Near real-time monitoring of cassava cultivation area
No hay miniatura disponible
Fecha
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
Descripción
Remote sensing technologies and deep learning/machine learning approaches play valuable roles in crop inventory, yield estimation, cultivated area estimation, and crop status monitoring. Satellite-based remote sensing has led to increased spatial and temporal resolution, leading to a better quality of land-cover mapping (greater precision, and detail in the number of land cover classes). In this work, we propose to use a long short-term memory neural network (LSTM), an advanced technical model adapted from artificial neural networks (ANN) to estimate cassava cultivation area in southern Laos. LSTM is a modified version of a Recurrent Neural Network (RNN) that uses internal memory to store the information received prior to a given time. This property of LSTMs makes them advantageous for time series regression. We employ Landsat-7/8 and Sentinel-2 time-series datasets and crop phenology information to identify and classify cassava fields using multi-sources remote sensing time-series in a highly fragmented landscape. The results indicate an overall accuracy of > 89% for cassava and > 84% for all-class (barren, bush/grassland, cassava, coffee, forest, seasonal, and water) validating the feasibility of the proposed method. This study demonstrates the
potential of LSTM approaches for crop classification using multi-temporal, multi-sources remote sensing time series.
Palabras clave
cassava, forest cover, remote sensing, machine learning, conservación de la naturaleza, satélites de observación terrestre, mandioca-yuca