Parallel Session 3
Sunday, July 18 10:30 - 11:45
10:30 A Comparison of Several Approaches for Image Recognition used in Food Recommendation System
In food recommendation systems, users can use mobile devices to capture images of the dishes they eat. The types of dishes in images will be automatically recognized and input to the recommendation system to suggest other fishes which the users likely to enjoy. Food image recognition is therefore an essential part of the food recommendation system. This used to be a hard problem in computer vision as many food are very similar in color and texture. Thanks to the development of artificial intelligent and especially deep learning techniques, it is much easier to build a program to recognize the type of foods in the image. This paper examines several approaches, from traditional machine learning to state-of-the-art deep learning techniques for food image recognition to provide a comparison of the performance of these techniques. To this end, a new dataset of Vietnamese cuisine including 12,017 photos of 15 dishes has been built to test algorithms. Traditional machine learning techniques including Histogram of Gradient (HOG) and Scale-Invariant Feature Transform (SIFT) and state-of-the-art deep learning models including VGG16, MobileNet, ANN, Resnet18, Resnet50, Densenet121 have been used for extracting features in the food images. Logistic Regression and softmax classification have been used for classification using the extracted features. Based on the comparison results provided in this paper, one can choose an appropriate techniques for image recognition to build a good food recommendation system.
10:45 Interpolation Effect on FBP-Based Image Reconstruction of Measured L-Band Microwave Tomography
This paper presents an interpolation effect on image reconstruction using Filtered Back Projection (FBP) algorithm for L-band microwave tomography. The FBP-based image reconstruction is conducted by changing the image of the object using radon transformation. The FBP method is chosen for the reconstruction due to no time consuming on its computation process. After conducting the reconstruction process, the interpolation is applied on the reconstructed image so that the image produced can be more stable. Here, the interpolation method is using linear, nearest, and cubic method at the L-band frequency of 1.5 GHz. The results show that the interpolation using linear method has Mean Square Error (MSE) value of 0.1335. Meanwhile, the nearest and cubic methods has the MSE value of 0.1392 and 0.1323, respectively. In case of using the Structure Similarity Indexing Method (SSIM), it shows that the linear method produces greater similarity than other methods, with the value of 0.3399. Whilst nearest-neighbor and cubic methods have the SSIM value of 0.3506 and 0.3404, respectively.
11:00 Experimental Approach of L-Band Microwave Tomography and Its Image Reconstruction Using CS Method
Tomography is a technique for observing internal parts of an object through microwave radiation emitted from various angles. This paper contributes to image reconstruction through an experimentally microwave tomography approach with a compressive sensing (CS) method at L-band frequency. The method is chosen due to its efficiency in image reconstruction and the ability to reconstruct an image with a limited input. The process is carried out by taking cross-sectional images of tree trunks resulting from the reconstruction of measured data with different pixel size. The results indicate that the CS method yields the image reconstruction with good quality. The peak signal noise to ratio (PSNR) value at 36 x 36 pixel resolution is about 57.423 dB.
11:15 Designing a tourism recommendation system using a hybrid method
Recommendation system is an application model based on observing the customer’s circumstances and desires. In the recommendation system, several methods are used to support how the system works in producing information. This paper’s recommendation system uses the Hybrid method (Collaborative Filtering & Content-Based Filtering) and the Nearest Neighbor (NN) algorithm. The method is a weakness of collaborative Filtering when an item does not have a rating at all. Therefore, it takes a method that can overcome the weaknesses of the method. The method used is the Hybrid method, which combines several methods found in the recommendation system to produce recommendation items according to the user’s wishes.This research will also utilize tools, namely, search engine optimization (SEO) and google analytics. This technique can help the system that has been developed appear on the first page of the search engine when visitors search for references about tourist attractions. In addition, this system will use Whatsapp Messenger as a promotional medium. This paper aims to propose a recommendation system that will help some historical tourism in Palembang become the leading destination in search engines.
