Initial Investigation into the Effect of Image Degradation on the Performance of a 3-Category Classifier Using Transfer Learning and Data Augmentation

摘要

This paper documents an initial investigation into the effect of image degradation on the performance of transferlearning (TL) as the number of retrained layers is varied, using a well-documented, commonly-used, and well-performing deep learning classifier (VGG16). Degradations were performed on a publicly-available data set tosimulate the effects of noise and varying optical resolution by electro-optical (EO/IR) imaging sensors. Perfor-mance measurements were gathered on TL performance on the base image-set as well as modified image-setswith different numbers of retrained layers, with and without data augmentation. It is shown that TL mitigatesagainst corrupt data, and improves classifier performance with increased numbers of retrained layers. Data aug-mentation also improves performance. At the same time, the phenomenal performance of TL cannot overcomethe lack of feature information in severely degraded images. This experiment provides a qualitative sense ofwhen transfer learning cannot be expected to improve classification results.