改进的粒子群神经网络检测种蛋成活性

Improved PSO neural network system for automatic detecting fertility of hatching eggs

提出了一种基于改进粒子群神经网络进行孵化种蛋成活性自动检测的方法。提取HSI图像的H分量作为孵化种蛋表面颜色特征,通过主成分分析,找到了6个主成分特征向量,减少了神经网络输入节点数。利用改进粒子群算法优化多层前馈神经网络的拓扑结构,提高了神经网络的学习质量和速度。训练集的样本具有足够代表性和全面性,提高了网络的泛化能力。实验证明,该方法检测准确性较高,具有鲁棒性和高效率。

An improved particle swarm optimization neural network for automatic detection fertility of hatching eggs is presented. RGB images of eggs are obtained using computer vision technique, and are converted into HSI color images. Hue values of the images are extracted as the related feature of egg＇s surface. The primary components of feature parameters are extracted and selected using primary component analysis （PCA）. Structure of multi-layer feedback forward neural network is optimized using an improved particle swarm optimization algorithm. Learning quality and training speed of the neural network are improved. The samples of network training are representative and comprehensive, it improved generalization ability ofthe neural network. The experimental results show that the neural network model for fertility of hatching eggs detection had a high accuracy and efficiency and the algorithm was robust.