10:30 A Study on Diacritic Restoration Problem in Vietnamese Text using Deep Learning based Models
Diacritic restoration is a challenging problem in natural language processing (NLP). With diacritic restoration, one can text faster and easier. Diacritic restoration is also helpful in making use of diacritic-missing texts, which are normally discarded in many NLP applications. This paper deals with the diacritic restoration problem for Vietnamese text. Three state-of-the-art deep learning models including Gated Recurrent Unit, Bidirectional Long-short Term Memory and Bidirectional Gated Recurrent Unit have been examined for the problem and the last one turned out to be the best among them. Besides deep learning models, it was found in this paper that word tokenization, which is the final pre-processing step applied on the data before feeding it to deep learning models also have influences on the final accuracy. Between two examined word tokenization methods: morpheme-based tokenization and phrase-based tokenization, the former yield better results regardless of the applied deep learning models. The experimental results show that the combination of morpheme-based tokenization and Bidirectional-GRU achieve the best performance of diacritic restoration with the Bleu-score of 88.06\%.
10:45 30VNFoods: A Dataset for Vietnamese Foods Recognition
In this paper, we construct an image dataset of typical Vietnamese foods taken from the Internet. We build the dataset by analyzing, collecting, preprocessing, and labeling, then conducting experiments on some modern convolutional networks such as VGG, Inception, ResNet. The accuracy is not very good (69.08%) but the top 5-accuracy is quite high (94.12%). From that, we draw the difficulties and methods to develop the dataset as well as solve the problem posed and define some new future tasks.
11:00 Face Recognition for Identification and Verification in Attendance System: A Systematic Review
Attendance is an important concept to be implemented in order to enhance organizational performance. Measuring attendance may be a concern issue for many organizations, especially with the rapid changes that have occurred in this era of digitalisation. Face recognition is one of many approaches to support long distance attendance, unfortunately, there are some issues that need to be overcome regarding face recognition, for example, its verification. The purpose of this systematic literature review is to provide either academics and practitioner with insight and knowledge related to face recognition implementation, how to verify it, and what is its critical success factors towards an attendance system. Therefore, the results of this study hopes to give both practitioners and academics knowledge on how to build a relevant attendance system based on the findings of this study. The review was conducted through a systematic literature review stages that was adopted from Kitchenham. It began with declaring protocol review and ended with an analysis of the prior studies that was obtained from five relevant sources. There were 22 of 516 studies that met the criteria after several filtering stages. It is found that academic reasons are mostly used on verifying face recognition. Moreover, based on the findings, information security has been considered the most utilized security methods regarding face recognition in attendance system. Furthermore, 78% of the analysed papers stated that security issues is the most critical factors towards implementing a successful attendance system.
11:15 Image Classification for Egg Incubator using Transfer Learning of VGG16 and VGG19
Research in the field of image classification is proliferating and providing benefits to the community. This research focuses on image classification on incubators, incubators in addition to monitoring temperature and humidity. Monitoring of conditions in incubators is also required. This study proposes a classification of eggs, hatching eggs, and chicks. Many images of each object are needed, a total of 3,924 images from all three classes. The data set collected from google image and the collection of images owned by the author obtained from shooting with a smartphone camera on a private farm. Data preprocessing is carried out, such as changing the shape of the image to a square that VGG required 224×224 pixels, augmentation to reproduce data, and sharing training data are the input size 80 and 20 percent validation. After preprocessing the data, the model formation and training process was carried out with the results for each model: Custom CNN model yielded an accuracy of 0.8687, VGG16 produced an accuracy of 0.90, and VGG19 produced an accuracy of 0.92. This study shows that transfer learning has the highest accuracy in image classification.
