantitative evaluation of five edge-based contour detection algorithm on multiclass fruit images. The algorithms are evaluated using direct pixel comparison and chamfer matching with regard to ground-truth boundaries. The results reported here provide a useful benchmark for current and future research efforts in contour detection.
We introduce a method for image segmentation that constraints the clustering with map and point d... more We introduce a method for image segmentation that constraints the clustering with map and point data. The method is showcased by applying the spectral clustering algorithm on aerial images for building detection with constraints built from a height map and address point data. We automatically detect the number of clusters using the elongated K-means algorithm instead of using the standard spectral clustering approach of a predefined number of clusters. The results are refined by filtering out noise with a binary morphological operator. Point data is used for semi-supervised labelling of building clusters. Our evaluation show that the combination of constraints have a positive impact on the clustering quality achieved. Finally we argue how the presented constraint types may be used in other applications.
2015 International Conference on Science in Information Technology (ICSITech), 2015
Color palette design is one important stage in color quantization. Color palette can reduce the n... more Color palette design is one important stage in color quantization. Color palette can reduce the number of color on images with minimal distortion, so the produced color looks similar to original image visually. In this paper, we proposed the use of color palette as a color descriptor, especially in fruit recognition system. We used k-means clustering method to cluster RGB pixels in order to construct fruit color palette. Removal of redundant RGB pixels, noise removal, determining of centroid of cluster, and determining threshold to refine cluster from outlier data are some steps we were conducted in order to build fruit color palette. Cluster with the highest percentage was chosen as fruit color palette representation of fruit images. Further, the centroid of the chosen cluster will be used as color descriptor to recognize fruit images based on color.
2015 International Conference on Electrical Engineering and Informatics (ICEEI), 2015
Color histogram has been widely used in feature extraction to represent color feature of an objec... more Color histogram has been widely used in feature extraction to represent color feature of an object in the image. In this paper, we identify which features that give high contribution in classification performance, because not all features are directly correlated with object category. In the case of n-bins color histogram, features were referred to color intensity range of color histogram. On the one hand, we consider fruit classification, where the feature space contains various properties of pixel intensities of RGB (Red-Green-Blue) channel. On selecting feature subset, we consider filter method of feature selection. In the filter method, we successively reduce the size of the feature sets and investigate the changes in the classification results. Specifically, we followed the filtering approach to feature selection: selecting features in a single pass first and then applying a classification algorithm independently. We used chi square feature selection to determine relevant features from RGB histogram. Further, we used and evaluated those relevant features in a classification system, using K-Nearest Neighbor (KNN) as classifier. In this paper we show that by conducting feature selection techniques combined with KNN we would be able to prune non-relevant intensities value of Red, Green, and Blue channel. Furthermore, we use the relevant subset of features to identify intensities range of RGB channel that was needed to represent 32 subcategories fruit image efficiently.
Product review banyak digunakan oleh calon pembeli untuk melihat opini pembeli lainnya, sebelum m... more Product review banyak digunakan oleh calon pembeli untuk melihat opini pembeli lainnya, sebelum memutuskan untuk membeli suatu produk. Kumpulan review produk tersebut juga digunakan oleh perusahaan untuk mengidentifikasi permasalahan produk yang dipasarkannya serta menemukan informasi strategi pemasaran perusahaan pesaingnya. Akan tetapi karena tidak ada quality control tentang penulisan review yang dapat dipercaya, maka setiap orang bebas untuk memberikan opininya sehingga tidak jarang kesempatan tersebut dimanfaatkan oleh spammer. Ada tiga jenis spam pada product review yaitu untruthfull opinions, reviews on brand only, dan non-review. Metode yang dipilih untuk menyelesaikan kasus review spam ini adalah Logistic Regression karena dapat menghasilkan estimasi probabilitas suatu review merupakan spam atau bukan seperti yang dibutuhkan. Dalam penelitian ini sistem dapat melakukan klasifikasi untuk beberapa tipe spam yaitu untruthfull opinions, reviews on brand only dan non-review. Berdasarkan hasil pengujian dan analisis diperoleh bahwa karakteristik untruthfull opinions spam yang banyak menjadi target spammer memiliki rating yang baik dan rata-rata yaitu rating antara 3-5, untuk selanjutnya akan diberi review spam negatif untuk menjatuhkan produk tersebut.
