基于卷积神经网络的图像识别研究

Research on image recognition based on convolution neural network

为降低图像识别误识率,文章采用卷积神经网络结构对图像进行识别研究。首先,对输入图像进行初始化;然后,初始化后的图像经卷积层与该层中卷积核进行卷积,对图像进行特征提取,提取的图像特征经过池化层进行特征压缩,得到图像最主要、最具代表性的点;最后,通过全连接层对特征进行综合,多次迭代,层层压缩,进而对图像进行识别,输出所识别图像。与原始算法相比,该网络构造可以提高图像识别准确性,大大降低误识率。实验结果表明,利用该网络模型识别图像误识率低至16.19%。

In order to reduce the false recognition rate of image recognition, this paper uses convolution neural network structure to identify the image. First, the input image is initialized; then, the initialized layer of the image is convoluted with the convolution kernel in the layer, and the feature extraction is carried out. The extracted image features are compressed by the pool layer to get the most representative and most representative feature points of the image. The features of the extracted layer are integrated through the full link layer, iterated repeatedly, compressed layer by layer, and then the image is identified and the identified image is output. Compared with the original algorithm, the network structure can improve the accuracy of image recognition and greatly reduce the false recognition rate. The experimental results show that the false recognition rate of the image recognition system using this network model is as low as 16.19%.