10:30 Cloud-RAN and Coverage Gap in Rural Areas
This article deals with Cloud Radio Access Network (Cloud-RAN), one of the potential technologies that can help meet the mobile telephony and mobile broadband universal access’ requirements. After presenting Cloud-RAN concepts and the existing literature review, we point out that an open Cloud-RAN network based on openBTS or OpenAirInterface in which dynamic spectrum management is performed could be very beneficial in terms of cost and performance for the extension of mobile broadband networks into rural areas. We propose a performance and cost indicators model to judge the quality of a mobile phone system. Using the indicators, we conduct a comparative study of a basic Cloud-RAN network with a traditional network. Our results have highlighted the Cloud-RAN potential to ensure network profitability and thus viability, which is a prerequisite for commercial network operators.
10:45 Real-time Signboards using Panels P5 RGB and NodeMCU esp8266 with Library PxMatrix
The causes of traffic congestion are divided into two, one of which is natural factors and the other is human factors. One way to resolve this is by diverting the flow. Usually, the diversion is carried out by the police. The other way to divert the flow is by using a signboard that can change its direction. This board is made using the P5 RGB led panel and the NodeMCU esp8266 microcontroller. This panel uses the Internet to communicate with the database. The database used is the Firebase that contains the name of the city and the direction to the city. It used the HTML Protocol to send the data from the database to the board. From the tests that have been done, the results obtained shows that NodeMCU esp8266 is not suitable to be used by the PxMatrix Library for large panel configurations. But on a small panel, the test was successfully carried out. The test was carried out by changing all directions of the city. The result of this test is that the direction of the entire city can be changed according to changes in the Firebase.
11:00 Prototype Smart Door Lock By Using Wireless Network Based on Arduino Uno
Most of the activities carried out all day are done outside the home. This condition happens in every major city. Therefore, it is ascertained that the occupants must leave many empty houses for working hours, even more so for the holiday season for Eid, Christmas, or New Year. These empty houses were targeted by thieves, especially those without a flawed security system. In this final project, the authors make a smart home prototype that protects the house from strangers. This final project makes an automatic house locking device using FPM10A Fingerprint components, HC-SR501 PIR Sensor, Arduino UNO, and Solenoid Door lock. In this final project, a Telegram application will be added, which will be connected through the Arduino IDE software, which functions to notify when the door of the house/room is open.
11:15 SOLSR Protocol Performance Analysis For Voip Application In Mesh Topology
Indonesia has a high level of potential for natural disasters that can cause damage to infrastructure; on the other hand, communication is critical when a natural disaster occurs, which causes communication to be hampered. The solution to dealing with this problem is building Ad-hoc Network (Manet) technology for communication media. By building alternative VoIP-based communication on Raspberry Pi. The protocol used in VOIP (Voice Over Internet Protocol) based communication using Raspberry Pi is the SOLSR (Secure Optimized Link State Routing) protocol. The VoIP work system is voice passing through the internet network in the VoIP system workflow; two components are connected, the VoIP client and VoIP server, VoIP server as the center. On the VoIP, the server uses Raspberry Pi, and the client also uses Raspberry Pi and is registered via SIP to the server to communicate to other clients and test its performance utilizing QOS testing using SOLSR. The test scenario is done by measuring throughput, delay, jitter, packet loss. The measurement results will produce the performance of a VoIP-based communication system using the SOLSR protocol and determine the feasibility of the SOLSR protocol to be applied to a VoIP-based communication system for disaster management.
10:30 Development and Characterization of Narrowband BPF Made of Substrate Integrated Waveguide
The utilization of substrate integrated waveguide (SIW) structure has penetrated into numerous applications for radio frequency (RF) devices. This paper discusses the development of narrowband bandpass filter (BPF) made of SIW structure for 5 GHz band wireless communication application. The proposed filter is design on a Duroid 5880 dielectric substrate with the thickness of 1.575 mm and the total dimension of 36.5 mm (length) and 19 mm (width). The characterization is carried out through simulation software to investigate and analyze the effect of different filter parameters to the output response of BPF. Meanwhile, parametric studies as part of characterization process are applied on the proposed BPF to obtain the optimum design of filter architecture. The results of characterization shows that the proposed narrowband SIW BPF with the input/output impedances of 50 Ohm has the bandwidth characteristic of 84 MHz in the frequency range of 5.827 GHz to 5.911 GHz with a good frequency selectivity suitable for the desired application.