2021 13th International Conference on Machine Learning and Computing, 2021
We apply a transformer called tensor2tensor toolkit, which is based on Tensorflow, to overcome th... more We apply a transformer called tensor2tensor toolkit, which is based on Tensorflow, to overcome the Grapheme-to-Phoneme conversion problem. This study performs conversions to produce pronunciation symbols for certain letter sequences in Indonesian particularly. The unavailability of the G2P conversion system in Indonesian is currently being faced, so research is being carried out to create a system that can solve this problem by applying the Transformer. The transformer has a simple network architecture based solely on the attention mechanism, so we took advantage of eliminating convolution and redundancies—complex recurrent and convolution neural networks including encoders and decoders as the basis for the sequence transduction model. The excellent performance of the model is obtained through the attention mechanism by connecting the encoder and decoder. By using this tool, we carry out to compare among KBBI and CMU dictionary datasets. We attained a word error rate (WER) of 6,7% on the KBBI data set after training for three days on two core CPUs, which has an accuracy of 93,3%, improving over the existing best results CMU dictionary dataset for 26% word error rate. In this study, we carried out a detailed experimental evaluation by assessing the processing time and the error rate of words and then compared it with state of the art. By demonstrating this Transformer, this tool successfully generalizes and then applies it to several Indonesian elements with limited training data and large training data. We concluded that the transformer model is suitable for dealing with the G2P problem at hand for this task.
2020 8th International Conference on Information and Communication Technology (ICoICT), 2020
Cancer is the body' s tissue cells that continue to grow beyond normal and out of control so ... more Cancer is the body' s tissue cells that continue to grow beyond normal and out of control so that cancer cells push normal cells and cause death in normal cells. One type of cancer is cancer that attacks breast tissue or is called breast cancer. The sooner breast cancer is detected, it will increase the chance the patient will survive. One of the techniques in the early detection of breast cancer is mammography screening. To minimize human error in checking the results of mammography, a CAD system is needed in checking the results of mammography. Therefore, in this research, a system that can classify breast tissue from mammogram into three classes, namely normal, benign, and malignant has been built. The performance of the system reaches F1-Score 74.02%, Recall 76.15% and Precision 74.02%. The system achieves this performance by combining the Uniform Local Binary Pattern and GLCM features and the Random Forest classification method.
Jurnal Teknologi Informasi dan Ilmu Komputer, 2021
Ras dapat digunakan untuk mengkategorikan manusia dalam populasi atau kelompok besar. Oleh karena... more Ras dapat digunakan untuk mengkategorikan manusia dalam populasi atau kelompok besar. Oleh karena itu, pengenalan ras dapat berguna untuk mempermudah dalam mengidentifikasi seseorang dan membantu dalam mempersempit lingkup pencarian. Penggunaan wajah sebagai dasar pengenalan ras mengarahkan penelitian pada identifikasi penggunaan bagian wajah yang berpengaruh signifikan terhadap kinerja pengenalan ras. Pada penelitian ini bagian wajah berupa hidung dan mulut diidentifikasi untuk digunakan sebagai dasar pengenalan ras Mongoloid, Kaukasoid, dan Negroid. Ciri Gray Level Co-occurrence Matrix (GLCM) diekstrak dari bagian hidung dan mulut untuk selanjutnya diklasifikasi menggunakan Random Forest. Hasil eksperimen menunjukkan bahwa penggunaan ciri gabungan dari hidung dan mulut mampu menghasilkan kinerja sistem yang paling baik jika dibandingkan penggunaan hidung atau mulut saja.
