Korean Journal of Artificial Intelligence, KODISA Foundation

PDF Citaion Utilization of Google Street View to Estimate Green View Index: a case study from Bandung, Indonesia

Emir LUTHFI LUTHFI(BPS Statistics Indonesia) ; Setia PRAMANA(Politeknik Statistika STIS) pp.1-7 https://doi.org/10.24225/kjai.2024.12.4.1

초록보기

Abstract

The use of street view has many benefits with its popular source being Google Street View (GSV). One of the processing methods uses semantic segmentation which can classify each pixel according to the category of the pre-trained pyramid scene parsing network (PSPNet) model used. The Green View Index (GVI) is one of the semantic segmentation research trends in viewing Green Open Space (GOS) based on human perception of an area. Green Open Space (GOS) provides many benefits and more attractiveness to the community to be able to live in the vicinity. The GVI obtained gives an average value of 22.5% capturing the presence of GOS which is higher than the green open space data collected by Housing and Settlement Area, Land and Parking Offices Bandung City.

PDF Citaion Implementation of Machine Learning for Spam Detection and Topic Modeling for Emails in Bahasa Indonesia

Masna Novita RAHMANIAR(BPS Statistics Indonesia) ; Ahmad HARTIONO(BPS Statistics) ; Setia PRAMANA(STIS Polytechnic of Statistics) pp.9-19 https://doi.org/10.24225/kjai.2024.12.4.9

초록보기

Abstract

Indonesia ranks fifth as the country of origin for spammers. Attention is urgently needed to tackle spam, especially in Bahasa Indonesia (Indonesian language), which can be achieved by building the best spam detection model. This study aims to compare machine learning models for spam detection, study spam email modeling topics, and design the implementation on the REST API. Spam detection is carried out using machine learning algorithms, i.e., Long Short Term Memory (LSTM), K-Nearest Neighbours (KNN), Naive Bayes, Random Forest, Adaboost, and Support Vector Machine (SVM) combined with slang preprocessing convert and translate. Furthermore, Latent Dirichlet Allocation (LDA) is used for topic modeling of spam emails. The results show that slang processes convert and translate can improve accuracy and f1-score, Long Short Term Memory (LSTM) was the best method with accuracy 93.15% and f1-score of 93.01%, compared to the other methods. In addition, there were five main topics on data categorized as spam: promotions, job vacancies, educational offers, bulletins and news, and investment and finance. A REST API model was successfully developed to separate spam categories based on promotional and other topics.

PDF Citaion Effectiveness of Noise Reduction in LDCT Images Based on SRCNN

Doo Bin KIM(Cheju Halla University) ; Hyun MEE PARK(National Medical Center) ; Sang Hoon JOON(National Medical Center) ; Joo Wan HONG(Eulji University) pp.21-26 https://doi.org/10.24225/kjai.2024.12.4.21

초록보기

Abstract

This study aims to evaluate the performance of noise reduction in LDCT images using an SRCNN based AI model. Using the Lungman phantom, images of effective mAs 72 recommended by AAPM and effective mAs 709 without using the AEC function were acquired. SRCNN model input image used a GT, label image and GT image was used as a low-resolution image. Image evaluation was conducted in the lung apex, middle level lung, and carina of trachea regions, and PSNR, SSIM, SSIM error map, SNR, MSE, and RMSE were used as evaluation indices are based on label image. Lung apex results showed increase of 19.52, 29.69 and 23.8%, and decreased of 71.37, 46.43% respectively. Middle level lung results showed increase of 20.99, 20.0 and 26.26%, and decreased of 72.67, 47.72% respectively. Carina of trachea results showed increase of 22.05, 32.31 and 28.18%, and decreased of 73.93, 48.93% respectively. Image evaluation results were improvement in image quality due to noise reduction was confirmed using the SRCNN based AI model. Therefore, confirmed that applying the SRCNN to LDCT images can improve image quality by reducing noise, and it is considered that AI based post processing will be useful for CT images without AI.

