Claim Missing Document
Check
Articles

Found 17 Documents
Search

Sistem Identifikasi Kesegaran Ikan Berbasis Android Menggunakan Convolutional Neural Network (CNN) Justam Justam; Muh. Nashir Takbir; Sitti Mawaddah Umar; Erlita Erlita; Revah Oktria Lewa
Jurnal Ilmiah Sistem Informasi dan Teknik Informatika (JISTI) Vol 7 No 2 (2024): Jurnal Ilmiah Sistem Informasi dan Teknik Informatika (JISTI)
Publisher : Lembaga Penelitian dan Pengabdian Masyarakat Universitas Lamappapoleonro

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.57093/jisti.v7i2.271

Abstract

Ikan merupakan sumber protein hewani penting dengan kandungan vitamin dan mineral esensial yang tinggi. Sebagai negara kepulauan, Indonesia memiliki potensi perikanan yang besar, namun masih banyak masyarakat yang kesulitan membedakan ikan segar dan tidak segar. Penelitian ini mengembangkan sistem berbasis Android untuk mengidentifikasi kesegaran ikan menggunakan Convolutional Neural Network (CNN). Model dilatih dengan 540 sampel gambar dalam tiga kategori (segar, baik, dan tidak layak) dengan resolusi 256 × 256 piksel RGB. CNN yang digunakan terdiri dari tiga lapisan konvolusi dan dua fully connected layer, dengan optimizer Adam dan fungsi aktivasi ReLU serta Softmax. Model dilatih di Google Colaboratory, lalu dikonversi ke TensorFlow Lite untuk diterapkan pada Android. Hasil pengujian menunjukkan akurasi 98% pada data uji dan 96,67% pada aplikasi Android dengan 60 sampel baru, membuktikan sistem mampu berfungsi dengan baik dalam mengidentifikasi kesegaran ikan
IoT dan Pengolahan Citra untuk Sistem Pakan Otomatis Udang dalam Kolam Bioflok Justam Justam; M Hasanuddin; Husni Mubarak; Rahmah Rahmah; Titin Nurjanah
Jurnal Ilmiah Sistem Informasi dan Teknik Informatika (JISTI) Vol 7 No 2 (2024): Jurnal Ilmiah Sistem Informasi dan Teknik Informatika (JISTI)
Publisher : Lembaga Penelitian dan Pengabdian Masyarakat Universitas Lamappapoleonro

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.57093/jisti.v7i2.272

Abstract

Pemberian pakan dalam budidaya udang vaname sangat penting karena mempengaruhi pertumbuhan udang. Saat ini, pengukuran berat rata-rata udang untuk menentukan jumlah pakan dilakukan secara manual, yang memakan waktu dan meningkatkan biaya operasional. Oleh karena itu, diperlukan sistem yang dapat mengestimasi berat rata-rata udang secara otomatis. Penelitian ini mengimplementasikan teknologi pengolahan citra digital untuk mengestimasi berat rata-rata udang vaname berdasarkan citra udang yang diambil. Fitur-fitur citra, seperti jumlah piksel, digunakan untuk estimasi berat rata-rata udang. Hasil penelitian menunjukkan bahwa regresi linier dengan Mean Absolute Error (MAE) sebesar 0,7752 dan regresi polinomial dengan MAE sebesar 0,6869 dapat digunakan untuk estimasi berat udang vaname. Selain itu, penggunaan metode Mask R-CNN terbukti efektif dalam mendeteksi dan menghitung jumlah udang dengan MAE sebesar 0,373. Penelitian ini juga menunjukkan bahwa pemberian pakan sesuai takaran yang dihitung berdasarkan estimasi berat udang dapat meningkatkan efisiensi dalam budidaya udang vaname. Sistem ini memberikan solusi cepat dan akurat untuk memperbaiki proses pemberian pakan, mengurangi biaya operasional, serta mengurangi risiko cedera bagi pembudidaya
Perbandingan Kinerja YOLO vs Faster R-CNN untuk Deteksi & Estimasi Berat Ikan Justam Justam; Abdul Malik; Erlita Erlita; Deo Mangellak; Yuyun Yuyun
Jurnal Ilmiah Sistem Informasi dan Teknik Informatika (JISTI) Vol 7 No 2 (2024): Jurnal Ilmiah Sistem Informasi dan Teknik Informatika (JISTI)
Publisher : Lembaga Penelitian dan Pengabdian Masyarakat Universitas Lamappapoleonro