Abstrak Parkir di tempat parkir besar itu sangat sulit. Artinya, pengemudi harus berkeliaran di d... more Abstrak Parkir di tempat parkir besar itu sangat sulit. Artinya, pengemudi harus berkeliaran di dalam area parkir untuk mencari tempat parkir kosong. Selain itu, diperlukan tenaga ekstra untuk mencari tempat parkir yang kosong, Sistem parkir konvensional saat ini menitik beratkan pengemudi untuk mencari lahan parkir sendiri yang dapat menambah pembuangan emisi gas dan waktu, Oleh karena itu dibutuhkan sebuah sistem untuk mencari ruang kosong yang terdapat pada tempat parkir. Pada penelitian ini penulis menggunakan metode Convolutional Neural Networks (CNN), dan untuk dataset yang digunakan berupa citra dari dataset publik PKLot. dalam mengenali ketersediaan lahan parkir yang kosong penulis menggunakan arsitektur AlexNet yang memiliki parameter lebih sedikit dari arsitektur lain. Hasil terbaik pada penelitian ini menunjukan rata rata akurasi 99% dengan data uji. Kata kunci: Convolutional Neural Networks (CNN), tempat parkir, PKLot. Abstract Parking in a large parking lot is very diff...
Di dunia akademik, mahasiswa tidak jarang mengalami kesulitan dalam menentukan mata kuliah yang a... more Di dunia akademik, mahasiswa tidak jarang mengalami kesulitan dalam menentukan mata kuliah yang akan diambil pada saat pendaftaran ulang. Terkadang mahasiswa berkonsultasi dengan dosen wali ataupun bertanya pada mahasiswa lain tentang mata kuliah yang sebaiknya ia ambil. Berdasar kondisi tersebut timbul gagasan untuk membangun suatu recommender system akademik yang akan menghasilkan rekomendasi mata kuliah pilihan untuk mahasiswa. Dengan sistem ini diharapkan mahasiswa dapat memilih mata kuliah pilihan yang tepat. Teknik yang diimplementasikan pada recommender system akademik adalah teknik hibrid, yaitu mengkombinasikan teknik collaborative filtering dengan content based filtering untuk menghasilkan rekomendasi tunggal. Teknik ini dipilih dengan tujuan agar dapat mengambil manfaat dan mengeliminasi kelemahan yang ada pada tiap teknik. Karena domain permasalahan yang spesifik, yaitu akademik, pada penelitian ini ditambahkan pula teknik knowledge based dengan tujuan untuk mendapatkan ...
2020 the 6th International Conference on Communication and Information Processing, 2020
In 2016 to 2017, the value of Asia's coffee market has increased rapidly with an annual rate ... more In 2016 to 2017, the value of Asia's coffee market has increased rapidly with an annual rate of 6% on average. Each region of coffee bean production has its own quality, which will affect the price and flavor. Conducting the sorting process manually will cost a lot of time and probably inaccurate. Therefore, technology will be required to make the selection process faster. We conduct a study to measure the quality of arabica green coffee beans based on their defect level and further classify them into five grades: specialty grade, premium grade, exchange grade, below grade, and off grade, by using the computer vision approach. We use the color histogram and Local Binary Pattern (LBP) to extract color and texture features of arabica green coffee beans. These feature extraction methods could well represent the quality of the beans. For the grade determination process, we compare the performance of Random Forest and K-Nearest Neighbor (KNN). From the experimental result, we successfully showed that the combination of color and texture visual feature and machine learning approach achieved promising results with an accuracy of 87.87% and 80.47%, by using Random Forest and KNN respectively.
2021 International Conference on Communication & Information Technology (ICICT), 2021
The range of voices is an essential aspect that a singer needs to know. This knowledge is necessa... more The range of voices is an essential aspect that a singer needs to know. This knowledge is necessary so that the singer can maximize their singing potential. This study discussed about how to classify someone's vocal range into four classes commonly used in choir using Mel-frequency Cepstral Coefficient (MFCC) for its feature extraction and Convolutional Neural Network (CNN) for the classification. This study emphasized how MFCC and CNN was able to solve human vocal type classification problem. It is assisted by WavAugment for augmentation to maximize the learning process. In this study, the data used were primary so that the data were collected through surveys and experiments conducted directly by the researchers. The data used also affect the classification result, where the data need to be sparse enough to avoid the model being overfitted. The experiment is giving a good result where the training accuracy reaches 91.83% and testing accuracy is 91.14%. This model (specifically the feature extractor) was able to outperform the STFT model that usually has a competitive result with 3.11% in training accuracy and 1.15% in testing accuracy. This study is a multi-disciplinary science that has a strong influence on music, especially in the choir. This study was conducted to improve choir music and computer technology continuity by combining music with computer science.