10:45 Design of a Compact Circular Patch Antenna Operating at ISM-Band for the WiMAX Communication Systems
In recent times, people’s dependency on high-speed internet has increased to a significant extent through the means of Wireless Communication Systems. However, in developing and underdeveloped nations, it is a common scene that rural areas do not receive proper internet connectivity due to their price ranges and implantation costs. Considering this issue, a Circular Patch Antenna has been designed in this Research work for WiMAX Communication Systems using FR-4 Substrate Material with an Operating Frequency of 2.45 GHz (ISM-Band). The whole Simulation of the Antenna was carried out in the CST Studio Suite 2019 Software. Pure Copper Material was used to create the Ground Plane and the Circular shape Patch of the Antenna. The simulated Patch Antenna Model has an Architectural Dimensions of 20.66*26.76 mm2, which has been thoroughly compared with the past Research works to justify the validity of the work. Some of the Parameters that are obtained from the Simulation include- Return Loss (S1,1) of -37.198 dB, VSWR of 1.0282, Surface Current of 175 A/m, and other vital parameters were also successfully determined, which proves that the Antenna can be a better choice for WiMAX Communication Systems.
11:00 Design and Characterization of ADM-based Dual-band SIW Bandpass Filter
This paper deals with the design and characterization of dual-band substrate-integrated waveguide (SIW) bandpass filter (BPF) developed based on artificial dielectric material (ADM). The SIW structure is implemented using dielectric substrates flanked by metal layers on the top and bottom sides whereby these metal layers are connected by array of conductor wires called as vias which resemble sidewalls of a waveguide. While the ADM structure is composed of a dielectric substrate incorporated by array of conductor wires in a specific configuration. Both SIW and ADM structures are designed and developed using an FR4 Epoxy dielectric substrate with the relative permittivity of 4.4. Parametric studies are applied upon both structures to obtain an optimum performance of the filter. The characterization result demonstrates that a dual-band SIW BPF configured by a two-stage ADM in face-centered orthorombic configuration could achieve the bandwidth responses of 150 MHz and 215 MHz with the center frequencies of 1.55 GHz and 2.375 GHz, respectively.
11:15 A Design of Extreme Learning Machine Based Receiver for 2×2 MIMO-OFDM System
Machine learning has raised the research interest for future communication network. The implementation of machine learning technologies such as Deep Neural Network (DNN), Support Vector Machine (SVM) and Extreme Learning Machine (ELM) have enabled the derivation of channel estimation techniques which provide more accurate estimation in comparison with accuracy provided by other well-known estimation techniques. ELM has faster learning ability compared to that of the other aforementioned technologies. This paper focuses on the ELM based receiver design for 2×2 MIMO OFDM system. The determination of the output layer neuron, a component in ELM technique, on the basis of symbol type (number of bits per symbol) has been proposed. The performance of the proposed receiver is evaluated in terms of Bit Error Rate (BER) and compared with that of the Minimum Mean Square Error (MMSE) based receiver. Improved performance has been achieved by means of the proposed scheme.
11:30 Secret and Trustable Communication Channel over Blockchain Public Ledger
Blockchain public ledger is a trusted medium that provides incorruptible data storage and traceable transactions. However, due to its public verifiability, it is hard to compromise two functionalities at once, a covert channel and a verifiable communication medium. The existing methods are either vulnerable to statistical analysis, or lack of delivery notifications due to off-chain message delivery. This paper fills the gap by proposing a Multi-addresses Random Set Encoding (MaRSE) to increase message covertness by retaining the natural transactions pattern and resisting statistical attacks. The method uses on-chain message delivery to preserve blockchain-based verification that provides a trustable communication medium for the communicating parties.
As a result, our proposed method is robust to a brute force attack due to its unpredictable transaction sequence and high number possibility of address combinations.
It also scales to message capacity since a bigger message requires a larger number of addresses that in return generate a higher set variation.