基于BP神经网络的啤酒瓶口检测方法

Method of Beer Bottle Mouth Inspection Based on BP Neural Network

运用BP神经网络可以实现啤酒瓶口的破损检测.首先获取啤酒瓶口图像,并进行图像处理.然后计算啤酒瓶口的周长、面积、圆形度和相对圆心距离4种特征参数,由这4种特征参数构成特征向量组.其次建立结构为4-7-1的BP神经网络模型,将特征向量组作为神经网络的输入.最后对啤酒瓶口破损情况进行训练,根据训练结果获得权值和阈值矩阵,通过逻辑转换关系获得啤酒瓶口的破损情况.经实验验证该方法具有很好的准确度和检测效率.

Damage identification of beer bottles could be realized by using the BP neural network. First, beer bottle mouth images were collected and processed. Four characteristic parameters including the perimeter, area, circularity, and the relative center distance were calculated and constituted a feature vector group. Then, the BP neural network model with the 4-7-1 model was established and the feature vectors group was inputted to the neural network. Finally, beer bottle damages would be trained. The weight and threshold matrix were acquired according to the training results. Beer bottle mouth damages would be easily obtained through logic relations. The experimental results verified that this method had good accuracy and inspection efficiency.