2021 International Conference on Software Engineering & Computer Systems and 4th International Conference on Computational Science and Information Management (ICSECS-ICOCSIM), 2021
Children’s emotions can affect the learning process, especially positive emotions, making them mo... more Children’s emotions can affect the learning process, especially positive emotions, making them more focused on learning. In addition, in terms of identifying someone’s emotions, we can represent them through facial expressions by combining the local binary pattern (LBP) and the local ternary patterns (LTP) method, known as Co-ChiLeRFE. The reasons for combining the two methods are that the LBP has proven to be very good at performing feature extraction, especially in describing textures. At the same time, LTP is adept at dealing with uniform motifs such as those on the face area. Subsequently, in this study, we used the NIMH child emotional faces picture set (NIMH-ChEFS), which has five-class expressions: sad, neutral, happy, angry, and afraid. To achieve optimal results in the Co-ChiLeRFE method, we set the LBP parameter as P = 8, R = 8, and the LTP parameter threshold value of one. The results we got from this experiment achieved a system performance superior accuracy of 92.51%.
2021 9th International Conference on Information and Communication Technology (ICoICT), 2021
An employee should be competent and expertise in their respective fields. An evaluation is needed... more An employee should be competent and expertise in their respective fields. An evaluation is needed to maintain the quality of employee’s performance, one of which can be done by observing their activity during working hours. This research discusses the classification of the employee’s activity in desk work. Classification of employee’s activity is investigated using ResNet and the Cyclical Learning Rate method in a novel dataset, i.e. vision-based employee activity. Classification is done by looking at three types of employee activities: talking on the phone, using a PC, and playing smartphone. The most optimal result of this research is ResNet50 using CLR with image input of 224x224x3 which has an accuracy of 87.01% and 12.99% error rate for talking on the phone, 99.95% accuracy and 0.05% error rate for using a pc, 81.67% accuracy and 18.83% error rate for playing smartphone, and has a decreasing loss value. In addition, this research shows that cyclical learning rate significantly affects the model performance.
Jurnal Teknologi Informasi dan Ilmu Komputer, 2021
Penggunaan ponsel sudah sangat erat dengan kehidupaan anak usia dini sehingga menimbulkan beberap... more Penggunaan ponsel sudah sangat erat dengan kehidupaan anak usia dini sehingga menimbulkan beberapa dampak negatif bagi anak usia dini terutama berkurangnya interaksi dengan dunia sekitarnya. Salah satu teknologi yang dapat dikembangkan pada ponsel adalah computer vision. Salah satu penggunaan computer vision adalah object recognition yang memberikan solusi untuk membantu mengenali objek. Pada penelitian ini dibangun sistem pengenalan objek benda di dalam rumah yang diaplikasikan pada ponsel yang diharapkan membantu anak usia dini mengenali benda disekitarnya. MobileNet merupakan salah satu feature extraction yang memiliki kinerja yang baik dan ringan digunakan pada perangkat ponsel. Arsitektur MobileNet terdiri dari layer depthwise convolution dan layer pointwise convolution dalam mengekstraksi fitur. Percobaan ini juga menggunakan arsitektur Single Shot Multibox Detector (SSD) sebagai metode dalam mendeteksi objek. Pre-trained model dari dataset COCO digunakan pada eksperimen, untu...
