Attention Gate with Bidirectional Long Short-Memory for Crop Type Classification Using Multi-Spectral Satellite Images

The satellite sensors are used for the monitoring of the crops based on an automated crop detection framework. The Attention Gate with Bidirectional Long Short-Memory (AG-BiLSTM) is developed for crop type classification using multi-spectral satellite images. The min-max normalization is used for preprocessing which scales every feature into common scale because satellite images have different ranges thereby leading different scales of features. The Principal Component Analysis (PCA) is used for feature extraction which minimizes the data dimensionality that retains required features when minimizing redundancy and noise. It leads enhanced computational effectiveness and performance through concentrates on significant data. The AG-BiLSTM enhance the capability to concentrate on relevant features which improves classification accuracy. The BiLSTM captures contextual data from both previous and future information which enhance the performance. AG-BiLSTM reached 98.61% of accuracy, 97.49% of precision, 96.28% of recall and 96.88% of f1-score for Sentinel-1 dataset compared to 3D Convolution Neural Network (3D-CNN).