(0721) 8030188    [email protected]   

KLASIFIKASI CITRA PENGINDERAAN JAUH BERBASIS TIME SERIES MENGGUNAKAN RESIDUAL NETWORK (RESNET) UNTUK PEMETAAN JENIS TANAMAN (Studi Kasus : Desa Girimulyo, Lampung Timur)


Most of Indonesia's population works in the agricultural sector. Map types of plants are needed for various types of food. Remote sensing is a technique that can be used to classify plant species and produce land cover information in the form of an effective and efficient plant classification map. This study aims to look at the results of time series classification using the Residual Network (ResNet) method for classifying plant species maps. This study used time series data from the Sentinel2A satellite with an observation range of 5 May 2021 to 5 May 2022, producing 120 sample data in the form of the coordinates of five types of plants, namely the classes of Corn, Coconut, Non-plants, Bananas and Other Plants. Time series data is used as input in making (EO) a regularized data cube, then sample data is used as input for training and validation. In the training, 120 sample data were used, then the quality or clustering was checked using the Self-Organizing Map (SOM) method, then a filtering process was carried out with two experimental schemes, namely single clustering (SC) and double clustering (DC). The next step is to train the model using the ResNet method, using filtered training data and regularized data cubes. A number of epochs are applied to obtain optimal epoch values. The results of the study were in the form of a Map of Plant Types in the Girimulyo Region, East Lampung, which were smoothed using the Bayesian method and then tested for accuracy. The accuracy of the SC scheme at epoch 100 which has reached 87% proves that the ResNet-based time series classification is effective for mapping plant species in the study area. The experiment with the DC scheme was not better than the SC scheme because of the clustering and reduced amount of training data, as well as an increase in the number of uncertain epochs. Keywords: Remote Sensing, Time Series Based Classification, Satellite Image Time Series (SITS), ResNet.

URI
https://repo.itera.ac.id/depan/submission/SB2304050005

Keyword