Khazanah Informatika: Jurnal Ilmu Komputer dan Informatika, 2020
Corn is one of the agricultural products that are essential as daily food sources or energy sourc... more Corn is one of the agricultural products that are essential as daily food sources or energy sources. Corn selection or sorting is important to produce high-quality seeds before its distribution to areas with varying conditions and agricultural characteristics. Hence, it is necessary to build corn seeds identification. In this paper, we propose a corn seed identification technique that incorporates the advantage of combining shape and color features. The identification process consists of three main stages, namely, ROI selection, feature extraction, and classification using the Artificial Neural Network (ANN) algorithm. The shape feature originates from the eccentricity value or comparison value between a distance of minor ellipse foci and major ellipse foci of an object. Meanwhile, the color features are extracted based on the HSV (Hue-Saturation-Value) channel. The experimental result shows that the proposed system achieves excellent performance for the identification of poor and good corn quality for BIMA-20 and NASA-29 species. The classification result for BIMA-20 Good vs. BIMA-20 Bad gives an accuracy of 89%, while the classification accuracy of BIMA-20 Good vs. NASA-29 Good is 97%.
Bulletin of Electrical Engineering and Informatics, 2020
In Indonesia, cancer is very burdensome financially for sufferers as well as for the country. Inc... more In Indonesia, cancer is very burdensome financially for sufferers as well as for the country. Increasing the access to early detection of cancer can be a solution to prevent the situation from worsening. Regarding the problem of cancer lesion detection, a whole-body bone scan image is the primary modality of nuclear medicine for the detection of cancer lesions on a bone. Therefore, high segmentation accuracy of the whole-body bone scan image is a crucial step in building the shape model of some predefined regions in the bone scan image where metastasis was predicted to appear frequently. In this article, we proposed an automatic whole-body bone scan image segmentation based on constrained local model (CLM). We determine 111 landmark points on the bone scan image as the input for the model building step. The resulting shape and texture model are further used in the fitting step to estimate the landmark points of predefined regions. We use the CLM-based approach using regularized land...
Advances in Intelligent Systems and Computing, 2016
Low-level image feature extraction is the basis of content based image retrieval (CBIR) systems. ... more Low-level image feature extraction is the basis of content based image retrieval (CBIR) systems. In that process, the usage of more than one descriptors has tremendous impact on the increasing of system accuracy. Based on that fact, in this paper we combined color and texture feature in the feature extraction process, namely Color Layout Descriptor (CLD) for color feature extraction and Edge Histogram Descriptor (EHD) for texture feature extraction. We measure the system performance on retrieving top-5, top-10, top-15, and top-20 relevant images. We successfully demonstrated in the experiment, that the combination of color and texture descriptor might be improved the performance of retrieval system, significantly. In our proposed system, the combination of CLD and EHD reaches 72.82% in accuracy, using adaptive weight in Late Fusion Method.
2016 8th International Conference on Information Technology and Electrical Engineering (ICITEE), 2016
Contours are one of the most commonly used shape descriptors in object recognition problem. In th... more Contours are one of the most commonly used shape descriptors in object recognition problem. In this paper, we proposed object recognition system based on shape. The shape is obtained by extracting the contour of the object in the image using common techniques in image processing domain. Further, the shape is represented by using chain coding technique and the chain coded representation is modified into the set of segments, with each segment has a particular weight in accordance with its length in its polygonal approximation of the object shape. For the purpose of similarity calculation, we modified a common algorithm used in Bioinformatics field, namely Needleman-Wunsch algorithm, in the term of scoring function. We created a new definition and implementation of the substitution matrix (for the purpose of scoring function), according to the characteristics of set of line segment. From the experiment we have conducted, we successfully shown that the weight of each segment of the object shape has positive impact in the similarity calculation, shown by the precision and recall value.
2016 4th International Conference on Information and Communication Technology (ICoICT), 2016
Object boundary has been proven as a powerful shape representation in several approaches, since t... more Object boundary has been proven as a powerful shape representation in several approaches, since they stable to illumination and variations in object color and texture. Besides that, commonly, the objects in the image are recognized based on the outline of two dimensions of the objects. However, extracting the object boundary automatically using a computer with a result that is consistent with human perception is very difficult. Regarding to that, in this paper we present a quantitative evaluation of five edge-based contour detection algorithm on multiclass fruit images. The algorithms are evaluated using direct pixel comparison and chamfer matching with regard to ground-truth boundaries. The results reported here provide a useful benchmark for current and future research efforts in contour detection.
