ESTIMASI BERAT BADAN SAPI PADA GAMBAR MONOKULAR MENGGUNAKAN FITUR BERBASIS RASIO TUBUH DAN FULLY CONNECTED NEURAL NETWORK (FCNN)

Apri Anggara Yudha, - and Lala Septem Riza, - and Muhamad Nursalman, - (2025) ESTIMASI BERAT BADAN SAPI PADA GAMBAR MONOKULAR MENGGUNAKAN FITUR BERBASIS RASIO TUBUH DAN FULLY CONNECTED NEURAL NETWORK (FCNN). S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Estimasi berat badan sapi dari data visual menawarkan alternatif tanpa kontak terhadap metode penimbangan tradisional, dengan potensi meningkatkan efisiensi dan kesejahteraan hewan dalam manajemen peternakan. Penelitian ini mengeksplorasi pendekatan tiga tahap yang meliputi segmentasi siluet sapi dari citra RGB, ekstraksi fitur berbasis rasio melalui image processing, dan regresi menggunakan Fully Connected Neural Network (FCNN). Mask hasil segmentasi digunakan untuk menghasilkan fitur berbasis rasio, sementara integrasi dengan fitur Histogram of Oriented Gradients (HOG) juga dievaluasi melalui arsitektur paralel. Model segmentasi menunjukkan performa tinggi dengan nilai MIoU sebesar 0.96, menegaskan keandalannya dalam mengekstraksi siluet sapi. Namun, tahap regresi menghasilkan performa terbatas, di mana model FCNN terbaik tanpa HOG mencapai RMSE 31.82 kg, MAPE 16.51%, dan R² 0.247. Perbandingan dengan beberapa model baseline sederhana menunjukkan hasil yang sebanding, dengan regresi linear sedikit lebih baik, sehingga mengindikasikan keterbatasan utama berasal dari fitur yang digunakan, bukan dari arsitektur model. Penelitian selanjutnya disarankan untuk mengeksplorasi representasi fitur yang lebih informatif, seperti fitur jenis ras sapi, representasi visual langsung dengan CNN, atau fitur morfologi tubuh yang lebih detail guna meningkatkan akurasi estimasi berat sapi berbasis citra. Estimating cattle body weight from visual data offers a contactless alternative to traditional weighing methods, with the potential to improve efficiency and animal welfare in farm management. This study explores a three-stage approach consisting of cattle silhouette segmentation from RGB images, ratio-based feature extraction through image processing, and regression using a Fully Connected Neural Network (FCNN). Segmentation masks were used to generate ratio-based features, while integration with Histogram of Oriented Gradients (HOG) features was also evaluated through a parallel architecture. The segmentation model achieved high performance with an MIoU of 0.96, confirming its reliability in extracting cattle silhouettes. However, the regression stage yielded limited performance, with the best FCNN model without HOG reaching an RMSE of 31.82 kg, MAPE of 16.51%, and R² of 0.247. Comparisons with several simple baseline models showed comparable results, with linear regression performing slightly better, indicating that the primary limitation lies in the features used rather than the model architecture. Future research is recommended to explore more informative feature representations, such as cattle breed, direct visual representations using CNN, or more detailed morphological body features to improve image-based cattle weight estimation accuracy.

	Text S_KOM_2102585_Title.pdf Download (578kB)
	Text S_KOM_2102585_Chapter1.pdf Download (278kB)
	Text S_KOM_2102585_Chapter2.pdf Restricted to Staf Perpustakaan Download (1MB)
	Text S_KOM_2102585_Chapter3.pdf Download (411kB)
	Text S_KOM_2102585_Chapter4.pdf Restricted to Staf Perpustakaan Download (3MB)
	Text S_KOM_2102585_Chapter5.pdf Download (219kB)

Item Type:	Thesis (S1)
Additional Information:	https://scholar.google.com/citations?hl=en&user=kJlGzUYAAAAJ&view_op=list_works ID SINTA Dosen Pembimbing: Lala Septem Riza: 5975668 Muhamad Nursalman: 6143456
Uncontrolled Keywords:	Fully Connected Neural Network (FCNN), Estimasi Berat Sapi, Computer Vision, Machine Learning Fully Connected Neural Network (FCNN), Cattle Weight Estimation, Computer Vision, Machine Learning
Subjects:	Q Science > QA Mathematics Q Science > QA Mathematics > QA75 Electronic computers. Computer science S Agriculture > SF Animal culture
Divisions:	Fakultas Pendidikan Matematika dan Ilmu Pengetahuan Alam > Program Studi Ilmu Komputer
Depositing User:	Apri Anggara Yudha
Date Deposited:	08 Sep 2025 09:37
Last Modified:	08 Sep 2025 09:37
URI:	http://repository.upi.edu/id/eprint/137960

Actions (login required)

View Item