论文部分内容阅读
针对BP神经网络容易陷入局部极小值的问题,借鉴模拟退火算法中metropolis接受准则的思想并加以改进,并引入禁忌(taboo)搜索算法中的禁忌表,在系统跳出局部极小值时对极小点进行记录比较,使得在处理多局部极小值系统时更加高效与精确。将改进后的算法应用于BP网络,从而构造出一种更易跳出局部极小值的改进的神经网络。最后,运用改进后的神经网络算法进行图像压缩与重构,实验结果表明改进后的神经网络收敛速度更快,具有更高的效率与精度。
Aiming at the problem that BP neural network is apt to fall into local minima, the idea of metropolis in simulated annealing algorithm is accepted and improved, and taboo tabu search tabu search is introduced. When the system jumps out of local minima, Dot comparison recording, making multi-local minimum system more efficient and accurate. The improved algorithm is applied to the BP network to construct an improved neural network that can jump out of the local minimum easily. Finally, the improved neural network algorithm is used to compress and reconstruct the image. The experimental results show that the improved neural network converges faster and has higher efficiency and precision.