Strategic Target Classification with Transfer Learning

Shoulin Yin ； Jie Liu ； Lin Teng

《International Journal of Electronics and Information Engineering》 9卷1期 (2018/09) Pp. 22-28

https://doi.org/10.6636/IJEIE.201809_9(1).03

摘要

Target classification plays an important in target detection, target recognition or target localization. Therefore, finding a common feature representation is crucial to tackle the problem of domain shift where the source domain and target domain have different distribution. Remote sensing technologies develop rapidly, the demand for rapid and accurate extraction of target is higher and higher, especially for military?targets. This paper proposes a transfer learning algorithm for strategic target classification. This work would fine tune a pre-trained network for a new recognition task with faster region convolutional neural network (Faster R-CNN). Then, transfer learning mechanism is adopted to transform the feature extraction in the faster R-CNN training process into strategic target classification. Meanwhile, images are labeled automatically according to the shape, size and texture. A very efficient GPU implementation of the convolution operation is used to make training faster. Finally, we make a model comparison on AlexNet, GoogleNet, and VGG.

關鍵字

AlexNet ； Faster Region Convolutional Neural Network ； Target Classification ； Transfer Learning