Weather classification method based on spiking neural network

Abstract

People's life and production activities are directly or indirectly affected by the weather. It is very necessary to accurately and quickly predict weather conditions. At present, the weather prediction system needs a series of sensors and manual assistance, but it cannot be arranged in high density due to high cost, which leads to inaccurate weather prediction. Computer vision technology can classify weather conditions through images, which reduces the cost and can be arranged in high density to ensure the accuracy of weather prediction. Because the training and reasoning of traditional p Convolutional Neural Network has very large energy consumption, while Spiking Neural Network has the characteristics of ultra-low energy consumption, which can further reduce the energy cost. In this paper, a shallow Spiking Neural Network for weather classification is constructed, which is trained and tested on a dataset containing four categories (cloudy, rainy, sunny and sunrise). Experiments show that the classification accuracy of the model is 93.45%, which is higher than that of the Convolutional Neural Network based on vgg19. In addition, the computational complexity of Spiking Neural Network and Convolutional Neural Network are analyzed to show the advantages of Spiking Neural Network in energy consumption.