Journal of ICT Research and Applications
Not a member yet
359 research outputs found
Sort by
Improving Robustness Using MixUp and CutMix Augmentation for Corn Leaf Diseases Classification based on ConvMixer Architecture
Get PDFCorn leaf diseases such as blight spot, gray leaf spot, and common rust still lurk in corn fields. This problem must be solved to help corn farmers. The ConvMixer model, consisting of a patch embedding layer, is a new model with a simple structure. When training a model with ConvMixer, improvisation is an important part that needs to be further explored to achieve better accuracy. By using advanced data augmentation techniques such as MixUp and CutMix, the robustness of ConvMixer model can be well achieved for corn leaf diseases classification. We describe experimental evidence in this article using precision, recall, accuracy score, and F1 score as performance metrics. As a result, it turned out that the training model with the data set without extension on the ConvMixer model achieved an accuracy of 0.9812, but this could still be improved. In fact, when we used the MixUp and CutMix augmentation, the training model results increased significantly to 0.9925 and 0.9932, respectively
Quality Analysis of Telemetry Tracking and Command at Ground Stations using the Association Rule Mining Approach
Get PDFLAPAN built several remote ground stations to support the telemetry tracking and command (TTC) system for the LAPAN-A2 and LAPAN-A3 satellites. These remote ground stations are located in Kototabang/KT (West Sumatra), Biak/BK (Papua), Parepare/PR (South Sulawesi), Rumpin/RP, Rancabungur/RB (Bogor, West Java), and Svalbard/SV (Norway). Problems that often arise in the TTC process are telecommands not being sent (commands sent from the ground station to the satellite) or telemetry packages not being received (feedback on telecommands sent by the satellite to the ground station). This research attempted to calculate and analyze the quality of TTC using a data-mining approach, i.e., rule mining. The calculations were performed using five main parameters: satellite name, ground station, azimuth, altitude, and communication status. The research output consisted of a combination of remote ground station parameters that may result in a successful or failed TTC. For the LAPAN-A3 satellite at the Svalbard ground station, 19 failed communication combinations were generated with a dataset of 57,029. Communication failures occur in azimuth and elevation, i.e., areas blocked by obstacles
Scene Segmentation for Interframe Forgery Identification
Get PDFA common type of video forgery is inter-frame forgery, which occurs in the temporal domain, such as frame duplication, frame insertion, and frame deletion. Some existing methods are not effective to detect forgeries in static scenes. This work proposes static and dynamic scene segmentation and performs forgery detection for each scene. Scene segmentation is performed for outlier detection based on changes of optical flow. Various similarity checks are performed to find the correlation for each frame. The experimental results showed that the proposed method is effective in identifying forgeries in various scenes, especially static scenes, compared with existing methods
CNN Based Covid-19 Detection from Image Processing
Get PDFCovid-19 is a respirational condition that looks much like pneumonia. It is highly contagious and has many variants with different symptoms. Covid-19 poses the challenge of discovering new testing and detection methods in biomedical science. X-ray images and CT scans provide high-quality and information-rich images. These images can be processed with a convolutional neural network (CNN) to detect diseases such as Covid-19 in the pulmonary system with high accuracy. Deep learning applied to X-ray images can help to develop methods to identify Covid-19 infection. Based on the research problem, this study defined the outcome as reducing the energy costs and expenses of detecting Covid-19 in X-ray images. Analysis of the results was done by comparing a CNN model with a DenseNet model, where the first achieved more accurate performance than the second
A Decoupling Technique for Beamforming Antenna Arrays Using Simple Guard Trace Structures
Get PDFThis paper discusses decoupling techniques for suppressing electromagnetic coupling between elements of beamforming antenna arrays. Guard trace structures, which are commonly used for crosstalk reduction on printed circuit board technology, are proposed to be inserted between the array elements for coupling reduction. Two types of guard trace structures, i.e., straight guard traces and serpentine guard traces, were explored, and the effect of using via holes on both types of guard traces was studied. For this purpose, two-element antenna arrays with guard trace structures inserted between array elements were designed and simulated. The simulation results showed that a straight guard trace with vias (straight GTV) and a serpentine guard trace without vias (serpentine GT) could effectively reduce EM coupling between elements of array antennas. To verify the simulation results, prototypes of antenna arrays with straight GTV and serpentine GT were realized and measured. The measurement results showed coupling reductions of 5 dB and 6.4 dB could be achieved when straight GTV and serpentine GT are inserted between two array elements separated by edge-to-edge distances of 4 mm and 9.05 mm, respectively. Therefore, the proposed decoupling technique is suitable for beamforming antenna arrays with a very close distance between array elements
Leveraging Data Management Capabilities for Innovation Capabilities: The Moderating Role of Cross-Functional Integration
