Convolutional long short-term memory network space-time sequence prediction method improved by utilizing attention mechanism

Abstract

The invention discloses a convolutional long-term and short-term memory network space-time sequence prediction method improved by utilizing an attention mechanism. The method relates to the field of computer prediction, and specifically comprises the following steps: (1) extracting spatial features by an asymmetric convolution block high-dimensional feature extractor; (2) a ConvLSTM encoder decoder architecture embedded with an attention module predicting extrapolation features; (3) reversely reconstructing a feature result; (4) L1 and L2 regularization optimization algorithms being carried out; and (5) predicting a space-time sequence image. According to the method, high-dimensional features of spatio-temporal sequence data can be well extracted through the multi-layer convolutional neural network, and the high-dimensional features are used as input of the model, so that the problem that high dimensions cannot be calculated is solved, and spatial key information is emphasized; according to the improved ConvLSTM, the spatial and temporal features can be better learned to realize more accurate extrapolation; the method is suitable for all sequential images.

Description

technical field [0001] The invention relates to the field of computer prediction, in particular to an improved convolutional long-short-term memory network spatio-temporal sequence prediction method using an attention mechanism. Background technique [0002] Time-space sequence prediction is an image extrapolation technology based on deep learning. It predicts future M-frame images based on the previous N-frame image sequence. Video prediction, human action prediction, robotics and other fields have become research hotspots in computer vision, but the current technology still has great limitations; on the one hand, when the target changes rapidly, it should be based on nearby frames instead of Generate future images on distant frames, which requires the predictive model to learn short-term video dynamics; on the other hand, when moving objects in the scene are frequently entangled, it is difficult to separate them into the generated frames, which requires the predictive mode...

AI Technical Summary

Problems solved by technology

[0005] The invention discloses a time-space sequence prediction method of convolutional long-term short-term memory network improved by using attention mechanism, and constructs a sequence-to-sequence model to solve the shortcoming of too much forgetting long-term information in the time-space sequence prediction process; attention The mechanism can judge the attention needs of the current step according to the output of the previous step, so as to learn to emphasize important data or suppress unimportant data; the attention mechanism designed in this method is a neural network, which can be well embedded Into ConvLSTM, and all the hidden states of the encoder and the hidden state of the previous step of the decoder are used as input, which overcomes the disadvantage of the ConvLSTM encoder-decoder compressing all input sequences into a fixed-size vector; in addition, the attention mechanism "Selectivity" establishes weights and mappings for sequence data, so that input and output can be linked, avoiding the "black box" of the algorithm

The beneficial effect of the present invention is: the high-dimensional features of time-space sequence data can be well extracted through the multi-layer convolutional neural network, and as the input of the model, it not only solves the problem of high-dimensional incomputability, but also emphasizes the key information of space

Images