应用深度学习方法的汽车轮毂类型识别

Automotive wheel hub type recognition by deep learning method

针对汽车轮毂大批量生产中人工分类效率低、错误率高的问题,研究应用机器视觉和深度学习方法对汽车轮毂进行在线识别和分类。根据轮毂的特点对AlexNet进行修改,将Leaky Relu函数作为激活函数,应用批归一化,改变卷积核大小,简化全连接层并增加平均池化,且在全连接层之前加入通道注意力机制模块,使得修改的模型可以更好地进行特征提取。综合试验结果表明,采用修改后的AlexNet模型进行轮毂分类,其测试精确度达到99.80%,训练时间大幅缩短,综合效果优于常用机器视觉分类方法,具有实际应用价值。

In order to solve the problems of low efficiency and high error rate in manual classification of automobile wheel hubs mass production, the application of machine vision and deep learning method to online identification and classification of automobile wheel hubs was studied.According to the characteristics of the wheel hub, AlexNet was modified, using Leaky Relu function as the activation function, the batch normalization was applied and the size of convolution kernels was changed, the connection layer was simplified, average pooling was increased, and the channel attention mechanism module was joined before the connection layer, the modified model can better extract feature. The comprehensive experimental results show that the recognizable precision of the modified AlexNet model achieves 99.80 % in wheel hub classification test, and the training time is greatly shorted, the comprehensive effect is superior to the common machine vision classification methods, and it has practical application value.