基于卷积神经网络的作战目标识别方法的研究

Research on Target Recognition Method Based on Convolutional Neural Network

在当前复杂多变的战场环境下,采用传统的目标识别手段存在成本高、识别率低、难以快速定位目标等问题,亟需一种成本低、识别率高的识别手段。为此提出一种战场目标识别方法,首先对连续拍摄且拍摄时有抖动的图像进行预处理以降低拍摄图像时抖动产生的影响,之后再基于卷积神经网络(Convolutional Neural Networks,CNN)改进模型实现对战场目标的识别。实验结果表明,基于改进的CNN模型的方法可以取得较高的战场目标识别准确率。

In the complex and changeable battlefield environment, traditional target recognition methods have some problems, such as high cost, low recognition rate and difficult to locate target quickly. Therefore, a low-cost and high recognition method is needed. This paper presents a target recognition method in the battlefield. Firstly, the continuous shooting image and the jittered shooting image are pre-processed to reduce the impact of jitter when captu- ring an image, then based on improved convolutional neural network （CNN） model, the recognition of battlefield tar- gets is achieved. Experimental resuhs show that the method based on improved CNN model can achieve higher recog- nition accuracy of battlefield target than traditional methods.