A flower recognition system based on MobileNet for smart agriculture

Abstract

Flower identification is the basic research in many fields, such as botany research, so the study of flower identification has far-reaching significance. The traditional flower recognition has the characteristics of weak generalization ability, high cost, and low accuracy. Because of the above problems, this paper proposes to use a lightweight neural network MobileNet model based on CNN to recognize flowers. The depth-separable convolution model reduces the number of parameters by dividing the convolution into Depthwise Convolution and Pointwise Convolution. That is to say, compared with the traditional convolutional neural network, the number of layers of the separable convolutional neural network can be deeper in the case of the same number of parameters. In this experiment, the CNN and MobileNet neural network models are trained by using public data sets, and the final experimental results show that the correct rate of the MobileNet model reaches 88. 37%, which is 22. 3% higher than that of the traditional CNN network model.