Search Results

You are looking at 1 - 2 of 2 items for :

  • Author or Editor: B. B. Hicks x
  • Journal of Atmospheric and Oceanic Technology x
  • Refine by Access: All Content x
Clear All Modify Search
A. Hicks and B. M. Notaroš


Taking advantage of the recent developments in machine learning, we propose an approach to automatic winter hydrometeor classification based on utilization of convolutional neural networks (CNNs). We describe the development, implementation, and evaluation of a method and tool for classification of snowflakes based on geometric characteristics and riming degree, respectively, obtained using CNNs from high-resolution images by a Multi-Angle Snowflake Camera (MASC). These networks are optimal for image classification of winter precipitation particles due to their high accuracy, computational efficiency, automatic feature extraction, and application versatility. They require little initial preparation, enable the use of smaller training sets through transfer learning techniques, come with large supporting communities and a wealth of resources available, and can be applied and operated by nonexperts. We illustrate both the ease of implementation and the usefulness of operation the CNN architecture offers as a tool for researchers and practitioners utilizing in situ optical observational devices. A training dataset containing 1450 MASC images is developed primarily from two storm events in December 2014 and February 2015 in Greeley, Colorado, by visual inspection of recognizable snowflake geometries. Defined geometric classes are aggregate, columnar crystal, planar crystal, small particle, and graupel. The CNN trained on this dataset achieves a mean accuracy of 93.4% and displays excellent generalization (ability to classify new data). In addition, a separate training dataset is developed by sorting snowflakes into three classes and showcasing distinct degrees of riming. The CNN riming degree estimator yields promising initial results but would benefit from larger training sets.

Full access
C. Key, A. Hicks, and B. M. Notaroš


We present improvements over our previous approach to automatic winter hydrometeor classification by means of convolutional neural networks (CNNs), using more data and improved training techniques to achieve higher accuracy on a more complicated dataset than we had previously demonstrated. As an advancement of our previous proof of concept study, this work demonstrates broader usefulness of deep CNNs by using a substantially larger and more diverse dataset, which we make publicly available, from many more snow events. We describe the collection, processing, and sorting of this dataset of over 25 000 high-quality Multi-Angle Snowflake Camera (MASC) image chips split nearly evenly between five geometric classes: aggregate, columnar crystal, planar crystal, graupel, and small particle. Raw images were collected over 32 snowfall events between November 2014 and May 2016 near Greeley, Colorado, and were processed with an automated cropping and normalization algorithm to yield 224 × 224 pixel images containing possible hydrometeors. From the bulk set of over 8 400 000 extracted images, a smaller dataset of 14 793 images was sorted by image quality and recognizability (Q&R) using manual inspection. A presorting network trained on the Q&R dataset was applied to all 8 400 000+ images to automatically collect a subset of 283 351 good snowflake images. Roughly 5000 representative examples were then collected from this subset manually for each of the five geometric classes. With a higher emphasis on in-class variety than our previous work, the final dataset yields trained networks that better capture the imperfect cases and diverse forms that occur within the broad categories studied to achieve an accuracy of 96.2% on a vastly more challenging dataset.

Restricted access