A Method for Assessing Battlefield Situation of Mission Planning Agent Based on Improved TextCNN

Autonomous unmanned systems generate action plans through the mission planning agent. Realizing real-time battlefield situation assessment helps the agent to regulate or replan the battle plan in real time. Improving the agent’s situation assessment capability can enhance the autonomy and intelligence of the unmanned system’s autonomous mission planning. Therefore, this paper proposes a battlefield situation assessment method based on an improved text convolutional neural network (TextCNN). First, a multi-frame sequence situation encoding perception method is constructed, which incorporates both the temporal characteristics of the situation and the spatial distribution characteristics of combat units. Then, an improved TextCNN algorithm for situation assessment is developed, which uses a multi-hidden layer residual network to improve the learning capability and feature information transfer efficiency of the evaluation model. Experimental validation reveals several significant results. The convergence and accuracy of the evaluation using multi-frame sequence coded data surpasses that of situation data lacking temporal dimensions. The improved TextCNN shows a 2.2% accuracy improvement over the original TextCNN model, and its evaluation is better than that of SVM, SAE, MLP, LSTM networks. In conclusion, the proposed battlefield situation assessment method, which integrates multi-frame sequence coding and an improved TextCNN network, can effectively assess the battlefield situation.