基于改进的卷积神经网络图像识别方法

当前的图像特征识别大多采用的是传统的机器学习方法与卷积神经网络方法。传统的机器学习对图像识别的研究,特征提取多是通过人工完成,泛化能力不够强。最早的卷积神经网络也存在诸多缺陷,如硬件要求高,需要的训练样本量大,训练时间长。针对以上问题,提出了一种改进的神经网络模型,在LeNet-5模型的基础上并在保证识别率的情况下,简化网络结构,提高训练速度。将改进的网络结构在MINIST字符库上进行识别实验,分析网络结构在不同参量中的识别能力,并与传统算法进行对比分析。结果表明提出的改进结构在当前识别正确率上,明显高于传统的识别算法,为当前的图像识别提供新的参考。

Research on Image Recognition Method Based on Improved Convolution Neural Network

Most of the current image feature recognition methods use traditional machine learning methods and convolution neural network methods.In traditional machine learning research on image recognition, feature extraction is mostly done manually, and generalization ability is not strong enough.The earliest convolution neural network also has many defects, such as high hardware requirements, large training samples and long training time. To solve the above problems, an improved neural network model was proposed, which simplifies the network structure and improves the training sped on the basis of LeNet-5 model and under the condition of ensuring the recognition rate. The improved network structure was tested on MINIST character database to analyze the recognition ability of the network structure in different parameters and compared with the traditional algorithm. The results show that the improved structure proposed in this paper is significantly higher than the traditional recognition algorithm in the current recognition accuracy, providing a new reference for the current image recognition.