IPB University Logo

SCIENTIFIC REPOSITORY

IPB University Scientific Repository collects, disseminates, and provides persistent and reliable access to the research and scholarship of faculty, staff, and students at IPB University

AI Repository
 
Building and Categories


      View Item 
      •   IPB Repository
      • Final Assignments
      • Doctoral Final Assignments
      • DF - School of Data Science, Mathematic and Informatics
      • View Item
      •   IPB Repository
      • Final Assignments
      • Doctoral Final Assignments
      • DF - School of Data Science, Mathematic and Informatics
      • View Item
      JavaScript is disabled for your browser. Some features of this site may not work without it.

      Peningkatan Kualitas Citra Melalui Optimasi Fusi CLAHE-MSR dengan Artificial Bee Colony untuk Deteksi Ikan di Padang Lamun

      Thumbnail
      View/Open
      Cover (537.7Kb)
      Fulltext (4.189Mb)
      Date
      2026
      Author
      Asri, Sri Dianing
      Jaya, Indra
      Buono, Agus
      Wijaya, Sony Hartono
      Metadata
      Show full item record
      Abstract
      Ekosistem padang lamun memiliki peran ekologis penting sebagai habitat, area pembesaran, tempat berlindung, dan sumber makanan bagi berbagai spesies ikan. Namun, pemantauan keberadaan ikan pada ekosistem ini masih menghadapi tantangan besar, terutama ketika dilakukan menggunakan citra bawah air. Citra yang dihasilkan sering tampak buram, memiliki kontras rendah, warna tidak seimbang, dan detail objek ikan sulit dikenali. Kondisi tersebut berdampak langsung terhadap kinerja sistem deteksi ikan berbasis visi komputer. Penelitian ini bertujuan untuk mengembangkan pendekatan peningkatan kualitas citra bawah air dan mengevaluasi pengaruhnya terhadap deteksi ikan pada ekosistem padang lamun. Pendekatan yang dikembangkan adalah fusi CLAHE dan MSR dengan optimasi bobot menggunakan Artificial Bee Colony (ABC). Underwater Image Quality Measure (UIQM) digunakan sebagai fungsi fitness untuk mengoptimalkan kombinasi bobot terbaik antara Contrast Limited Adaptive Histogram Equalization (CLAHE) dan Multi-Scale Retinex (MSR). Hasil optimasi menghasilkan bobot optimal sebesar 0,34 untuk CLAHE dan 0,66 untuk MSR. Komposisi ini menunjukkan bahwa kontribusi MSR lebih dominan dalam memperbaiki pencahayaan, warna, dan ketajaman citra bawah air, sedangkan CLAHE tetap berperan dalam menjaga kontras lokal dan ketegasan detail visual. Metode penelitian diawali dengan persiapan dataset citra bawah air yang terdiri atas citra UIEB sebagai data baseline, citra padang lamun sebagai data utama, serta citra hutan bakau dan terumbu karang dari Data Deepfish sebagai data pengujian robustness. Peningkatan kualitas citra dilakukan melalui beberapa skenario, yaitu CLAHE, MSR, pipeline CLAHE–MSR, pipeline MSR–CLAHE, fusi CLAHE–MSR dengan bobot tetap, dan fusi CLAHE–MSR dengan bobot optimal berbasis ABC. Evaluasi kualitas citra dilakukan menggunakan UIQM, komponen UIQM, entropi informasi, distribusi histogram RGB, serta pengamatan visual. Citra hasil peningkatan kualitas kemudian digunakan sebagai masukan dalam proses fine-tuning YOLO-Fish untuk mengevaluasi pengaruh image enhancement terhadap performa deteksi ikan. Hasil pengujian pada data UIEB menunjukkan bahwa citra awal memiliki nilai UIQM sebesar 2,124. Setelah dilakukan peningkatan kualitas secara individual, CLAHE menghasilkan nilai UIQM sebesar 2,419, sedangkan MSR menghasilkan nilai UIQM sebesar 1,897. Pipeline CLAHE–MSR menghasilkan