Get PDFIn today’s dynamic and competitive business environment, data are crucial for sustaining a competitive advantage. Organizations are also constantly seeking ways to enhance their innovation capabilities in order to stay ahead of the competition. One critical factor that has been identified as influential in enabling innovation are the organization’s data management capabilities. Past studies have found that cross-functional integration may enhance the impact of data management on innovation. Hence, this study aimed to investigate the influence of data management capabilities on explorative and exploitative innovation by considering the role of cross-functional integration as a moderating variable. This study used 116 data samples from medium and large companies across different industries in Indonesia. The PLS-SEM analysis was applied to test the research hypotheses. The results indicate that data management capabilities as a third-order construct, consisting of three dimensions, namely data governance, technology, and skills, have significant direct influences on explorative and exploitative innovation. This study demonstrated that cross-functional integration still plays an important role in amplifying the relationship between data management capabilities and innovation capabilities, especially in relation to explorative innovation
Smart Card-based Access Control System using Isolated Many-to-Many Authentication Scheme for Electric Vehicle Charging Stations
Get PDFIn recent years, the Internet of Things (IoT) trend has been adopted very quickly. The rapid growth of IoT has increased the need for physical access control systems (ACS) for IoT devices, especially for IoT devices containing confidential data or other potential security risks. This research focused on many-to-many ACS, a type of ACS in which many resource-owners and resource-users are involved in the same system. This type of system is advantageous in that the user can conveniently access resources from different resource-owners using the same system. However, such a system may create a situation where parties involved in the system have their data leaked because of the large number of parties involved in the system. Therefore, ‘isolation’ of the parties involved is needed. This research simulated the use of smart cards to access electric vehicle (EV) charging stations that implement an isolated many-to-many authentication scheme. Two ESP8266 MCUs, one RC522 RFID reader, and an LED represented an EV charging station. Each institute used a Raspberry Pi Zero W as the web and database server. This research also used VPN and HTTPS protocols to isolate each institute’s assets. Every component of the system was successfully implemented and tested functionally
An Efficient Intrusion Detection System to Combat Cyber Threats using a Deep Neural Network Model
Get PDFThe proliferation of Internet of Things (IoT) solutions has led to a significant increase in cyber-attacks targeting IoT networks. Securing networks and especially wireless IoT networks against these attacks has become a crucial but challenging task for organizations. Therefore, ensuring the security of wireless IoT networks is of the utmost importance in today’s world. Among various solutions for detecting intruders, there is a growing demand for more effective techniques. This paper introduces a network intrusion detection system (NIDS) based on a deep neural network that utilizes network data features selected through the bagging and boosting methods. The presented NIDS implements both binary and multiclass attack detection models and was evaluated using the KDDCUP 99 and CICDDoS datasets. The experimental results demonstrated that the presented NIDS achieved an impressive accuracy rate of 99.4% while using a minimal number of features. This high level of accuracy makes the presented IDS a valuable tool
Sentiment Classification for Film Reviews in Gujarati Text Using Machine Learning and Sentiment Lexicons
Get PDFIn this paper, two techniques for sentiment classification are proposed: Gujarati Lexicon Sentiment Analysis (GLSA) and Gujarati Machine Learning Sentiment Analysis (GMLSA) for sentiment classification of Gujarati text film reviews. Five different datasets were produced to validate the machine learning-based and lexicon-based methods’ accuracy. The lexicon-based approach employs a sentiment lexicon known as GujSentiWordNet, which identifies sentiments with a sentiment score for feature generation, while in the machine learning-based approach, five classifiers are used: logistic regression (LR), random forest (RF), k-nearest neighbors (KNN), support vector machine (SVM), naive Bayes (NB) with TF-IDF, and count vectorizer for feature selection. Experiments were carried out and the results obtained were compared using accuracy, precision, recall, and F-score as performance evaluation criteria. According to the test results, the machine learning-based technique improved accuracy by 3 to 10% on average when compared to the lexicon-based approach
Generative Adversarial Networks Based Scene Generation on Indian Driving Dataset
Get PDFThe rate of advancement in the field of artificial intelligence (AI) has drastically increased over the past twenty years or so. From AI models that can classify every object in an image to realistic chatbots, the signs of progress can be found in all fields. This work focused on tackling a relatively new problem in the current scenario-generative capabilities of AI. While the classification and prediction models have matured and entered the mass market across the globe, generation through AI is still in its initial stages. Generative tasks consist of an AI model learning the features of a given input and using these learned values to generate completely new output values that were not originally part of the input dataset. The most common input type given to generative models are images. The most popular architectures for generative models are autoencoders and generative adversarial networks (GANs). Our study aimed to use GANs to generate realistic images from a purely semantic representation of a scene. While our model can be used on any kind of scene, we used the Indian Driving Dataset to train our model. Through this work, we could arrive at answers to the following questions: (1) the scope of GANs in interpreting and understanding textures and variables in complex scenes; (2) the application of such a model in the field of gaming and virtual reality; (3) the possible impact of generating realistic deep fakes on society