Tegar Arifin Prasetyo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Tegar Arifin Prasetyo has a total of 37 co-authors. Here are the top co-authors by number of shared publications:
- Rudy Chandra
Institut Teknologi Del
8 shared publications - Dewi Handayani
Institut Teknologi Bandung
2 shared publications - Roberd Saragih
Institut Teknologi Bandung
2 shared publications - Wesly Mailander Siagian
Institut Teknologi Del
2 shared publications - Asido Saragih
Institut Teknologi Del, Laguboti
1 shared publication Cristian Nicolas Tambun
1 shared publication
Latest Publications
- Open Access
- 31 May 2026
- Published by Society of Visual Informatics in JOIV : International Journal on Informatics Visualization
- Vol. 10 (3) , 1269
Optimized CNN-NDRE Model Development for Detecting Diseased Regions in Potato Fields Using Drone Imagery
Potato (Solanum tuberosum L.) is a key agricultural crop that requires continuous monitoring to maintain productivity. This study evaluates potato plant health using drone-captured aerial imagery. A total of 3,040 images were collected from field surveys and TSTH2 datasets, acquired with multiple spectral sensors to ensure data variation. The dataset was manually labeled, augmented, and split into training, validation, and testing sets. Health classification was conducted using the NDRE (Normalized Difference Red Edge) vegetation index, which distinguishes healthy from diseased areas based on spectral differences. A Convolutional Neural Network (CNN) with the VGG16 architecture was employed, trained for a pre-trained and fine-tuned model. In the pretraining stage, all convolutional layers of VGG16 are frozen to preserve the pretrained ImageNet weights, while only the newly added classification layers are trained using a higher learning rate and data augmentation. In the fine-tuning stage, several of the final layers are unfrozen, allowing the model to adjust high-level feature representations to better fit the target dataset domain. Fine-tuning involved unlocking the final four layers, resulting in a validation accuracy of 97% with stable performance and no overfitting. Additional segmentation using NIR–Red reflectance, optimized Otsu thresholding, and K-Means clustering effectively separated potato plants from surrounding vegetation, with a threshold of -0.05 yielding accurate, low-noise results consistent with field observations. Finally, a web-based interface was developed to classify potato fields as healthy or diseased, enabling more efficient, technology-driven land monitoring.
- 31 May 2026
- Published by Society of Visual Informatics in JOIV : International Journal on Informatics Visualization
- Vol. 10 (3) , 1269
Open AccessOptimized CNN-NDRE Model Development for Detecting Diseased Regions in Potato Fields Using Drone Imagery
Potato (Solanum tuberosum L.) is a key agricultural crop that requires continuous monitoring to maintain productivity. This study evaluates potato plant health using drone-captured aerial imagery. A total of 3,040 images were collected from field surveys and TSTH2 datasets, acquired with multiple spectral sensors to ensure data variation. The dataset was manually labeled, augmented, and split into training, validation, and testing sets. Health classification was conducted using the NDRE (Normalized Difference Red Edge) vegetation index, which distinguishes healthy from diseased areas based on spectral differences. A Convolutional Neural Network (CNN) with the VGG16 architecture was employed, trained for a pre-trained and fine-tuned model. In the pretraining stage, all convolutional layers of VGG16 are frozen to preserve the pretrained ImageNet weights, while only the newly added classification layers are trained using a higher learning rate and data augmentation. In the fine-tuning stage, several of the final layers are unfrozen, allowing the model to adjust high-level feature representations to better fit the target dataset domain. Fine-tuning involved unlocking the final four layers, resulting in a validation accuracy of 97% with stable performance and no overfitting. Additional segmentation using NIR–Red reflectance, optimized Otsu thresholding, and K-Means clustering effectively separated potato plants from surrounding vegetation, with a threshold of -0.05 yielding accurate, low-noise results consistent with field observations. Finally, a web-based interface was developed to classify potato fields as healthy or diseased, enabling more efficient, technology-driven land monitoring.
