Ensemble Learning Based Multi-Color Space in Convolutional Neural Network

Abstract

Convolutional neural networks is a common mean of accomplishing image classification tasks in recent years. The input of the existing networks are in single color space (RGB color space). In this paper, we propose an ensemble learning based multi-color space in the convolutional neural network, which can combine the advantages of multiple color spaces on the image. In addition, the color space conversion process can bring more nonlinear components to the network, which can increase the effectiveness of solving real-world classification tasks. Moreover, this article optimizes the method mentioned, so that the parameter quantity and calculation amount of the final network model is basically maintained at the original scale, and the accuracy rate is similarly improved. We conduct comparative experiments and show that ensemble learning based multi-color space in convolutional neural network achieves better performance than the original network.