nilai UIQM sebesar 1,675, sedangkan pipeline MSR–CLAHE menghasilkan nilai UIQM sebesar 1,971. Hasil ini menunjukkan, pemrosesan berurutan cenderung menyebabkan perubahan warna, pencahayaan, dan kontras yang berlebihan. Sebaliknya, strategi fusi dengan bobot tetap memberikan hasil yang lebih baik, yaitu UIQM sebesar 2,920 pada kombinasi bobot 0,25:0,75, UIQM sebesar 2,919 pada bobot 0,5:0,5, dan UIQM sebesar 2,735 pada bobot 0,75:0,25. Nilai tertinggi diperoleh pada fusi bobot optimal berbasis ABC dengan UIQM sebesar 2,969. Hasil ini menunjukkan bahwa optimasi bobot menggunakan ABC mampu menentukan komposisi CLAHE dan MSR yang lebih tepat dibandingkan metode tunggal, strategi pipeline, maupun fusi bobot tetap. Penerapan bobot optimal pada citra lamun monospesifik menunjukkan bahwa metode fusi CLAHE–MSR mampu meningkatkan kualitas citra secara kuantitatif dan visual. Nilai UIQM citra awal sebesar 1,454 meningkat menjadi 1,805 setelah CLAHE, 2,394 setelah MSR, dan mencapai nilai tertinggi sebesar 2,513 pada citra fusi CLAHE–MSR. Analisis entropi juga menunjukkan peningkatan kandungan informasi visual, yaitu dari 5,77410 pada citra awal menjadi 7,20409 pada citra fusi. Pada citra lamun vegetasi campuran, metode fusi CLAHE–MSR juga menghasilkan performa terbaik. Nilai UIQM meningkat dari 2,315 pada citra awal menjadi 2,940 pada CLAHE, 2,653 pada MSR, dan mencapai nilai tertinggi sebesar 3,195 pada hasil fusi. Nilai entropi informasi juga meningkat dari 6,24813 pada citra awal menjadi 7,69423 pada citra fusi. Analisis histogram RGB memperkuat hasil evaluasi UIQM dan entropi informasi. Citra awal menunjukkan ketidakseimbangan distribusi warna, terutama karena kanal merah cenderung lebih lemah, sedangkan kanal hijau dan biru lebih dominan. Setelah proses fusi CLAHE–MSR, distribusi intensitas menjadi lebih baik, kanal merah meningkat pada rentang intensitas sedang, dan dominasi kanal hijau serta biru menjadi lebih terkendali. Pengujian robustness pada Data Deepfish menunjukkan bahwa bobot optimal hasil ABC juga mampu meningkatkan kualitas citra. Pada citra hutan bakau, nilai UIQM meningkat dari 2,053 pada citra awal menjadi 2,839 pada CLAHE, 2,520 pada MSR, dan mencapai nilai tertinggi sebesar 3,313 pada citra fusi. Pada citra terumbu karang, nilai UIQM meningkat dari 2,074 pada citra awal menjadi 2,488 pada CLAHE, 2,200 pada MSR, dan mencapai nilai tertinggi sebesar 3,311 pada citra fusi. Evaluasi kinerja deteksi menunjukkan bahwa peningkatan kualitas citra memberikan dampak positif terhadap performa model YOLO-Fish. Pada data uji, metode fusi CLAHE–MSR menghasilkan performa terbaik dengan mAP@0,5 sebesar 86,6%, meningkat 8,9 poin dibandingkan citra asli sebesar 77,7%, serta lebih tinggi dibandingkan CLAHE sebesar 84,0% dan MSR sebesar 84,7%. Pada metrik lain, metode fusi menghasilkan precision sebesar 91,0%, recall sebesar 82,0%, dan F1-score sebesar 86,0%. Kebaruan dari penelitian yang diusulkan menghasilkan pertama pendekatan fusi CLAHE–MSR berbasis optimasi ABC untuk meningkatkan kualitas citra bawah air pada ekosistem padang lamun. Pada pendekatan ini, solusi direpresentasikan sebagai pasangan bobot CLAHE (?CL) dan bobot MSR (?M) yang dioptimalkan menggunakan UIQM sebagai fungsi fitness. Kedua, penelitian ini mengintegrasikan citra hasil fusi CLAHE–MSR teroptimasi ke dalam proses fine-tuning YOLO-Fish untuk mengevaluasi pengaruh peningkatan kualitas citra terhadap performa deteksi ikan pada ekosistem padang lamun yang memiliki kompleksitas visual tinggi.
       
