论文部分内容阅读
提出了一种具有实用意义的形态滤波神经网络模型及其网络参数的模拟退火优化算法。通过分析指出,形态滤波网络的优化设计过程实际上就是网络参数(结构元素)不断调整、逐步适应图像环境的优化学习过程,从而将目标客体的特征规律反映到网络结构上来,赋予结构元素特定的知识,使形态滤波过程融入特有的智能,以实现对复杂变化的图像具有良好的滤波性能和稳健的适应能力。为结合运动图像目标的检测需要,采用了渐近收缩误差、适时校正网络权值的动态跟踪学习算法。通过实验结果可以看出,该算法不仅能适应复杂多样的背景环境,而且对运动目标的持续检测能力具有位移不变、伸缩不变和旋转不变的特性。
A practical morphological filter neural network model and its simulated annealing algorithm for network parameters are proposed. The analysis shows that the optimization design process of morphological filter network is actually the continuous adjustment of network parameters (structural elements), and gradually adapt to the optimization learning process of image environment, so that the characteristics of the target object is reflected in the network structure, and the structural elements are given Knowledge, morphology filtering process into the unique intelligence, in order to achieve the complex changes in the image has good filtering performance and robust adaptability. In order to meet the needs of moving target detection, a dynamic tracking learning algorithm that uses asymptotic shrinkage error and timely corrects network weights is adopted. The experimental results show that this algorithm not only can adapt to the complex and diverse background environment, but also has the characteristics of constant displacement, invariant and rotational invariance to the continuous detection ability of moving target.