TELKOMNIKA (Telecommunication Computing Electronics and Control)
Not a member yet
3120 research outputs found
Sort by
Smart hydroponic agriculture using genetic algorithm based k-nearest neighbors
Full text linkIn this research, researcher has implemented supervised machine learning, namely k-nearest neighbor (k-NN) which is optimized using genetic algorithms, and the internet of things (IoT) on the nutrient film technique (NFT) hydroponic system. The aim of this research is to improve the accuracy of classification of nutrient and light conditions in NFT system, and evaluating the harvest of hydroponic farming. The dataset was obtained by observing and recording nutritional and light conditions using sensors for 35 days during the growing period of lettuce in the NFT system, thus obtaining 1,680 data. Then, a training dataset is created based on that dataset. The system architecture is divided into 3 parts, namely the sensor system, data processing, and actuator system. The conclusion of this research is the IoT can be used to monitor the nutritional and light conditions of NFT system in real time and automatic control actions can be carried out using actuators controlled by the Raspberry Pi, the impact of applying the k-NN algorithm and the genetic algorithms is the accuracy of classifying nutritional and light conditions is 92%, the lettuce in a NFT system controlled by the system grow better than the lettuce in a NFT system controlled manually
An approach for liver cancer detection from histopathology images using hybrid pre-trained models
Full text linkHistopathological image analysis (HIA) plays an essential role in detecting cancer cell development, but it is time-consuming, prone to inaccuracy, and dependent on pathologist competence. This paper proposes an automated HIA that uses deep learning to improve accuracy and efficiency in liver cancer cell growth. The model uses whole slide image (WSI) input, open computer vision (OpenCV) libraries for image preprocessing, ResNet50 for patch-level feature extraction, and multiple instances learning for image-level classification. The suggested approach accurately distinguishes liver histopathological pictures as cancerous or non-cancerous. Assisting in the early detection of liver cancer cell development with potential invasion or spread
Tomato leaf disease recognition system using Faster R-CNN
Full text linkThe objective of this paper is to detect tomato leaf disease using Faster region-based convolutional neural network (R-CNN). The tomato leaf disease recognition system utilizes a dataset consisting of healthy tomato leaves and eight leaf diseases, including early blight, late blight, leaf mold, mosaic virus, septoria, spider mites, yellow leaf curl virus, and leaf miner. The dataset is obtained from various sources, such as Kaggle, Google Images, Bing Images, and Roboflow Universe. Pre-processing techniques, including collage, tile, static crop, and resize, are applied to prepare the dataset for training. Data augmentation methods, such as flipping, 90° rotation, exposure adjustment, and hue modification, are applied to enhance the model’s robustness and generalize its performance. Specifically, we implemented Faster R-CNN as part of Detectron2 using its base models and configurations. The results demonstrate that the X101-FPN base model for Faster R-CNN with the default configurations of Detectron2 is efficient and general enough to be applied to defect detection. This approach results in an average precision (AP) detection score of 87.01% for validation results
Increasing the perovskite cell performance using comparative layering method between PTAA and PEDOT: PSS layers
Full text linkToday, perovskite solar cells are introduced and used as a suitable alternative with high efficiency for silicon solar cells. The main problem of this type of cells until now has been mostly about their instability, because these materials that are used in perovskite solar cells are rapidly destroyed in reaction with air or their efficiency is greatly reduced. In this article, we increase the productivity to an acceptable amount and also increase its stability by using new high-quality synthesized materials and also by changing the manufacturing method of perovskite solar cells. Using the hole transport layers (HTLs), in the inverse planar perovskite solar cell structure, the effect of two layers of poly (triarylamine) (PTAA) and poly (3,4-ethylenedioxythiophene) (PEDOT): PSS as the bottom layer of the perovskite film on the morphology of the nanoparticles, the crystallinity of the perovskite layer and the photovoltaic parameters affecting the efficiency of the solar cell made in a single-step method were investigated. The result shows the PTAA layer has been very effective in controlling the morphology of the perovskite layer, so the efficiency reached to 23.9% while the maximum efficiency per solar cell based on PEDOT: PSS is 11.37%
Convolutional neural network enhancement for mobile application of offline handwritten signature verification
Full text linkThe increase in signature forgery cases can be attributed to the escape of forged signatures from manual signature verification systems. Researchers have developed various machine learning and deep learning methods to verify the authenticity of signatures, one of which uses convolutional neural networks (CNNs). This research aims to develop a mobile application for handwritten signature verification using CNN architecture by adding a batch normalization technique to its layer. The performance of our proposed method achieved a verification accuracy of 86.36%, with a 0.061 false acceptance rate (FAR), 0.303 false rejection rate (FRR), and 0.182 equal error rate (EER), which is compatible to be embedded in smartphones. However, there is still a need for further development of the CNN model and its integration with mobile applications
Synthesis of reduced graphene oxide decotate Cu2S nanoparticles for cathode of quantum dot solar cell
