Sistem Identifikasi Objek di Perairan Keruh (Tambak) Berbasis Underwater Image Processing

Mashuri, Mashuri (2019) Sistem Identifikasi Objek di Perairan Keruh (Tambak) Berbasis Underwater Image Processing. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 02311750010009-Master_Thesis.pdf Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Hamburan di dalam air menyebabkan efek kabur dalam hasil fotografi, penyerapan panjang gelombang menyebabkan pengurangan warna. Permasalahan yang terjadi pada citra bawah air antara lain adalah redaman cahaya, hamburan, pencahayaan yang tidak seragam, bayangan, dan perubahan warna. Perumusan masalah tersebut dapat diatasi melalui pemrosesan citra optik bawah air dengan pendekatan hardware dan software. Pendekatan hardware salah satunya dengan memasang polarisator di depan kamera yang digunakan. Pendekatan software dilakukan image processing secara digital dengan metode dark channel prior (DCP). Penelitian penguatan citra bawah air dengan mengkombinasikan metode polarisasi dan DCP tersebut perlu dilakukan untuk menjawab permasalah yang muncul. Penelitian dilakukan dengan melakukan analisis pengaruh perubahan turbidity terhadap kualitas citra yang didapatkan. Citra objek diambil dengan sensor kamera yang dipasang polarisator maupun yang tidak. Citra yang didapat selanjutnya diolah menggunakan perangkat lunak yang menerapkan prinsip DCP. Model You Only Look Once (YOLO) yang menerapkan konsep filter partikel merupakan metode dalam identifikasi objek berbasis struktur dan warna. Citra hasil proses underwater image processing diidentifikasi menggunakan prinsip filter partikel dalam deep learning yang dibangun dalam logika konvoluasi jaringan saraf tiruan (CNN). Penelitian menunjukkan bahwa terjadi pengurangan kualitas citra bawah air yang dipengaruhi meningkatnya nilai kekeruhan pada medium air. Penelitian menunjukkan bahwa citra yang diambil dengan polarisator mampu menghasilkan citra yang memiliki nilai matrix square error (MSE) lebih rendah dibandingkan dengan tanpa polaristor. Penelitian menunjukkan DCP dapat digunakan untuk mengurangi kabut yang terdapat pada citra yang disebabkan kekeruhan medium air. Penerapan deep learning yang mengacu prinsip filter partikel mampu mengidentifikasi dan menghitung objek di bawah perairan keruh dengan rata-rata 6,16 frame per second (FPS) dan rata-rata mAP sebesar 81,52%. Perhitungan objek memiliki rata-rata ketelitian 89% pada percobaan di tambak udang yang keruh, dan 90% untuk percobaan di air jernih.
=============================================================================================================================
Scattering in water causes a blurring effect in photographic results, while absorption of certain wavelengths causes color reduction. Problems that occur in underwater images include light attenuation, scattering, non-uniform lighting, shadows, and discoloration. These problems can be addressed through underwater optical image processing using both hardware and software approaches. One hardware approach is installing a polarizer in front of the camera used, while the software approach involves digital image processing using the Dark Channel Prior (DCP) method. Research on enhancing underwater images by combining polarization and DCP methods is needed to address these problems. The study was conducted by analyzing the effect of changes in turbidity on the resulting image quality. Images of the object were captured using a camera sensor, both with and without a polarizer mounted. The images obtained were then processed using software applying the DCP principle. The You Only Look Once (YOLO) model, incorporating the particle filter concept, was used as a method for identifying objects based on structure and color. The resulting images from underwater image processing were identified using the particle filtering principle in deep learning, built on convolutional neural network (CNN) logic. The research shows a reduction in underwater image quality as the turbidity value of the water medium increases. It also shows that images taken with a polarizer produce a lower mean square error (MSE) compared to those taken without one. Additionally, DCP proved effective in reducing the haze present in images caused by water turbidity. The application of deep learning based on the particle filter principle was able to identify and count objects in turbid water with an average speed of 6.16 frames per second (FPS) and an average mAP of 81.52%. Object counting achieved an average accuracy of 89% in experiments conducted in turbid shrimp ponds, and 90% in experiments conducted in clear water.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	turbidity, underwater image processing, identifikasi objek.
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing T Technology > TR Photography
Divisions:	Faculty of Industrial Technology > Physics Engineering > 30101-(S2) Master Thesis
Depositing User:	MASHURI .
Date Deposited:	21 Jul 2026 03:36
Last Modified:	21 Jul 2026 04:07
URI:	http://repository.its.ac.id/id/eprint/69826

Actions (login required)

View Item