- Open Access
- 1 December 2025
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 15 (6) , 5543-5554
Optimizing parameter selection in bidirectional encoder portrayal for transformers algorithm using particle swarm optimization for artificial intelligence generate essay detection
This research proposes a novel method for detecting artificial intelligence (AI)-generated essays by integrating the bidirectional encoder representations from transformers (BERT) model with particle swarm optimization (PSO). Unlike traditional approaches that rely on manual hyperparameter tuning, this study introduces a systematic optimization technique using PSO to improve BERT’s performance in identifying AI-generated content. The key problem addressed is the lack of effective, real-time detection systems that preserve academic integrity amidst rapid AI advancements. This optimization enhances the model’s detection accuracy and operational efficiency. The research dataset consisted of 46,246 essays, which, after data cleaning, were refined to 44,868. The model was then tested on 9,250 essays. Initial evaluations showed BERT's accuracy ranging from 83% to 94%. After being optimized with PSO, the model achieved an accuracy of 98%, an F1-score of 98.31%, precision of 97.75%, and recall of 98.87%. The model was deployed using a FastAPI-based web interface, enabling real-time detection and providing users with an efficient way to quickly verify text authenticity. This research contributes a scalable, automated solution for AI-generated text detection and offers promising implications for its application in various academic and digital content verification contexts.
- 1 December 2025
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 15 (6) , 5543-5554
Open AccessOptimizing parameter selection in bidirectional encoder portrayal for transformers algorithm using particle swarm optimization for artificial intelligence generate essay detection
This research proposes a novel method for detecting artificial intelligence (AI)-generated essays by integrating the bidirectional encoder representations from transformers (BERT) model with particle swarm optimization (PSO). Unlike traditional approaches that rely on manual hyperparameter tuning, this study introduces a systematic optimization technique using PSO to improve BERT’s performance in identifying AI-generated content. The key problem addressed is the lack of effective, real-time detection systems that preserve academic integrity amidst rapid AI advancements. This optimization enhances the model’s detection accuracy and operational efficiency. The research dataset consisted of 46,246 essays, which, after data cleaning, were refined to 44,868. The model was then tested on 9,250 essays. Initial evaluations showed BERT's accuracy ranging from 83% to 94%. After being optimized with PSO, the model achieved an accuracy of 98%, an F1-score of 98.31%, precision of 97.75%, and recall of 98.87%. The model was deployed using a FastAPI-based web interface, enabling real-time detection and providing users with an efficient way to quickly verify text authenticity. This research contributes a scalable, automated solution for AI-generated text detection and offers promising implications for its application in various academic and digital content verification contexts.
- Open Access
- 1 May 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (2) , 1204-1213
Refining tomato disease recognition: hyperparameter tuning on ResNet-101 architecture for precise leaf-based classification
Tomatoes plants are widely recognized as versatile vegetables globally. This study aims to develop a high-precision web interface for classifying various leaf diseases in tomatoes. Utilizing a convolutional neural network (CNN) algorithm using the residual network-101 (ResNet-101) architecture, this tool aids farmers in accurately identifying leaf diseases in tomatoes, thereby reducing the risk of crop failure. The dataset comprises 6,800 images, categorized into four classes: early blight, spider mites two spotted, tomato yellow leaf curl virus, and healthy tomatoes, each containing 1,700 images. Hyperparameter tuning was conducted as part of an experiment aimed at enhancing the performance of the model. Experiments involved varying epoch values (10, 25, 30, 50, 60, 75, 100, and 110), a fixed batch size of 4, different learning rates (0.1, 0.01, 0.001, 0.0001), and num workers (4, 8, 16). The results demonstrated an accuracy of 99% with 100 epochs, a batch size of 4, a learning rate of 0.001, and 16 num workers. Consequently, this research contributes to a deeper understanding of disease management in tomato plants, ensuring optimal quality and quantity of the harvest.
- 1 May 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (2) , 1204-1213
Open AccessRefining tomato disease recognition: hyperparameter tuning on ResNet-101 architecture for precise leaf-based classification
Tomatoes plants are widely recognized as versatile vegetables globally. This study aims to develop a high-precision web interface for classifying various leaf diseases in tomatoes. Utilizing a convolutional neural network (CNN) algorithm using the residual network-101 (ResNet-101) architecture, this tool aids farmers in accurately identifying leaf diseases in tomatoes, thereby reducing the risk of crop failure. The dataset comprises 6,800 images, categorized into four classes: early blight, spider mites two spotted, tomato yellow leaf curl virus, and healthy tomatoes, each containing 1,700 images. Hyperparameter tuning was conducted as part of an experiment aimed at enhancing the performance of the model. Experiments involved varying epoch values (10, 25, 30, 50, 60, 75, 100, and 110), a fixed batch size of 4, different learning rates (0.1, 0.01, 0.001, 0.0001), and num workers (4, 8, 16). The results demonstrated an accuracy of 99% with 100 epochs, a batch size of 4, a learning rate of 0.001, and 16 num workers. Consequently, this research contributes to a deeper understanding of disease management in tomato plants, ensuring optimal quality and quantity of the harvest.
