Severe convection weather recognition algorithm based on convolutional neural network learning

Abstract

The invention relates to the technical field of natural disaster prediction, in particular to a severe convection weather recognition algorithm based on convolutional neural network learning, which comprises the following steps: firstly, collecting a historical weather flow pattern model, then training and learning through a convolutional neural network, and obtaining a recognition model for the historical weather flow pattern through training and learning; secondly, adopting a real-time rainstorm weather flow pattern model through a radar real-time meteorological chart, and then recognizing the real-time weather flow pattern model through the recognition model obtained through training in the first step; judging whether the real-time weather system of the weather flow pattern model identified in the step 2 is located in a monitored key area or not; if yes, outputting a weather flow pattern conforming to geological disasters; and if not, continuing real-time identification. According to the method, real-time recognition is carried out on radar images of severe convection weather, recognition is carried out through a model obtained through convolutional neural network learning and training, and different weather models and disaster models judged in a monitored area can be effectively obtained.

Description

technical field [0001] The invention relates to the technical field of natural disaster prediction, in particular to a strong convective weather recognition algorithm based on convolutional neural network learning. Background technique [0002] Severe convective weather forecasting is still a difficult problem in the world, and it is impossible to determine fixed-point and quantitative data. At present, radar is often used to identify weather clouds, so as to calculate the formation and expansion process of strong convective weather. The work carried out by relevant departments of flood control and people's life and production are of great significance. [0003] In recent years, with the rapid development of pattern recognition, artificial intelligence and image processing technology, the water level recognition method based on image processing has received more and more attention. The issue of image readout is yet to be resolved. The over-complete signal sparse representa...

AI Technical Summary

Problems solved by technology

Although a multi-channel neural network proposed by Zhang Wenda et al. effectively reduces the impact of downsampling on feature extraction, it lacks weight sharing and the parameter design is slightly complicated; Yi Chaoren et al. proposed a multi-channel convolutional neural network recognition method. Random fusion of features from 4 different gradient directions effectively reduces the recognition error rate, but the network structure of this model is relatively complex. Compared with the single-channel convolutional neural network, the training time is nearly doubled, and the time for error backpropagation to be further shortened

Images