改进的Cifar-10模型在装甲目标二分类中的应用

Application of Cifar-10 Model for Improved in Armor Target Binary Classification

分析研究了Cifar-10模型的网络结构,搭建了实验平台,对传统的模型进行了训练,分析了实验中网络参数对识别效果的影响。提出了增加卷积核数量、合理选取迭代次数和改变激活函数的方法,实现了对传统的Cifar-10模型的改进,实验结果表明,同等实验环境,在合理选取迭代次数和激活函数的前提下,提高卷积核数量1个倍数后对图像的识别率可提高99%以上。在此基础上,利用改进的Cifar-10模型对装甲目标进行了二分类实验,实验结果表明分类准确率达到了99.93%,验证了改进模型在军事应用中的现实意义。

This paper analyzed and studied the network structure of the Cifar-10 model,built an experimental platform,trained the traditional model,and analyzed the influence of network parameters in the experiment on the recognition effect.It put forward the reasonable selection method to increase the number of convolution kernels,reasonably select the number of iterations and change the activation function,which improves the traditional Cifar-10 model.The experimental results show that under the premise of the same experimental environment,the reasonable selection of the number of iterations and activation function.After increasing the number of convolution kernel by one multiple,the recognition rate of image can be increased to more than 99%.On this basis,the improved Cifar-10 model was used to conduct binary classification experiments on armored targets,and the experimental results showed that the classification accuracy reached 99.93%,which verified the practical significance of the improved model in military applications.