- Open Access
- 1 May 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (2) , 1353-1366
Evaluating the efficacy of univariate LSTM approach for COVID-19 data prediction in Indonesia
The coronavirus disease 2019 (COVID-19) pandemic, originating in 2020, has emerged as a critical global issue due to its rapid and widespread transmission. Indonesia, among the affected nations, has taken measures to address the situation, including the development of a deep learning model for predicting future COVID-19 infection and spread. This predictive tool serves as a valuable reference for the government and stakeholders, aiding them in making informed decisions and implementing appropriate measures to contain the virus. The deep learning model employs the long short-term memory (LSTM) algorithm, chosen for its ability to recognize temporal patterns in the country’s COVID-19 data. The model creation process involves data collection, preprocessing, model architecture planning, modeling, training, and evaluation. Two LSTM models were developed: a univariate and a multivariate model. Following thorough training and evaluation, the univariate model emerged as the superior choice, boasting evaluation metrics of 16.72 for mean absolute percentage error (MAPE) and 66.36 for root mean squared error (RMSE). This model was then deployed on a publicly accessible website, presenting visualizations of past COVID-19 data and predictions of future cases through line graphs. This user-friendly platform enables the public to access and analyze the data easily.
- 1 May 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 34 (2) , 1353-1366
Open AccessEvaluating the efficacy of univariate LSTM approach for COVID-19 data prediction in Indonesia
The coronavirus disease 2019 (COVID-19) pandemic, originating in 2020, has emerged as a critical global issue due to its rapid and widespread transmission. Indonesia, among the affected nations, has taken measures to address the situation, including the development of a deep learning model for predicting future COVID-19 infection and spread. This predictive tool serves as a valuable reference for the government and stakeholders, aiding them in making informed decisions and implementing appropriate measures to contain the virus. The deep learning model employs the long short-term memory (LSTM) algorithm, chosen for its ability to recognize temporal patterns in the country’s COVID-19 data. The model creation process involves data collection, preprocessing, model architecture planning, modeling, training, and evaluation. Two LSTM models were developed: a univariate and a multivariate model. Following thorough training and evaluation, the univariate model emerged as the superior choice, boasting evaluation metrics of 16.72 for mean absolute percentage error (MAPE) and 66.36 for root mean squared error (RMSE). This model was then deployed on a publicly accessible website, presenting visualizations of past COVID-19 data and predictions of future cases through line graphs. This user-friendly platform enables the public to access and analyze the data easily.
- Open Access
- 23 April 2024
- Published by Sunan Gunung Djati State Islamic University of Bandung in Jurnal Online Informatika
- Vol. 9 (1) , 111-118
Development of a Mobile-Based Application for Classifying Caladium Plants Using the CNN Algorithm
Caladium is a popular ornamental plant and has business potential. However, difficulties in recognizing the type of Caladium often occur because of the similarities in shape, pattern, and color of the leaves between the different kinds of Caladium. To overcome this problem, research will use machine learning with the Convolutional Neural Network (CNN) algorithm to build a mobile application that can accurately classify four types of Caladiums. The data set used is 1200 data with four classes; each class has 300 data. The best model is found with the parameter epoch 100, learning rate 0.001, and batch size 64. The model is then implemented in a mobile application with two menus, "Take a photo" and "Choose an image," which will display the classification output and confidence values of the four types of Caladiums. Testing with 30 test data per class achieves 0.975 accuracy on both menus. On the “Take a photo” menu, precision is 0.974, recall is 0.9725, and f1-score is 0.965. Meanwhile, on the “Choose an image” menu a precision and recall value is 0.975, and f1-score value of 0.97.