Full text linkIn this paper, the results of making a reduced graphene oxide cathode electrode with Cu2S nanoparticles are shown so that it can be used as a counter electrode in quantum dot solar cells to replace other counter electrodes. An rGO-Cu2S paste obtained by hydrolysis was scanned onto the surface of the fluorine-doped tin oxide (FTO) conductive substrate when bound to Cu2S nano by a screen-printing process, then calcined at 350 °C to crystallize the film. Following calcination, the film was examined for structure using energy-dispersive X-ray (EDX) and X-ray diffraction (XRD) spectroscopy, as well as for type and particle size using scanning and transmission electron microscopy and transmission electron microscopy, respectively. Mott-schottky measurement is used to determine the semiconductor and carrier concentrations in the film, and an electrochemical device is used to assess the electrodes redox capacity in a polysulfide electrolyte solution. The operability of the rGO-Cu2S cathode at the peak of the current density in the C-V curve was 24 mA/cm2, a 30-fold increase compared to that of the Cu2S electrode. This result shows that the efficiency, Voc, FF, Jsc are 4.92%, 0.525 V, 0.418, and 22.4 mA/cm2, respectively
A good result of brain tumor classification based on simple convolutional neural network architecture
Full text linkBrain tumor disease has become a topic of research whether it is in the case of segmentation or classification. For the case of classification, the types of brain tumors that are grouped generally consist of high-grade glioma (HGG) and low-grade glioma (LGG) tumors. In this research we are doing, we propose a method for classifying 2 types of tumors, namely HGG and LGG, using the convolutional neural network (CNN) algorithm which is trained and will be tested against the 2018 and 2019 brain tumor segmentation (BRATS) datasets which have 4 modalities, namely fluid-attenuated inversion recovery (FLAIR), T1, T1ce, and T2 totaling 2048 images. The CNN algorithm was chosen because it can directly receive input in the form of a magnetic resonance image (MRI) with the feature extraction process as well as the classification algorithm. By forming a simple CNN algorithm architecture with only 3 convolutional layers which have an input layer in the form of a full MRI image with dimensions of 240×240×3, we obtained a relatively high accuracy result of 94.14%, it can even be said to be better than similar methods but with more complicated architecture
The game model of investing in the academic cloud
Full text linkThe article analyzes approaches to the use of cloud technologies in the process of teaching students at large universities. The model of the academic cloud of a modern university is considered. Examples of software and functional platforms that meet the needs of students in electronic learning resources are given. The deployment models of the cloud-oriented educational environment that includes private cloud infrastructure as a service (IaaS) and platform as a university service are analyzed. The cost of deploying an academic cloud based on the educational institution’s infrastructure and renting infrastructure from a vendor is compared. A multifactorial model for evaluating investment options in the university cloud in the context of fuzzy information is proposed. In contrast to the known approaches to solving such a problem, our model assumes that the dynamics of the financial states of the players are set through a system of discrete equations. These equations describe the dynamics of multidimensional variables. The latter made it possible to consider the general problem of investing in the academic cloud within the framework of a game scheme for tasks in a fuzzy formulation, with the financial resources available to the educational institution. Preference sets and optimal financial allocation strategies for building an academic cloud are found
Quantum binary particle swarm optimization for optimal on-load tap changing and power loss reduction
Full text linkOver time, there has been a continuous surge in the demand for electrical energy, necessitating the development of larger and more intricate electrical power networks. These extensive networks pose a significant challenge, primarily in the form of considerable loss of electrical energy, which, if not effectively addressed, may lead to persistent and imperceptible losses. In response to this challenge, this research proposes the application of quantum binary particle swarm optimization (QBPSO) for the coordinated management of on-load tap changers (OLTC) in loaded transformers within a distribution network, with a specific emphasis on reducing power losses. The experimental results demonstrate that the implementation of QBPSO results in a reduction of power loss from 21.756107 kW to 19.157321 kW and an increase in the average voltage from 19.00467941 kV to 19.93068 kV in a 20 kV 34-bus distribution network. This has the potential to significantly enhance overall system efficiency
Simple RNN-LSTM hybrid deep learning model for Bitcoin and EUR_USD forecasting
Full text linkThe popularity of deep learning in time series prediction has significantly increased compared to the past. In this article, we utilize deep learning methods, which encompass long short term memory (LSTM) networks, simple recurrent neural network (SimpleRNN) networks, and gated recurrent units (GRU) networks. This research introduces a hybrid foundational model for forecasting future closing prices of EUR_USD in financial time series and Bitcoin, combining SimpleRNN with LSTM, referred to as SimpleRNN-LSTM. To improve the precisions of our hybrid model, we incorporate twenty-one technical indicators into the training data. Then, we compute four measures to evaluate the performance of various prediction models. When predicting currency pairs EUR_USD and Bitcoin, our hybrid foundational model outperforms SimpleRNN, LSTM, and GRU models