Flower Image Classification Based on Multi-scale Dense Residual Network

Aiming at the problem of insufficient feature utilization, small convolution layer width and single convolution scale in traditional residual network in flower image classification, a flower image classification algorithm based on multi-scale dense residual network proposes. The first layer of the algorithm uses a convolution group composed of multi-scale convolutions, and then replaces the single-path convolution in the residual block with a multi-path parallel convolution group composed of different numbers of convolutions, and finally add dense connections between the difference block to strengthen the feature flow of different residual blocks of the network and enhance the feature utilization of the network. The experimental results show that the accuracy of this algorithm in flower classification is 87.91%, it is higher than the traditional residual network.