      Seagrass ecosystems play an important ecological role as habitats, nursery grounds, shelters, and food sources for various fish species. However, monitoring fish presence in these ecosystems remains challenging, particularly when using underwater images. The resulting images are often blurred, have low contrast, exhibit color imbalance, and contain fish object details that are difficult to recognize. These conditions directly affect the performance of computer vision-based fish detection systems. This study aims to develop an underwater image enhancement approach and evaluate its effect on fish detection in seagrass ecosystems. The proposed approach is based on the fusion of CLAHE and MSR, with weighting optimization using the Artificial Bee Colony (ABC) algorithm. Underwater Image Quality Measure (UIQM) is used as the fitness function to optimize the best weight combination between Contrast Limited Adaptive Histogram Equalization (CLAHE) and Multi-Scale Retinex (MSR). The optimization results produced an optimal weight of 0.34 for CLAHE and 0.66 for MSR. This composition indicates that MSR contributes more dominantly to improving illumination, color, and sharpness in underwater images, while CLAHE still plays a role in preserving local contrast and enhancing visual details. The research method began with the preparation of underwater image datasets, consisting of UIEB images as the baseline data, seagrass images as the main data, and mangrove and coral reef images from the DeepFish dataset as robustness testing data. Image enhancement was carried out through several scenarios, namely CLAHE, MSR, CLAHE–MSR pipeline, MSR–CLAHE pipeline, CLAHE–MSR fusion with fixed weights, and CLAHE–MSR fusion with ABC-based optimal weights. Image quality evaluation was performed using UIQM, UIQM components, information entropy, RGB histogram distribution, and visual observation. The enhanced images were then used as input in the YOLO-Fish fine-tuning process to evaluate the effect of image enhancement on fish detection performance. The experimental results on the UIEB data showed that the original images had a UIQM value of 2.124. After individual enhancement, CLAHE produced a UIQM value of 2.419, while MSR produced a UIQM value of 1.897. The CLAHE–MSR pipeline produced a UIQM value of 1.675, while the MSR–CLAHE pipeline produced a UIQM value of 1.971. These results indicate that sequential processing tends to cause excessive changes in color, illumination, and contrast. In contrast, the fixed-weight fusion strategy produced better results, with a UIQM value of 2.920 for the 0.25:0.75 weight combination, 2.919 for the 0.5:0.5 weight combination, and 2.735 for the 0.75:0.25 weight combination. The highest value was obtained by the ABC-based optimal weighted fusion, with a UIQM value of 2.969. These results show that weight optimization using ABC is able to determine a more appropriate CLAHE and MSR composition compared with single methods, pipeline strategies, and fixed-weight fusion. The application of the optimal weights to monospecific seagrass images showed that the CLAHE–MSR fusion method was able to improve image quality both quantitatively and visually. The UIQM value of the original images was 1.454, increasing to 1.805 after CLAHE, 2.394 after MSR, and reaching the highest value of 2.513 in the CLAHE–MSR fusion images. The entropy analysis also showed an increase in visual information content, from 5.77410 in the original images to 7.20409 in the fusion images. In mixed seagrass vegetation images, the CLAHE–MSR fusion method also produced the best performance. The UIQM value increased from 2.315 in the original images to 2.940 with CLAHE, 2.653 with MSR, and reached the highest value of 3.195 in the fusion images. The information entropy value also increased from 6.24813 in the original images to 7.69423 in the fusion images. The RGB histogram analysis further supported the UIQM and information entropy evaluation results. The original images showed an imbalance in color distribution, particularly because the red channel tended to be weaker, while the green and blue channels were more dominant. After the CLAHE–MSR fusion process, the intensity distribution improved, the red channel increased in the mid-intensity range, and the dominance of the green and blue channels became more controlled. Robustness testing on the DeepFish data showed that the optimal weights obtained using ABC were also able to improve image quality. In mangrove images, the UIQM value increased from 2.053 in the original images to 2.839 with CLAHE, 2.520 with MSR, and reached the highest value of 3.313 in the fusion images. In coral reef images, the UIQM value increased from 2.074 in the original images to 2.488 with CLAHE, 2.200 with MSR, and reached the highest value of 3.311 in the fusion images. The detection performance evaluation showed that image enhancement had a positive impact on the performance of the YOLO-Fish model. On the test data, the CLAHE–MSR fusion method achieved the best performance, with an mAP@0.5 of 86.6%, an increase of 8.9 percentage points compared with the original images at 77.7%, and higher than CLAHE at 84.0% and MSR at 84.7%. For the other metrics, the fusion method achieved a precision of 91,0%, a recall of 82,0%, and an F1-score of 86,0%. The novelty of the proposed research consists of two aspects. First, this study develops a CLAHE–MSR fusion approach based on ABC optimization to improve underwater image quality in seagrass ecosystems. In this approach, the solution is represented as a pair of CLAHE weight (?CL) and MSR weight (?M), which are optimized using UIQM as the fitness function. Second, this study integrates the optimized CLAHE–MSR fusion images into the YOLO-Fish fine-tuning process to evaluate the effect of image enhancement on fish detection performance in seagrass ecosystems with high visual complexity.
       
      URI
      http://repository.ipb.ac.id/handle/123456789/178936
      Collections
      • DF - School of Data Science, Mathematic and Informatics [54]

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository
        

       

      Browse

      All of IPB RepositoryCollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

      My Account

      Login

      Application

      google store

      Copyright © 2020 Library of IPB University
      All rights reserved
      Contact Us | Send Feedback
      Indonesia DSpace Group 
      IPB University Scientific Repository
      UIN Syarif Hidayatullah Institutional Repository
      Universitas Jember Digital Repository