Classification of Ice Crystal Habits by Deep Transfer Learning

Abstract

Abstract Ice crystal particle shape is an important factor affecting cloud microphysical processes. Accurately identifying the shapes of ice crystal particle within clouds is a fundamental requirement for calculating various cloud microphysical parameters. In this study, we set up an ice crystal image dataset, encompassing nine distinct habit categories with 8100 images. These images were captured using three probes with varying resolutions: the Cloud Particle Imager (CPI), the Two-dimensional Stereo Probe (2D-S), and the High Volume Precipitation Spectrometer (HVPS). In addition, we introduce a deep convolutional neural network (CNN) based on transfer learning for ice crystal particle shape classification model, TL-AlexNet, which demonstrates the capability to simultaneously classify ice crystal particle habits observed by both the Line Scan Imager and the Area Scan Imager. The results indicate that the TL-AlexNet model could achieve superior performance in ice crystal shapes classification for two types of imagers, and the classification with the accuracy of 97.16%. It is much higher than the traditional shape recognition methods, and has certain application value for Climate and cloud microphysics research.