- 23 April 2024
- Published by Sunan Gunung Djati State Islamic University of Bandung in Jurnal Online Informatika
- Vol. 9 (1) , 111-118
Open AccessDevelopment of a Mobile-Based Application for Classifying Caladium Plants Using the CNN Algorithm
Caladium is a popular ornamental plant and has business potential. However, difficulties in recognizing the type of Caladium often occur because of the similarities in shape, pattern, and color of the leaves between the different kinds of Caladium. To overcome this problem, research will use machine learning with the Convolutional Neural Network (CNN) algorithm to build a mobile application that can accurately classify four types of Caladiums. The data set used is 1200 data with four classes; each class has 300 data. The best model is found with the parameter epoch 100, learning rate 0.001, and batch size 64. The model is then implemented in a mobile application with two menus, "Take a photo" and "Choose an image," which will display the classification output and confidence values of the four types of Caladiums. Testing with 30 test data per class achieves 0.975 accuracy on both menus. On the “Take a photo” menu, precision is 0.974, recall is 0.9725, and f1-score is 0.965. Meanwhile, on the “Choose an image” menu a precision and recall value is 0.975, and f1-score value of 0.97.
- Open Access
- 29 March 2024
- Published by Ikatan Ahli Informatika Indonesia (IAII) in Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)
- Vol. 8 (2) , 242-249
Design and Implementation of the Shortest Path Navigation in Samosir District using Branch and Bound Algorithm
Samosir has a large area and several tourist attractions, making it difficult for visitors to explore the island. Unknowing the route can make the journey less fun and waste time. In general, tourists seek to know the fastest way to a tourist location to save time and money while on vacation. As a result, we require an application that will offer directions to the shortest path. This research aims to develop a web-based application that may display a map of the shortest travel to a tourist site. This website will display a map that marks the route from the origin point to the destination point. The Branch and Bound algorithm is used to determine the shortest path. The Python libraries OSMnx, Folium, and NetworkX modify paths and show a route map with OpenStreetMap. The error value of the distance between the branch, the bound algorithm, and Google Maps is used to obtain the RMSE accuracy value. The RMSE value is 3.02 and the MAPE value is 0.0023 indicating that the application produced already has a good implementation prototype. Furthermore, there is no significant distinction between the appearance of maps that implement OpenStreetMap and Google Maps.
- 29 March 2024
- Published by Ikatan Ahli Informatika Indonesia (IAII) in Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)
- Vol. 8 (2) , 242-249
Open AccessDesign and Implementation of the Shortest Path Navigation in Samosir District using Branch and Bound Algorithm
Samosir has a large area and several tourist attractions, making it difficult for visitors to explore the island. Unknowing the route can make the journey less fun and waste time. In general, tourists seek to know the fastest way to a tourist location to save time and money while on vacation. As a result, we require an application that will offer directions to the shortest path. This research aims to develop a web-based application that may display a map of the shortest travel to a tourist site. This website will display a map that marks the route from the origin point to the destination point. The Branch and Bound algorithm is used to determine the shortest path. The Python libraries OSMnx, Folium, and NetworkX modify paths and show a route map with OpenStreetMap. The error value of the distance between the branch, the bound algorithm, and Google Maps is used to obtain the RMSE accuracy value. The RMSE value is 3.02 and the MAPE value is 0.0023 indicating that the application produced already has a good implementation prototype. Furthermore, there is no significant distinction between the appearance of maps that implement OpenStreetMap and Google Maps.
- Open Access
- 14 September 2023
- Published by UKI Press in JURNAL Comunità Servizio : Jurnal Terkait Kegiatan Pengabdian kepada Masyarakat, terkhusus bidang Teknologi, Kewirausahaan dan Sosial Kemasyarakatan
- Vol. 5 (2) , 1293-1304
Pembuatan Website Kelompok Maduma Tani
Sipituhuta Village is one of the villages in Pollung District which has commodities in the agricultural sector. In 2016 a farmer group called the Maduma Tani Group was established as a means for sustainable agriculture to increase income. This farmer group provides farming tools that can be borrowed by farmers. However, the lack of information about the availability of farming tools and education on agricultural processing causes the process of borrowing tools and seeking agricultural processing education is still manual. The use of the website is one of the solutions in the field of information technology that helps increase farmers' knowledge in solving agricultural problems they face and borrowing farming tools in Sipituhuta Village. The development of this website uses the prototyping method, where this method is very suitable for building small-scale and customized websites that are created based on certain requests and needs. Website development using the PHP programming language, Laravel framework, and MySQL Database Management System (DBMS). Some of the features that have been successfully developed include lending tools, viewing history of tool lending, viewing education, viewing farming projects, viewing notifications. The result of this activity is that farmers in Sipituhuta Village can easily find information about agriculture and borrow farming tools. Website testing is done using the black box method. The test results show that the website that was built has been successfully running according to its function.
