基于卷积神经网络和随机森林的毁伤效果评估方法

Damage Effect Evaluation Method Based on Convolution Neural Network and Random Forest

目标毁伤效果评估是现代化战争中的重要一环。针对传统的毁伤效果评估方法无法区分目标特征与背景特征而导致评估结果不准确的问题,提出了卷积神经网络(convolutional neural network,CNN)和随机森林(random forest,RF)相结合的方法,记为CNN-F算法。通过卷积神经网络处理图像,提取图像特征,再使用随机森林替换卷积神经网络中的部分全连接层和softmax分类器进行目标毁伤结果分类。实验结果表明,该算法在准确度、精确度、召回率和F1值4个指标上都达到了较高的水平,达到了83.050%、83.585%、83.050%和82.945%,其评估结果可以为指挥员下一步决策提供参考。

Target damage effect evaluation is an important link of modernization war.Aiming at the problem that the traditional damage effect evaluation method can't distinguish the target characteristics from the background characteristics,resulting in the inaccurate evaluation results,a method combining convolution neural network(CNN)and random forest(RF)is proposed,which is recorded as CNN-F algorithm.The image is processed by convolution neural network,and the image features are extracted.Then the random forest is used to replace part of the full connection layer and softmax classifier in convolution neural network to classify the target damage results.The experimental results show that the algorithm achieves a higher level in accuracy,precision,recall rate and F1 value,reached 83.050%,83.585%,83.050%and 82.945%.The evaluation results can provide a reference for the commander's next decision-making.