基于混合卷积神经网络的花卉种类识别研究

Flower Species Recognition Based on Hybrid Convolution Neural Network

花卉种类识别属于植物分类的重要分支之一,有着很高的研究和应用价值。但是,目前的花卉识别方法普遍存在着时间长、准确率较低的问题。针对这些问题,本文以花卉图像作为研究对象,首先选择Gabor滤波器对花卉图像进行纹理分析,然后采用改进后的LeNet-5和GoogLeNet神经网络模型对处理后的花卉图片分别进行特征提取,并对得到的两个不同的特征向量进行特征融合,以此进行花卉种类的识别和模型训练。为了验证模型的优劣,使用Tensorflow框架对实验进行仿真,基于混合卷积神经网络无论在泛化能力还是拟合能力上都有显著的提高。

In order to avoid time consuming and low accuracy of the present flower recognition method,this paper presents a method based on hybrid convolution neural network.It first chooses Gabor filter to analyze the flower image texture,and then uses the improved LeNet-5 and GoogLeNet neural network model to extract the feature of the processed flower pictures respectively.Secondly,two obtained different eigenvectors are fused to achieve the process of recognition of flower species and the model training.Finally,the Tensorflow framework is adopted to simulate the experiments.The results show that both generalization ability and fitting ability have been significantly improved.