Abstract
Collaborative applications of physical systems and algorithms bring the rapid development of cyber physical systems (CPS). Establishing CPS with image classification systems, however, is difficult, because both categories of algorithms, deep learning methods and traditional feature extraction methods, are independent and individual currently. Therefore, in this paper, we propose a fast feature fusion algorithm to satisfy the requirement of CPS in the area of image classification from a comprehensive perspective. First, we fuse the shallow-layer network feature, large pre-trained convolutional neural network feature and traditional image features together by genetic algorithm, in order to guarantee high accuracy with little training time and hardware cost. Second, we increase the distance between different centers by dynamic weight assignment to improve distinguishability of different classes. Third, we propose a partial selection method to reduce the length of the fused feature vectors and to improve the classification accuracy further by centralizing the features within the same class. Finally, experimental results show that our method can achieve high classification accuracy with lower training time and hardware consumption, which can greatly improve the efficiency and flexibility of image classification in cyber physical systems.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have