基于卷积神经网络的农作物图像识别方法研究

Research on Crop Image Recognition Method Based on Convolution Neural Network

为了使计算机视觉技术更好地应用于农业自动化领域,本文采用两种识别方法:一种是使用卷积神经网络算法对图像直接进行识别分类;另一种是先对图像进行预处理及分割,然后提取图像的颜色、纹理、形状和内在低维流形特征等特征参数,使用BP神经网络和Elman神经网络算法对提取的特征参数进行识别。通过对两种图像识别方法识别结果的比较,发现卷积神经网络直接对图片进行识别的最高正识率为100%,更适合作为本研究的识别算法。

In order to classify crop images,two recognition methods are used in this paper:one is to directly identify and classify images using convolution neural network algorithm;the other is to preconditioning and segment the images,then extract the color,texture,shape and intrinsic low-dimensional manifold features of the images,and use BP neural network and Elman neural network algorithm to identify the extracted feature parameters.The results reveal that the highest positive recognition rate of convolution neural network is 100%,hence more suitable for this study.