【摘要】 声纳目标特征级融合的主要任务是实现信息压缩、目标身份确定(分类) ,以利于实时处理、决策分析。基于数学模型的各种算法,由于情况复杂,往往很难建立。而人工神经网络通过样本的学习,具有存储记忆、在相似输入下能恢复记忆等特性,从而避免了烦琐而复杂的建模。在神经网络声纳目标识别前的噪声预处理方法中,选用了功率谱特征提取、双谱特征提取算法;在研究了提取的特征后,选取反向传播神经网络(BP)模型;在此基础上构造了BP神经网络,并对网络进行训练与测试,给出识别实验结果。仿真模拟分析证明,基于神经网络的声纳特征级信息的融合,对目标分类有一定效果,为进一步实现声纳信息融合奠定了基础更多还原

【Abstract】 The main tasks in sonar object character information fusion are the realization of information compression and the determination of object identity (classification) for real-time processing and decision making analysis. It is difficult to establish a mathematical model-based algorithm due to complicated conditions. Artificial Neural Network can store memory and restore memory under the case of resembling input through studies of specimens so as to avoid setting up model. In noise pre-processing before recognition of sonar objects two methods of features extraction based on power spectrum and bi-spectrum are selected. Furthermore, a method of sonar object classification based on BP Neural Network is explored. On this basis, BP NN is constructed, trained and tested as well as the experimental recognition results are given. The results indicate that the sonar information fusion based on neural network has certain effects on object classification and can be the groudwork of sonar information fusion.更多还原