基于量子遗传模糊神经网络的苹果果实识别

Apple Recognition Based on Fuzzy Neural Network and Quantum Genetic Algorithm

针对田间苹果采摘机器人视觉系统中彩色图像边界像素的模糊性和不确定性影响苹果果实识别精度和速度问题,提出了一种将量子遗传算法的全局搜索能力和模糊推理神经网络的自适应性相结合的算法来识别苹果果实。利用量子遗传算法对模糊神经网络的可调整参数初始值进行了全局优化,加快了网络学习速度,避免了传统BP误差反向传播学习算法易陷入局部极小值、迭代次数多等弊端。实验表明:该识别模型高速且稳定,鲁棒性好,对于果实本身颜色不均匀样本正确识别率为100%,对自然光照引起颜色不均匀样本正确识别率为96.86%,对邻接图像正确识别率为94.29%,对重叠图像正确识别率为92.31%。

The apple images were hard to be identified at a faster speed and a higher accuracy because of fuzzy and uncertain factors existing in the color image boundary pixels, so in order to overcome the disadvantages above, a model combined quantum genetic algorithm and fuzzy neural network was built up which showed the capability of global search capability and adaptation. In the proposed model, quantum genetic algorithm was used to optimize the initial value of adjustable parameter in fuzzy neural network, which avoided redundant iteration and the incline to fall into the local minimum value of traditional BP algorithm. The experimental results showed that the proposed model achieved accuracy of 100% for the uneven color samples, 96.86% for sunlight influenced samples, 94.29% for the adjacent samples, and 92.31% for the overlapping samples.