基于改进卷积神经网络的图像分类方法

Image Classification Method based on Modified Convolutional Neural Network

为解决传统图像分类算法存在特征提取复杂、训练时间长、分类精度较差等问题,提出一种用于图像分类的改进卷积神经网络方法。首先对卷积神经网络结构、池化运算、激活函数、分类器等方面进行介绍,其次设计了该模型的图像分类算法,最后用TensorFowl对MNIST数据集进行试验,并与传统分类算法进行对比分析。结果表明,将改进卷积神经网络应用于图像分类中,能避免对图像进行复杂的预处理,防止过拟合现象,收敛速度快,分类准确率高达99.22%,其分类性能明显优于其他方法。

In order to solve the problems of traditional image classification algorithms such as complex feature extraction, long training time and poor classification accuracy, a modified convolutional neural network method for image classification is proposed. Firstly, the structures, pooling operation, activation function, classifier, etc. of the convolutional neural network are introduced. Then, the image classification algorithm of the model is designed. Finally, the MNIST dataset is tested with TensorFlow and compared with traditional classification algorithms. The experiment results indicate that the modified convolutional neural network can be applied to image classification, and this can avoid complex preprocessing of images, prevent over-fitting, and achieve fast convergence. Its classification classification performance achieved is significantly better than accuracy rate is as high as 99.22%, and the that by other methods.