Jurnal Sistem Cerdas

Balance Control of a Hexapod Robot Using Fuzzy Logic and Inverse Kinematics Algorithm with Real-Time IMU Sensor Measurement

2025-08-31T21:57:53+07:00

Robotic technology is a crucial pillar in modern civilization, especially in high-risk environments such as post-disaster evacuation scenarios. Hexapod legged robots are designed to navigate uneven terrains that are inaccessible to humans. Although hexapods offer superior mobility and flexibility, they face stability challenges when moving on inclined surfaces due to uneven load distribution, which can affect servo motor performance. To address this issue, this study implements a control system combining fuzzy logic and inverse kinematics to maintain body stability. An Inertial Measurement Unit (IMU) sensor is also integrated to detect the robot’s orientation angle in real-time, enabling adaptive posture correction. This research focuses on three main problems: first, how inverse kinematics can stabilize hexapod posture on sloped surfaces; second, how IMU sensors detect inclination and orientation; and third, how fuzzy logic control contributes to balance regulation. The methodology involves system design, experimental testing, and performance analysis based on the robot's body tilt measurements across various inclinations. The results show that the proposed system responds effectively to surface tilt, particularly in pitch angle correction and maintaining a neutral position. Inverse kinematics successfully calculates leg configurations to keep the body posture stable. The IMU sensor demonstrates high accuracy in angle detection, while fuzzy logic provides flexibility in decision-making for posture control. The integration of these three approaches proves effective in maintaining hexapod balance on inclined terrains, thus supporting their potential use in complex, unstable environments.

Analysis of MP3 Bitrate on the Accuracy of Academic Audio Transcription Using Whisper large-v3

2025-08-31T21:57:53+07:00

In the digital era, automatic transcription is a crucial solution for converting audio content into text, especially in the context of academic documentation. The main challenge in this process is transcription accuracy, which can be affected by the quality of the audio file, including its bitrate and file size. This study aims to analyze the impact of MP3 bitrate and file size on transcription accuracy using the Whisper large-v3 model. Five academic audio files were converted into five different bitrate levels, ranging from 64 kbps to 320 kbps, and then transcribed automatically using the Whisper model. Evaluation was conducted by calculating the Word Error Rate (WER) as an indicator of transcription accuracy. In addition, processing time and file size were recorded to analyze transcription efficiency. The results show that increasing bitrate does not always lead to higher accuracy. Bitrates of 128–192 kbps provided the best balance between transcription accuracy, processing efficiency, and file size. This study makes a significant contribution to the development of automatic transcription systems based on ASR models, particularly for audio documentation needs in educational institutions. These findings serve as a technical reference for developing efficient and accurate audio documentation systems in academic environments.

Analysis of Defense Mechanisms Against FGSM Adversarial Attacks on ResNet Deep Learning Models Using the CIFAR-10 Dataset

2025-08-31T21:57:55+07:00

Adversarial attacks threaten the reliability of deep learning models in image classification, requiring effective defense mechanisms. This study evaluates how defense distillation and adversarial training protect ResNet18 models trained on CIFAR-10 data against Fast Gradient Sign Method (FGSM) attacks. The baseline model achieves 85.01% accuracy on clean data but its accuracy falls to 19.23% when FGSM attacks at epsilon 0.3. The accuracy of defense distillation drops to 23.68% when epsilon reaches 0.3 but adversarial training maintains 0.34% accuracy at epsilon 0.25 although it reduces clean data accuracy to 57.08%. The analysis shows that classes with similar visual characteristics such as cats and dogs remain vulnerable to attacks. The study demonstrates the requirement for balanced defense approaches while indicating additional work needs to improve model robustness.

Smart Rice Disease Detection Based on Leaf Analysis Using the YOLO Algorithm with an Interactive User Interface

2025-08-31T21:57:55+07:00

Rice is an important commodity for human life. The application of appropriate technology continues to be developed and researched as an effort to create food security. Indonesia is one of the largest rice producers but has not fully implemented agricultural technology. The lack of application of technology causes agricultural techniques to be still traditional. This causes the younger generation to be less interested in working as farmers. One of the challenges for novice farmers is how to handle plant diseases. This study aims to design a disease detection system so that it can be easier to handle. This plant detection uses a deep learning method with the YOLO V5 Algorithm. To obtain the best model, each YOLO V5 version was compared. The experimental results showed that the detection of healthy plants could be predicted better (0.99) than the other classes. Based on the predicted value, it means that the extra-large version is better (0.83) than the other versions. In addition, this study also designed the user interface with website application media. This website can be accessed via a laptop or smartphone so that its use is more effective and efficient. The user interface design is designed simply so that farmers and novices can easily learn and use it. With this research, it is hoped that rice production can be increased and one way to attract the interest of the younger generation.

Leveraging BERT and T5 for Comprehensive Text Summarization on Indonesian Articles

2025-08-31T21:57:55+07:00

One of the main challenges in the field of Natural Language Processing (NLP) is developing systems for automatic text summarization. These systems typically fall into two categories: extractive and abstractive. Extractive techniques generate summaries by selecting important sentences or phrases directly from the original text, whereas abstractive techniques focus on rephrasing or paraphrasing the content, producing summaries that resemble human-written ones. In this research, models based on Transformer architectures, including BERT and T5, were used, which have been shown to effectively summarize texts in various languages, including Indonesian. The dataset used was INDOSUM, consisting of Indonesian news articles. The best results were achieved with the T5 model, using the abstractive approach, recorded ROUGE-1, ROUGE-2, and ROUGE-L scores of 69.36%, 61.27%, and 66.17%, respectively. On the other hand, the extractive BERT model achieved ROUGE-1, ROUGE-2, and ROUGE-L scores of 70.82%, 63.99%, and 58.40%.

Mini Drone-Based Precision Agriculture for Indonesian MSMEs: A Low-Cost AI-Assisted Monitoring System

2025-09-06T19:34:45+07:00

This research introduces a cost-effective drone-based agricultural monitoring system targeted at Indonesia’s smallholder farming enterprises (MSMEs). By leveraging mini drones (DJI Mini 2 SE) and lightweight AI models, farmers can segment land, detect vegetation health, and count crops using simple RGB video analysis. The system utilizes a mobile-to-YouTube private livestream pipeline and performs video processing offline using semantic segmentation (U-Net) and object detection (YOLOvX). The prototype system—tested on a 300m² vegetable plot—shows promising results with over 90% detection accuracy and effective land use visualization. The interface, built with Streamlit, provides real-time insights, affordability, and aligns with Smart City goals of accessibility and sustainability.

Feature Selection and Reduction in Happiness Index Analysis: A Systematic Literature Review

2025-08-31T21:57:55+07:00

This study investigates the role and effectiveness of feature selection and feature reduction techniques in improving the accuracy, validity, and efficiency of predictive models for survey-based happiness indices. A Systematic Literature Review (SLR) was conducted following the PRISMA 2020 protocol, evaluating 40 peer-reviewed articles published between 2020 and 2025. The results demonstrate that feature selection methods namely wrapper, filter, and embedded approaches can significantly enhance model performance, yielding higher coefficients of determination (R²) and lower prediction errors. Furthermore, the identification of relevant features has been shown to improve construct validity and the reliability of happiness indicators. The integration of feature selection and feature reduction techniques also contributes to more efficient and stable models, particularly in high-dimensional data contexts. However, the limited number of studies directly addressing happiness and the methodological heterogeneity across works pose challenges to the generalizability of the findings. This review provides valuable insights for establishing evidence-based practices and guiding strategic developments in future happiness index analytics