Rice Growth Monitoring Based on 3D-CNN Satellite Remote Sensing Data

EasyChair Preprint no. 14005

Abstract

Rice is the most important food crop in China. Timely and accurate acquisition of large-area rice planting area and growth information is of great significance to China’s food security. Satellite remote sensing can sensitively respond to the development of large-area rice plants and changes in soil moisture, and is an important means of rice growth monitoring. Remote sensing research on rice growth monitoring mainly focuses on three major focuses: extraction of rice planting area, inversion of rice physiological parameters, and identification of rice maturity, and the research methods mainly include mathematical analysis, machine learning, and multi-source collaboration. In response to the problem that traditional remote sensing vegetation indices are difficult to accurately monitor large-area rice growth and have low accuracy, this study proposes a 3D-CNN neural network model that uses hyperspectral satellite remote sensing data sets to monitor the growth of rice in China’s main rice planting areas. The model uses three-dimensional convolutional neural networks (3D-CNN) and temporal convolutional neural networks (TCN) to process the spatiotemporal information and spectral information of rice satellite remote sensing images. The model first extracts spatiotemporal features through multiple three-dimensional convolutional layers, then performs spatial feature analysis by compressing and reducing the dimensionality of the output features through two-dimensional convolutional layers, and finally, the high-level feature maps are flattened and category prediction is performed through fully connected layers. At the same time, a new loss function is introduced in the neural network model to eliminate the impact of the imbalance of rice yield label distribution. Finally, the new model is verified through the prediction of rice yield data in China. The results are compared with the main deep learning methods in use.

Keyphrases: 3D CNN, chlorophyll, growth monitoring, hyperspectral remote sensing, Rice

@Booklet{EasyChair:14005,
  author = {Yindong Li and Yang Chen and Wang Wang},
  title = {Rice Growth Monitoring Based on 3D-CNN Satellite Remote Sensing Data},
  howpublished = {EasyChair Preprint no. 14005},

  year = {EasyChair, 2024}}