A Real-time Deep Convolution Image Recognition Method Based on Data Mining
Shuai Wang, Jinyuan Shen, Runjie Liu · 2020
The traditional convolution neural networks(CNN) usually have slow target recognition speed due to its complex models and computational complexity. To improve the recognition speed, two mining data methods are employed to optimize the deep convolution recognition structure in this paper. Features of image which are obtained by deep CNN are firstly screened based on the dispersion ratio algorithm then genetic algorithm is adopted to get less useful features. These data mining methods greatly reduce the feature dimensions so that the recognition computations are reduced tremendously. To verify this idea, synthetic aperture radar (SAR) images are recognized by the proposed models. The correct recognition rate and the complexities of the proposed method are analyzed and compared to the traditional CNN. The experimental results show that the correct recognition rate gets better and the complexities reduce immensely so the target recognition speed is greatly improved.