We introduce a method for image segmentation that constraints the clustering with map and point d... more We introduce a method for image segmentation that constraints the clustering with map and point data. The method is showcased by applying the spectral clustering algorithm on aerial images for building detection with constraints built from a height map and address point data. We automatically detect the number of clusters using the elongated K-means algorithm instead of using the standard spectral clustering approach of a predefined number of clusters. The results are refined by filtering out noise with a binary morphological operator. Point data is used for semi-supervised labelling of building clusters. Our evaluation show that the combination of constraints have a positive impact on the clustering quality achieved. Finally we argue how the presented constraint types may be used in other applications.
2015 International Conference on Science in Information Technology (ICSITech), 2015
Color palette design is one important stage in color quantization. Color palette can reduce the n... more Color palette design is one important stage in color quantization. Color palette can reduce the number of color on images with minimal distortion, so the produced color looks similar to original image visually. In this paper, we proposed the use of color palette as a color descriptor, especially in fruit recognition system. We used k-means clustering method to cluster RGB pixels in order to construct fruit color palette. Removal of redundant RGB pixels, noise removal, determining of centroid of cluster, and determining threshold to refine cluster from outlier data are some steps we were conducted in order to build fruit color palette. Cluster with the highest percentage was chosen as fruit color palette representation of fruit images. Further, the centroid of the chosen cluster will be used as color descriptor to recognize fruit images based on color.
2015 International Conference on Electrical Engineering and Informatics (ICEEI), 2015
Color histogram has been widely used in feature extraction to represent color feature of an objec... more Color histogram has been widely used in feature extraction to represent color feature of an object in the image. In this paper, we identify which features that give high contribution in classification performance, because not all features are directly correlated with object category. In the case of n-bins color histogram, features were referred to color intensity range of color histogram. On the one hand, we consider fruit classification, where the feature space contains various properties of pixel intensities of RGB (Red-Green-Blue) channel. On selecting feature subset, we consider filter method of feature selection. In the filter method, we successively reduce the size of the feature sets and investigate the changes in the classification results. Specifically, we followed the filtering approach to feature selection: selecting features in a single pass first and then applying a classification algorithm independently. We used chi square feature selection to determine relevant features from RGB histogram. Further, we used and evaluated those relevant features in a classification system, using K-Nearest Neighbor (KNN) as classifier. In this paper we show that by conducting feature selection techniques combined with KNN we would be able to prune non-relevant intensities value of Red, Green, and Blue channel. Furthermore, we use the relevant subset of features to identify intensities range of RGB channel that was needed to represent 32 subcategories fruit image efficiently.
Product review banyak digunakan oleh calon pembeli untuk melihat opini pembeli lainnya, sebelum m... more Product review banyak digunakan oleh calon pembeli untuk melihat opini pembeli lainnya, sebelum memutuskan untuk membeli suatu produk. Kumpulan review produk tersebut juga digunakan oleh perusahaan untuk mengidentifikasi permasalahan produk yang dipasarkannya serta menemukan informasi strategi pemasaran perusahaan pesaingnya. Akan tetapi karena tidak ada quality control tentang penulisan review yang dapat dipercaya, maka setiap orang bebas untuk memberikan opininya sehingga tidak jarang kesempatan tersebut dimanfaatkan oleh spammer. Ada tiga jenis spam pada product review yaitu untruthfull opinions, reviews on brand only, dan non-review. Metode yang dipilih untuk menyelesaikan kasus review spam ini adalah Logistic Regression karena dapat menghasilkan estimasi probabilitas suatu review merupakan spam atau bukan seperti yang dibutuhkan. Dalam penelitian ini sistem dapat melakukan klasifikasi untuk beberapa tipe spam yaitu untruthfull opinions, reviews on brand only dan non-review. Berdasarkan hasil pengujian dan analisis diperoleh bahwa karakteristik untruthfull opinions spam yang banyak menjadi target spammer memiliki rating yang baik dan rata-rata yaitu rating antara 3-5, untuk selanjutnya akan diberi review spam negatif untuk menjatuhkan produk tersebut.
Uploads
Papers by Ema Rachmawati