Tree height parameter prediction method based on multi-scale spatio-temporal double-flow network

Abstract

The application discloses a tree height parameter prediction method based on a multi-scale space-time double-flow network, and belongs to the application of remote sensing technology and deep learning in the field of forestry resource management. The method processes multi-modal data in parallel through independent optical flow and radar flow branches, extracts local features by using multi-scale one-dimensional convolution, filters noise by combining a gate enhanced attention mechanism, captures time sequence dependence by a bidirectional LSTM, and finally fuses features to perform tree height regression prediction. The application significantly improves the precision and robustness of tree height parameter prediction, effectively overcomes the spectral saturation phenomenon in high canopy density forest areas, and provides an efficient solution for global climate change research, carbon sink estimation and fine forestry management.

Images