- 14 September 2023
- Published by UKI Press in JURNAL Comunità Servizio : Jurnal Terkait Kegiatan Pengabdian kepada Masyarakat, terkhusus bidang Teknologi, Kewirausahaan dan Sosial Kemasyarakatan
- Vol. 5 (2) , 1293-1304
Open AccessPembuatan Website Kelompok Maduma Tani
Sipituhuta Village is one of the villages in Pollung District which has commodities in the agricultural sector. In 2016 a farmer group called the Maduma Tani Group was established as a means for sustainable agriculture to increase income. This farmer group provides farming tools that can be borrowed by farmers. However, the lack of information about the availability of farming tools and education on agricultural processing causes the process of borrowing tools and seeking agricultural processing education is still manual. The use of the website is one of the solutions in the field of information technology that helps increase farmers' knowledge in solving agricultural problems they face and borrowing farming tools in Sipituhuta Village. The development of this website uses the prototyping method, where this method is very suitable for building small-scale and customized websites that are created based on certain requests and needs. Website development using the PHP programming language, Laravel framework, and MySQL Database Management System (DBMS). Some of the features that have been successfully developed include lending tools, viewing history of tool lending, viewing education, viewing farming projects, viewing notifications. The result of this activity is that farmers in Sipituhuta Village can easily find information about agriculture and borrow farming tools. Website testing is done using the black box method. The test results show that the website that was built has been successfully running according to its function.
- Open Access
- 1 August 2023
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 13 (4) , 4535-4546
Genetic algorithm to optimization mobility-based dengue mathematical model
Implementation of vaccines, mosquito repellents and several Wolbachia schemes have been proposed recently as strategies against dengue. Research showed that the use of vaccine and repellent is highly effective when implemented to individuals who are in area with high transmission rates, while the use of Wolbachia bacteria is strongly effective when implemented in area with low transmission rates. This research is to show a three-strategy combination to cope with the dengue using mathematical model. In dengue mathematical model construction, several parameters are not yet known, therefore a genetic algorithm method was used to estimate dengue model parameters. Numerical simulation results showed that the combination of three strategies were able to reduce the number of infected humans. The dynamic of the human population with the combination of three strategies on average was able to reduce the infected human population by 45.2% in immobility aspect. Furthermore, the mobility aspect in dengue model was presented by reviewing two areas; Yogyakarta and Semarang in Indonesia. The numerical solutions showed that the trend graph was almost similar between the two areas. With the maximum effort given, the combination control values decreased slowly until the 100th day.
- 1 August 2023
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 13 (4) , 4535-4546
Open AccessGenetic algorithm to optimization mobility-based dengue mathematical model
Implementation of vaccines, mosquito repellents and several Wolbachia schemes have been proposed recently as strategies against dengue. Research showed that the use of vaccine and repellent is highly effective when implemented to individuals who are in area with high transmission rates, while the use of Wolbachia bacteria is strongly effective when implemented in area with low transmission rates. This research is to show a three-strategy combination to cope with the dengue using mathematical model. In dengue mathematical model construction, several parameters are not yet known, therefore a genetic algorithm method was used to estimate dengue model parameters. Numerical simulation results showed that the combination of three strategies were able to reduce the number of infected humans. The dynamic of the human population with the combination of three strategies on average was able to reduce the infected human population by 45.2% in immobility aspect. Furthermore, the mobility aspect in dengue model was presented by reviewing two areas; Yogyakarta and Semarang in Indonesia. The numerical solutions showed that the trend graph was almost similar between the two areas. With the maximum effort given, the combination control values decreased slowly until the 100th day.