PDF Citaion Learnable Sobel Filter and Attention-based Deep Learning Framework for Early Forest Fire Detection

Sehun KIM(Kwangwoon University) ; Kyeongseok JANG(Kwangwoon University) ; Dongwoo LEE(Kwangwoon University) ; Seungwon CHO(Kwangwoon University) ; Seunghyun LEE(Kwangwoon University) ; Kwangchul SON(Kwangwoon University) pp.27-33 https://doi.org/10.24225/kjai.2024.12.4.27

초록보기

Abstract

Various techniques are being researched to effectively detect forest fires. Among them, techniques using object detection models can monitor forest fires over wide areas 24 hours a day. However, detecting forest fires early with traditional object detection models is a very challenging task. While they show decent accuracy for thick smoke and large fires, they show low accuracy for faint smoke and small fires, and frequently generate false positives for lights that are like fires. In this paper, to solve these problems, we focus on leveraging local characteristics such as contours and textures of fire and smoke, which are crucial for accurate detection. Based on this approach, we propose EDAM (Edge driven Attention Module) that performs enhancement by richly utilizing contour and texture information of fire and smoke. EDAM extracts important edge information to generate feature maps with emphasized contour and texture information, and based on this map, performs Attention Mechanism to emphasize key characteristics of smoke and fire. Through this mechanism, the overall model performance was improved, with AP_sincreasing from 0.154 to 0.204 and AP_0.5 from 0.779 to 0.784, resulting in a significant improvement in AP_Svalue to 32.47%. In practice, the model applying this technique showed excellent inference speed while greatly improving detection performance for small objects compared to existing models and reduced false positive rates for building and street light illumination in nighttime environments that are easily mistaken for fire.

PDF Citaion A Study on Improving Pressure Sensor Calibration Based on Multiple Calibration Points and Auto Target Setting*

Jonghyun OH(Daejeon University) ; Jae-Yong HWANG(Daejeon University) ; Tumenbat TENGIS(Daejeon University) ; Woo-Seong JUNG(Daejeon University) pp.35-42 https://doi.org/10.24225/kjai.2024.12.4.35

초록보기

Abstract

Pressure sensors are essential equipment for precise measurements in industrial and research fields, requiring calibration and target value setting for each sample to ensure high accuracy. This study proposes an automated target value prediction method based on a polynomial regression model to enhance pressure sensor accuracy and evaluates its effectiveness. Experiments were conducted over a pressure range of 0 to 45 bar and a temperature range of -5°C to 60°C. By expanding the calibration points from the previous two to four, linearity error was improved from 0.380% to 0.116%. In the conventional method, theoretical output values were manually calculated based on LDO voltage, and target values were set accordingly. However, this study employed a method that uses Polynomial Features (degree=2) transformation followed by a Linear Regression model to automatically predict target values. This approach allowed samples to more precisely follow the target voltage. This study demonstrates that an automated target value setting with multiple calibration points can contribute to improving the accuracy of pressure sensor measurements.

PDF Citaion Analysis of the Impact of ESG on Corporate Credit

Yeji KIM(Gyeongsang National University) ; Sangmok LEE(Gyeongsang National University) ; Doobae JUN(Gyeongsang National University) pp.43-51 https://doi.org/10.24225/kjai.2024.12.4.43

초록보기

Abstract

This study analyzed the effect of ESG on corporate credit ratings. Currently, interest in ESG at home and abroad is increasing, such as Korea's mandatory disclosure of ESG information in 2025 and the carbon neutrality policy in 2050. At the same time, this study assumed that ESG lists, which are non-financial factors, would have an indirect and partial effect on a company's credit rating, and analyzed it by year and industry. From 2011 to 2021, the importance of variables was measured using ESG division data provided by the Korea Institute of Corporate Governance and Sustainability and KIS-Value's financial statements. Also, Mean Decrease Impurity(MDI) and Recursive Feature Elimination(RFE) were used as variable importance measurement methods. As a result of the study, the importance of E(Environment), S(Social), and G (Governance) items all increased in 2021, compared to 2011, increasing the effect of ESG on corporate credit ratings. In particular, it was found that the importance of S increased the most. In addition, through analysis by industry, it was confirmed that the degree of impact of ESG lists varies from industry to industry. This is a result that can infer the discriminatory application of ESG by industry.

바로가기메뉴