Klasifikasi Tanaman Menggunakan Metode Deep Learning Residual Network (ResNet) Berbasis Data Time Series Penginderaan Jauh di Desa Girimulyo, Lampung Timur

Rahadianto, Muhammad Ario Eko
Sejati, Putri Wahyu
Fauzi, Adam Irwansyah
Atmojo, Aulia Try
Widayanti, Tika
Yudanegara, Rizky Ahmad

Jurnal Fisika Unand
Vol 15 No 1 (2026)

Most Indonesians work in agriculture, making crop-type maps essential for food security. This study evaluates time-series classification using Residual Network (ResNet) for crop mapping. Sentinel-2A imagery from May 2021 to May 2022 was used with 120 samples across five classes: Corn, Coconut, Non-crop, Banana, and Other Crops. The data were processed into a regularized Earth Observation (EO) data cube and trained using samples filtered with Self-Organizing Map (SOM) under two schemes: single clustering (SC) and double clustering (DC). The ResNet model was trained with filtered data and tested with varying epochs. The study produced a crop-type map of Girimulyo, East Lampung, smoothed with the Bayesian method. Accuracy assessment showed that SC at 100 epochs achieved 87%, exceeding the 85% threshold, while DC yielded lower accuracy due to reduced training data. These results confirm that ResNet-based time-series classification is effective for crop-type mapping in the study area.