Recurrent neural network-based remote sensing monitoring method and device for ground surface abnormal phenomenons

Abstract

The invention discloses a recurrent neural network-based remote sensing monitoring method and a device for ground surface abnormal phenomenons. The method comprises the steps: inputting a training front-segment sequence into an encoder constructed based on a bidirectional recurrent neural network, and obtaining a front-segment hidden state sequence for representing the global information of the training front-segment sequence; sequentially inputting the numerical value of the last moment into a decoder based on a single-layer recurrent neural network, so as to sequentially obtain predicted values corresponding to the moments in the training back-end sequence; adopting an encoder-decoder model to predict an observation location at a future moment. By the adoption of the above scheme, an overall change rule of a historical sequence can be mined and observation values at multiple moments in the future can be predicted; near-real-time anomaly monitoring and real-time monitoring and supervising are achieved, meanwhile, required data preprocessing and manually set experience parameters are few, the requirements for experience and professional backgrounds of implementers are not high, themethod is suitable for various geographic areas and land cover types, and the method is high in feasibility, robustness and prediction result accuracy.

Description

technical field [0001] The invention relates to the field of data processing, in particular to a remote sensing monitoring method and device for surface anomalies based on a cyclic neural network. Background technique [0002] Remote sensing time series data records the continuous change process of the earth's surface, and its change pattern over time deeply reflects a variety of natural geographical processes and vegetation phenological rhythms, and has complex nonlinear characteristics of periodicity, trend and randomness. By analyzing the historical remote sensing time series data accumulated over a long period of time in the same area, and mining its internal structure and time-varying laws, it is possible to perform near-real-time monitoring and early warning of various abnormal phenomena caused by human activities or natural disasters. [0003] Existing near-real-time monitoring and early warning of anomalies, the adopted remote sensing time series anomaly detection sc...

AI Technical Summary

Problems solved by technology

[0004] However, both of the above two schemes have disadvantages: on the one hand, the scheme based on time series decomposition assumes that the remote sensing time series conforms to strict periodicity, and can use a linear function to approximate the trend item, which is inconsistent with the real law of surface change; on the other hand, On the one hand, due to the complexity and variety of time-domain change patterns of different land cover types in the prediction-based scheme, it is difficult to find a suitable mathematical model to reflect the internal change law and essential characteristics of the time series, resulting in the gap between the model prediction value and the real value. Large error, the accuracy of anomaly monitoring is not ideal

Therefore, the above two types of methods have their own defects, and their robustness and applicability are not strong.

Images