基于改进遗传模拟退火算法的BP神经网络的畸变校正研究

The Study of Distortion Correction of BP Neural Network Based on Improved Genetic Simulated Annealing Algorithm

摘要:

针对相机所采集的图像大多都存在畸变现象的问题,设计了基于改进遗传模拟退火算法的BP神经网络校正算法。该算法针对传统遗传算法易于收敛局部最优的问题,提出分段选择策略与随机抽样相结合的选择算子,自适应交叉与变异算子。在畸变校正中,该算法通过网络的输入输出建立理想点与畸变点的关系,使用改进的遗传模拟退火算法来优化神经网络中的阈值与权值,然后使用基于LM算法的BP神经网络进行局部优化,最后通过插值算法得到校正后的图像。实验表明,该算法能过较好的对图像进行畸变校正,同时与传统的BP神经网络算法相比精度更高,收敛速度更快。

Abstract:

Aimed at the problem that the images collected by the camera are mostly distorted, a BP neural network correction algorithm based on improved genetic simulated annealing algorithm is designed. Aiming at the problem that traditional genetic algorithm is easy to converge local optimum, this algorithm proposes a selection operator combining segmentation selection strategy and random sampling, adaptive crossover operator and adaptive mutation operator. In the distortion correction, The algorithm establishes the relationship between the ideal point and the distortion point through the input and output of the network. Optimizing the threshold and weight in the neural network through improved genetic simulated annealing algorithm, then carring out local optimization through BP neural network based on LM algorithm, and finally obtaining the corrected images by interpolation. Experiments show that the proposed algorithm can correct the distortion of the image better than the traditional BP neural network algorithm, and the convergence speed is faster.