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.57093/jisti.v7i2.273

Abstract

Ikan kerapu dan ikan kakap memiliki nilai ekonomi tinggi di pasar global, sehingga identifikasi jenis dan estimasi beratnya menjadi aspek penting dalam perdagangan. Metode manual yang umum digunakan memerlukan waktu lama dan tenaga kerja besar. Oleh karena itu, penelitian ini membandingkan performa dua model deep learning, yaitu YOLO dan Faster R-CNN, dalam mendeteksi jenis dan mengestimasi berat ikan. Dataset terdiri dari 2.991 citra yang terbagi dalam 18 kelas dan diperluas melalui augmentasi menjadi 6.843 citra. Proses deteksi menggunakan detection threshold 0,8, dengan evaluasi berdasarkan precision, recall, accuracy, serta Mean Absolute Percentage Error (MAPE) untuk estimasi berat. Hasil menunjukkan bahwa model YOLO memiliki precision, recall, dan accuracy masing-masing sebesar 0,98, 0,98, dan 0,96, sedangkan Faster R-CNN mencapai 0,97, 0,98, dan 0,95. Untuk estimasi berat, MAPE YOLO pada citra sebesar 2,42% dan pada video 3,66%, sementara Faster R-CNN memiliki MAPE 14,62% pada citra dan 13,59% pada video. Dengan demikian, model YOLO menunjukkan kinerja lebih baik dibandingkan Faster R-CNN dalam mendeteksi jenis dan mengestimasi berat ikan
AI POWERED DIGITAL HISTOPATHOLOGY: PREDICTING IMMUNOTHERAPY RESPONSE USING DEEP LEARNING Loso Judijanto; Som Chai; Ming Pong; Justam Justam; Ardi Azhar Nampira
Journal of Biomedical and Techno Nanomaterials Vol. 2 No. 3 (2025)
Publisher : Yayasan Adra Karima Hubbi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.70177/jbtn.v2i3.2379

Abstract

Immunotherapy has revolutionized cancer treatment, yet predicting which patients will respond remains a major clinical challenge. Current predictive biomarkers, such as PD-L1 expression, have limited accuracy and fail to capture the complex interplay of cells within the tumor microenvironment. Digital histopathology, the analysis of digitized tissue slides, combined with artificial intelligence (AI), offers a novel approach to identify complex morphological patterns that could serve as more robust predictive biomarkers. Objective: A deep learning model, specifically a convolutional neural network (CNN), was trained on a large, multi-center cohort of digitized tumor slides from patients with non-small cell lung cancer who had received ICI therapy. The model was trained to identify subtle morphological features and the spatial arrangement of tumor cells and tumor-infiltrating lymphocytes. The model’s predictive performance was rigorously validated on an independent, held-out test cohort, and its performance was compared to the predictive accuracy of PD-L1 staining. The AI-powered model successfully predicted immunotherapy response with a high degree of accuracy, achieving an area under the receiver operating characteristic curve (AUC) of 0.88 in the validation cohort.
Development of a Motor Vehicle Rearview Image Pattern Recognition System for Detection of Traffic Flow Violations on One-Way Roads: Image processing Sitti Mawaddah Umar; Justam Justam
Jurnal Media Informatika Vol. 6 No. 3 (2025): Jurnal Media Informatika
Publisher : Lembaga Dongan Dosen

