BPUM: A Bayesian Probabilistic Updating Model Applied to Early Crop Identification

Authors:
Publisher website
Google Scholar
Abstract
Accurately predicting crop cultivation information in the early stages is important for national food security decision-making. However, due to limited time-series observation, early crop mapping is a difficult task. The existing works focus only on feature modeling, relying on uncertain time-series observations, which have been proved not to be a promising direction. Crop cultivation has a regular and cyclical pattern, which could be used to guide crop identification for the upcoming year. Building upon this, a Bayesian probabilistic updating model (BPUM) is proposed for early crop identification. The key of BPUM is iteratively optimizing the crop cultivation probability based on all possible knowledge and observations. Firstly, historical cultivation knowledge can be modeled by estimating the prior probability distribution. Meanwhile, BPUM designs to integrate prior probability and new stage observation. Furthermore, every new stage observation could contribute to updating this prior probability distribution. With the increase in observations, the intelligence of the model can be enhanced. Experiments were conducted in 2 study areas with different climatic conditions. The results indicate that this approach can identify crops 1 to 2 months earlier than traditional methods with overall accuracy of 94.66% and 96.00% at these areas and is applicable to various agricultural regions, demonstrating good stability and applicability.
Funding Information
  • The National Key R&D Program of China (2022YFB3903402)
  • National Natural Science Foundation of China (42222106)
  • National Natural Science Foundation of China (61976234)

This publication has 68 references indexed in Scilit: