RIZKY, MUHAMMAD (2026) KLASIFIKASI KESEGARAN IKAN MULTI-SPESIES MELALUI CITRA MATA MENGGUNAKAN FEATURE FUSION CNN–GLCM. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
01 Cover.pdf Download (592kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
02 Bab 1.pdf Restricted to Registered users only Download (102kB) |
|
|
Text (BAB II)
03 Bab 2.pdf Restricted to Registered users only Download (408kB) |
|
|
Text (BAB III)
04 Bab 3.pdf Restricted to Registered users only Download (233kB) |
|
|
Text (BAB IV)
05 Bab 4.pdf Restricted to Registered users only Download (866kB) |
|
|
Text (BAB V)
06 Bab 5.pdf Restricted to Registered users only Download (102kB) |
|
|
Text (DAFTAR PUSTAKA)
07 Daftar Pustaka.pdf Restricted to Registered users only Download (103kB) |
|
|
Text (LAMPIRAN)
08 Lampiran.pdf Restricted to Registered users only Download (1MB) |
Abstract
Fish freshness assessment is an important aspect of maintaining food quality and consumer safety. However, manual fish freshness assessment still relies on subjective visual observation, which may lead to inconsistent evaluations. This study aims to develop a multi-species fish freshness classification model based on fish eye images using a CNN–GLCM feature fusion approach. The dataset used in this study was obtained from the Freshness of the Fish Eyes (FFE) dataset. In the main experiment, this study used three fish species, namely Eleutheronema tetradactylum, Johnius trachycephalus, and Upeneus moluccensis, with three freshness levels: Highly Fresh, Fresh, and Not Fresh. This study applied a nineclass classification scenario based on the combination of fish species and freshness levels. The proposed approach utilized RGB images as visual input and Gray Level Co-occurrence Matrix (GLCM) texture features as additional input to strengthen the representation of texture characteristics in fish eye images. In the main experiment, three models were compared: Custom VGG Baseline CNN, Residual CNN, and Depthwise Separable CNN. The main experiment showed that the Custom VGG Baseline CNN achieved the best performance with a test accuracy of 87.67%, precision of 88%, recall of 88%, F1-score of 87%, and test loss of 0.3535. This study also conducted an additional experiment as a comparative analysis using a different data processing pipeline, in which the dataset was split before augmentation and augmentation was applied only to the training data. The additional experiment used a different species subset from the same dataset source, namely Chanos chanos, Nibea albiflora, and Rastrelliger faughni. In the additional experiment, Residual CNN + GLCM achieved the best performance with a test accuracy of 71.65%, test loss of 0.7867, macro F1-score of 70.54%, and weighted F1-score of 71.60%. The results indicate that the CNN–GLCM feature fusion approach can be used for multi-species fish freshness classification based on fish eye images. However, model performance is influenced by the architecture, amount of training data, augmentation pipeline, and evaluation data conditions. Kata kunci: fish freshness, fish eye image, multi-species, CNN, GLCM, feature fusion. Penentuan kesegaran ikan merupakan aspek penting dalam menjaga kualitas pangan dan keamanan konsumsi. Namun, proses penilaian kesegaran ikan secara manual masih bergantung pada pengamatan visual yang bersifat subjektif, sehingga berpotensi menghasilkan penilaian yang tidak konsisten. Penelitian ini bertujuan untuk mengembangkan model klasifikasi kesegaran ikan multi-spesies melalui citra mata menggunakan pendekatan feature fusion CNN–GLCM. Dataset yang digunakan berasal dari Freshness of the Fish Eyes (FFE). Pada eksperimen utama, penelitian ini menggunakan tiga spesies ikan, yaitu Eleutheronema tetradactylum, Johnius trachycephalus, dan Upeneus moluccensis, dengan tiga tingkat kesegaran, yaitu Highly Fresh, Fresh, dan Not Fresh. Penelitian ini menggunakan skenario klasifikasi sembilan kelas gabungan berdasarkan kombinasi spesies dan tingkat kesegaran. Pendekatan yang digunakan memanfaatkan citra RGB sebagai masukan visual dan fitur tekstur Gray Level Co-occurrence Matrix (GLCM) sebagai masukan tambahan untuk memperkuat representasi karakteristik tekstur pada citra mata ikan. Pada eksperimen utama, model yang dibandingkan terdiri dari Custom VGG Baseline CNN, Residual CNN, dan Depthwise Separable CNN. Hasil eksperimen utama menunjukkan bahwa Custom VGG Baseline CNN memperoleh performa terbaik dengan test accuracy sebesar 87,67%, precision sebesar 88%, recall sebesar 88%, F1-score sebesar 87%, dan test loss sebesar 0,3535. Penelitian ini juga melakukan eksperimen tambahan sebagai analisis pembanding dengan pipeline pengolahan data yang berbeda, yaitu pembagian dataset dilakukan sebelum proses augmentasi dan augmentasi hanya diterapkan pada data latih. Eksperimen tambahan menggunakan subset spesies berbeda dari sumber dataset yang sama, yaitu Chanos chanos, Nibea albiflora, dan Rastrelliger faughni. Pada eksperimen tambahan, Residual CNN + GLCM memperoleh performa terbaik dengan test accuracy sebesar 71,65%, test loss sebesar 0,7867, macro F1-score sebesar 70,54%, dan weighted F1-score sebesar 71,60%. Hasil penelitian menunjukkan bahwa pendekatan feature fusion CNN–GLCM dapat digunakan untuk klasifikasi kesegaran ikan multi-spesies melalui citra mata. Namun, performa model dipengaruhi oleh arsitektur yang digunakan, jumlah data latih, tahapan augmentasi, serta kondisi data evaluasi. Kata kunci: kesegaran ikan, citra mata ikan, multi-spesies, CNN, GLCM, feature fusion.
Actions (login required)
 |
View Item |
