Semantic Segmentation of Sentinel-2 Satellite Image for Rice Growth Phase Classification Using Deep Learning

Sentinel-2 satellite image is one of the big data for remote sensing. This research explores the usage of U-Net model for segmentation of rice fields and CNN Model for classification of rice growth phase on SentineL-2 L2A satellite image. This study investigates the use of the U-net method for segmentation. Rice growth phases are classified using CNN model. The results show that CNN model can work better for various computer vision tasks, including preprocessing of existing satellite data for rice growth phases. For the suggested semantic segmentation model, the accuracy and IoU score values for the proposed semantic segmentation model were 0.99 and 0.66, respectively. Conversely, the suggested CNN classification model for rice growth phase, the accuracy was 0.86, while specificity was 0.97 and sensitivity was 0.85. The goal of this research is to establish the viability of such a system and to pave the way for future research including rice field extraction and rice growth phase classification using satellite imagery. The segmentation and classification model has shown excellent performance. However, the model still struggles to accurately identify the correct pixels belonging to a specific mask for segmentation or correctly recognize the right class for a specific group of area based on the spectral indices, as evidenced by a few samples in the results.