- Open Access
- 1 July 2023
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 31 (1) , 423-432
Sales forecasting of marketing using adaptive response rate single exponential smoothing algorithm
Micro, small and medium enterprises (UMKM) is one of the important aspects to support the improvement of the economy in Indonesia. Zee Mart’s business is one of the UMKM shop in Pematang Siantar City with sales and purchase transaction activities for supplies. The purpose of this study is to predict the sales of Zee Mart store goods in the coming month using the adaptive response rate single exponential smoothing (ARRSES) method. ARRSES is a method with the advantage of having two parameters, alpha and beta, where alpha will change every period when the data pattern changes. The dataset obtained will be pre-processed through data selection, cleaning, and transformation. The best beta is determined based on the level of accuracy calculated using the mean absolute percentage error (MAPE). Model development using the ARRSES method will produce forecasting percentages and errors for each product using MAPE. The number of sales data is 23,092 before preprocessing and 23,021 after pre-processing, with the total quantity of goods sold being 149,764 of 1,492 products. The results obtained using sales data 23,021 show the lowest MAPE value of 9.85 at the best beta of 0.6 with the highest accuracy of 90.15% and the model is implemented into a web interface.
- 1 July 2023
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 31 (1) , 423-432
Open AccessSales forecasting of marketing using adaptive response rate single exponential smoothing algorithm
Micro, small and medium enterprises (UMKM) is one of the important aspects to support the improvement of the economy in Indonesia. Zee Mart’s business is one of the UMKM shop in Pematang Siantar City with sales and purchase transaction activities for supplies. The purpose of this study is to predict the sales of Zee Mart store goods in the coming month using the adaptive response rate single exponential smoothing (ARRSES) method. ARRSES is a method with the advantage of having two parameters, alpha and beta, where alpha will change every period when the data pattern changes. The dataset obtained will be pre-processed through data selection, cleaning, and transformation. The best beta is determined based on the level of accuracy calculated using the mean absolute percentage error (MAPE). Model development using the ARRSES method will produce forecasting percentages and errors for each product using MAPE. The number of sales data is 23,092 before preprocessing and 23,021 after pre-processing, with the total quantity of goods sold being 149,764 of 1,492 products. The results obtained using sales data 23,021 show the lowest MAPE value of 9.85 at the best beta of 0.6 with the highest accuracy of 90.15% and the model is implemented into a web interface.
- Open Access
- 1 June 2023
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 13 (3) , 2908-2920
Corn Plant Disease Classification Based on Leaf using Residual Networks-9 Architecture
Classification on corn plants is used to classify leaf of corn plants that are healthy and have diseases consisting of Northern Leaf Blight, Common Rust and Gray Leaf Spot. Convolutional Neural Network (CNN) is one of algorithms from the branch of deep learning that utilizes artificial neural networks to produce accurate results in classifying an image. In this study, ResNet-9 architecture implemented to build the best model CNN for classification corn plant diseases. After that we doing comparisons of epochs have been carried out to obtain the best model, including comparisons of epochs of 5, 25, 55, 75 and 100. After the epoch comparison, the highest accuracy value was obtained in the 100 epoch experiment so that in this study 100 epochs were used in model formation. The number of datasets used is 9145 data which is divided into two, there are training data (80%) and testing data (20%). In this study, three hyperparameter tuning experiments were carried out and the results of hyperparameter tuning experiments where num_workers is 4 and batch_size is 32. This classification obtained an accuracy rate of 99% and the model is implemented into a web interface.
- 1 June 2023
- Published by Institute of Advanced Engineering and Science in International Journal of Electrical and Computer Engineering (IJECE)
- Vol. 13 (3) , 2908-2920
Open AccessCorn Plant Disease Classification Based on Leaf using Residual Networks-9 Architecture
Classification on corn plants is used to classify leaf of corn plants that are healthy and have diseases consisting of Northern Leaf Blight, Common Rust and Gray Leaf Spot. Convolutional Neural Network (CNN) is one of algorithms from the branch of deep learning that utilizes artificial neural networks to produce accurate results in classifying an image. In this study, ResNet-9 architecture implemented to build the best model CNN for classification corn plant diseases. After that we doing comparisons of epochs have been carried out to obtain the best model, including comparisons of epochs of 5, 25, 55, 75 and 100. After the epoch comparison, the highest accuracy value was obtained in the 100 epoch experiment so that in this study 100 epochs were used in model formation. The number of datasets used is 9145 data which is divided into two, there are training data (80%) and testing data (20%). In this study, three hyperparameter tuning experiments were carried out and the results of hyperparameter tuning experiments where num_workers is 4 and batch_size is 32. This classification obtained an accuracy rate of 99% and the model is implemented into a web interface.