Find in Library

Deep convolutional neural network model has achieved excellent results in many computer vision applications. How to use deep learning technology to complete auxiliary guidance and aiming points positioning in complex battlefield environment is the key for army to win the modern information war. To solve this problem, this paper proposes a recognition algorithm for types and aiming point of the time-sensitive target to improve the quality of time-sensitive target detection, and provides the strike value of various components of military resources for subsequent modules. This algorithm redesigns the YOLOv3 trunk network and uses the residual block of the depth-wise separable convolutional neural network to extract the features of the input image, then sents the obtained feature maps into the attention model, and assigns corresponding weights to feature maps with important semantic information such as target components. Finally, feature maps processed by the attention model is sent into the regression network for the recognition of time-sensitive target types and aiming points. The experimental results on COCO and VOC data sets show that the feature extraction network and attention module used in this algorithm effectively improve the mean average precision of deep convolutional neural network in common target detection. The experimental results on the data set of the battlefield military resource model established in this paper show that this algorithm can accurately recognize the aiming points of non-cooperative time-sensitive targets.