Weak supervision LSTM recurrent neural network rice field identification method based on time sequence remote sensing data

Abstract

The invention discloses a weak supervision LSTM recurrent neural network rice field identification method based on time sequence multispectral and microwave remote sensing data, and the method comprises the steps: constructing an SAR standard time sequence curve based on ground measured data, carrying out the DTW distance calculation of SAR time sequence remote sensing data, carrying out the sample labeling according to the DTW distance, constructing a high-confidence weak label sample set; and fusing multispectral and SAR input features, training a deep learning classifier based on LSTM by adopting a training set of a weak label sample and an actual measurement sample, and extracting rice classification in a prediction result as a final rice field identification result. The method can be used for pixel-level rice field identification and prediction, high-confidence training data can be obtained through annotation of the DTW distance of the SAR curve, dependence on ground sampling data is reduced, and the ground sampling cost can be effectively reduced.

Description

technical field [0001] The invention relates to the field of crop remote sensing information analysis, in particular to a weakly supervised LSTM cycle neural network paddy field recognition method based on time series remote sensing data. Background technique [0002] Rice is a staple food widely planted in my country, and its yield has an important impact on my country's food security. Remote sensing technology can quickly and accurately dynamically obtain spatial information such as the spectral characteristics of crops, which greatly improves the work efficiency and technological level of agricultural statistics. Crop area identification and yield estimation based on satellite remote sensing technology can not only serve as guidance for actual agricultural production, but also provide an important source of information for global grain trade. Deep learning algorithms have the ability to learn complex spectral and spatial features, and have great potential for application ...

AI Technical Summary

Problems solved by technology

The acquisition of the training set limits the application of the deep learning model to a certain extent, so a weakly supervised LSTM recurrent neural network rice field recognition method based on time series remote sensing data is needed

Images