152 / 2019-07-12 21:24:41

基于深度学习的空间近距离目标特征库构建和智能识别

空间近距离目标; 探测识别; 人工智能

采用传统方法开展空间近距离目标特征识别，存在目标选择策略无针对性，难以适应目标多样变化等问题。而通过深度学习识别空间近距离目标或者其相关部件，也存在难以获得满足模型训练需要的大量图像集的问题。本文通过分析已有目标的特征信息，获取全目标和相关部件的三维模型，采用三维仿真技术生成目标的相关部件图像，通过部件间组合形成目标，标注图像信息辅助训练，采用基于深度学习的目标特征智能识别方法进行识别，克服训练样本数量少的缺点，提升目标特征感知准确性。
There are some problems by using traditional method to carry out space short-range target feature recognition, , such as the target selection strategy with no strong points and difficulty to adapt to the variety of targets. It is also difficult to obtain a large number of images to meet the needs of model training through deep learning to identify short-range target or their related components. The paper analyses the feature information of the existing target, generates the relevant component images of the target by the three-dimensional simulation technology, identify targets by the target feature intelligent recognition method based on deep learning, overcome the shortage of training samples, and improve the accuracy of target feature perception.