Phase Recognition of Supercooled Droplet Based on Convolutional Neural Network

Abstract

The supercooled droplet in the cloud is an important research object in the field of cloud precipitation physics and artificial weather modification or in the field of aircraft icing. The measurement of its microphysical characteristics currently mainly depends on aircraft observations. The high-resolution cloud particle imager is currently an important instrument in the field of airborne cloud particle imaging and observation, but this instrument only images the particles in the cloud and does not recognize the shape of the particles. By processing the original multi-particle image data taken by the high-resolution cloud particle imager, a single-particle gray-scale image data set is created. A three-layer convolutional neural network model is constructed and trained, and it is applied to the recognition of supercooled droplets. Experiments show that the phase recognition based on convolutional neural network has a high accuracy rate and is better than existing recognition methods. It can be applied to cloud precipitation physics and artificial weather influencing and aircraft icing research.