Target Intention Recognition with Uncertain Samples Based on Discriminative Parameter Learning and Naive Bayesian Networks

In battlefield environments, target intention recognition based on Bayesian networks often requires parameter learning with uncertain samples. However, existing parameter learning methods for Bayesian networks do not account for the uncertainty information inherent in the samples, which prevents effective parameter learning and reduces its accuracy.. In order to solve this problem, this study proposes a Bayesian network parameter learning method that directly use uncertain samples without loss of sample data information, thereby improving parameter learning accuracy. Firstly, from the perspective of exact Bayesian network inference and combining message propagation inference with discriminative learning, a conditional log-likelihood function under uncertain samples is established as the objective function for parameter learning. To alleviate the overfitting in small-sample scenarios, a norm regularization term for the parameters is constructed based on the principle of maximum entropy. The parameters are then estimated by optimizing the objective function using gradient descent. In experiments on target intention recognition based on naive Bayes classification, the proposed method is compared with other six methods. The results show that the proposed method effectively improves both parameter learning accuracy and target intention recognition performance under uncertain samples. Tests on small sample sets with different sample sizes show that the proposed method consistently achieves higher recognition accuracy than the main comparative methods, indicating that it effectively alleviates the over-fitting problem and enhances the generalization performance of target intention recognition in small-sample settings.