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.55338/jumin.v6i3.6052

Abstract

This study aims to detect traffic violations, specifically motorcycles riding against the flow on one-way roads, by utilizing computer vision technology to recognize the rearview patterns of vehicles. The method employed involves applying the deep learning model Faster-RCNN for object detection, using image data captured from an IP camera mounted on a pole at a height of 2.5 meters with a 45-degree tilt angle to optimally monitor vehicles from behind. Image labeling was performed using the LabelImg application, while model training and classification were conducted using the TensorFlow framework. The developed system achieved a detection accuracy of 88%, demonstrating the effectiveness of this approach in identifying motorcycles violating traffic direction. These findings highlight the potential of computer vision as an automatic and real-time solution for traffic monitoring, which can help reduce dangerous violations and enhance road safety. Therefore, this research contributes significantly to the development of more advanced and efficient traffic violation detection systems.
UTILIZING MACHINE LEARNING ALGORITHMS TO PREDICT PATIENT SURGES IN FUTURE EMERGENCY DEPARTMENTS Justam Justam; Luis Santos; Josefa Flores
Journal of World Future Medicine, Health and Nursing Vol. 4 No. 4 (2026)
Publisher : Yayasan Adra Karima Hubbi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.70177/health.v4i4.4278

Abstract

Emergency departments face increasing pressure from unpredictable patient surges that can intensify overcrowding, treatment delays, staff workload, and resource shortages. This study aimed to develop and evaluate machine learning algorithms for predicting emergency demand across multiple hospitals and forecast horizons. A quantitative multi-site predictive modeling design combined retrospective records from 1,427,860 emergency encounters with prospective validation data from 151,632 visits. Seasonal ARIMA, Poisson regression, support vector regression, random forest, long short-term memory, extreme gradient boosting, and hybrid ensemble models were compared using temporal and external validation. Results showed that the hybrid ensemble achieved the strongest twenty-four-hour performance, with a mean absolute error of 21.84 patients, mean absolute percentage error of 6.58%, and an R-squared value of 0.91. The model also produced high six-hour surge sensitivity, while recent arrivals, occupancy, ambulance activity, infectious disease indicators, bed availability, weather, air quality, and public events emerged as influential predictors. Predictive accuracy declined for longer horizons and less frequent clinical categories. The study concludes that machine learning can strengthen emergency preparedness when forecasts are explainable, continuously validated, and connected to predefined staffing, triage, bed-management, and ambulance-response protocols.  
ADVANCED SENSOR FUSION: SIGNAL PROCESSING ARCHITECTURES FOR AUTONOMOUS ENGINEERING PLATFORMS Justam Justam; Liu Yang; Chen Mei
Journal of Moeslim Research Technik Vol. 3 No. 4 (2026)
Publisher : Yayasan Adra Karima Hubbi

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.70177/technik.v3i4.4323

Abstract

The reliability of autonomous engineering platforms depends fundamentally on the seamless integration of high-bandwidth, multi-modal sensory data to perceive dynamic environments. This research addresses the persistent challenge of computational latency and perception failure in traditional sensor fusion architectures when operating under adverse conditions. The study aims to evaluate a novel hybrid signal processing architecture that optimizes the balance between edge-level feature extraction and centralized semantic synthesis. Utilizing a “Hardware-in-the-Loop” methodology, the proposed framework was tested on an embedded GPU testbed using synchronized LiDAR, RADAR, and camera datasets across 500 diverse navigational scenarios. Results demonstrate that the hybrid architecture achieves a 56% reduction in processing latency, maintaining a mean response time of 12.4 milliseconds without compromising positional accuracy, which remained stable at 0.11 meters RMSE. Furthermore, the implementation of an entropy-driven weighting mechanism allowed the system to maintain 99.2% anomaly detection accuracy during simulated sensor failures. This research concludes that decentralized feature processing is essential for the operational continuity of energy-constrained autonomous systems. The findings provide a scalable blueprint for developing resilient, low-power perception modules, asserting that hardware-aware signal processing is a prerequisite for achieving Level 5 autonomy in complex, real-world engineering applications.