Application of deep learning algorithms considering spatio-temporal features for crop classification

초록

The purpose of this study is to compare deep learning models that consider characteristics of crops in the classification of multi-temporal and high spatial resolution images. We applied 2D-convolutional neural network (2D-CNN) and long short-term memory (LSTM) for crop classification to consider spatial and temporal features, respectively. In addition, 3D-CNN and convolutional LSTM (Conv-LSTM), which can account for both temporal and spatial features, were also applied and compared. From a case study of crop classification with multitemporal unmanned aerial vehicle images, Conv-LSTM showed the best classification accuracy thanks to its ability to account for both spatial and temporal features. Since the growth cycles of crops should be properly considered for crop classification, LSTM-based models including LSTM and Conv-LSTM are more effective than CNN-based models. © 2020 40th Asian Conference on Remote Sensing, ACRS 2019: Progress of Remote Sensing Technology for Smart Future. All rights reserved.

키워드