基于轻量级卷积神经网络的苹果表面缺陷检测方法

Apple surface defect detection method based on lightweight convolutional neural network

为了解决目前农业信息领域对苹果表面缺陷检测准确率低的问题,提出一种基于轻量级卷积神经网络的苹果表面缺陷检测方法。首先采集苹果缺陷样本图片制作实验数据集用于模型训练和测试;其次在AlexNet网络结构的基础上,引入深度可分离卷积代替原有网络中的标准卷积运算来进行图像特征的提取;最后利用全局平均池化方法代替原有网络中的全连接层,从而将卷积层输出的多个特征图以自身为单位进行映射得到特征点。实验结果表明改进后网络对苹果缺陷识别精度达到了98.57%,较改进前提升1.55%;较改进前模型参数量减少99.3%、训练速度提高32.67%、FPS提高33.28%,改进后的轻量级卷积神经网络不仅减少了模型参数量和训练时间,而且提高了检测精度和速度。因此,新的检测方法在减少模型参数量的同时,还可保证模型的检测精度和效率,具有较强的工程实用性,可为苹果缺陷分类提供理论参考。

In order to solve the problem of low accuracy of apple surface defect detection in the field of agricultural information,a method of apple surface defect detection based on lightweight convolutional neural network was proposed.Firstly,apple defect sample images were collected to make experimental dataset for model training and testing;Secondly,on the basis of AlexNet network structure,depth separable convolution was introduced to replace the standard convolution operation in the original network to extract image features;Finally,the global average pooling method was used to replace the full connection layer in the original network,and the output feature maps of the convolution layer as a unit were mapped to get the feature points.The experimental results show that the recognition accuracy of the improved network is 98.57%,which is 1.55%higher than that before;Compared with the model before improvement,the model parameters reduce by 99.3%,training speed increases by 32.67%,and FPS increases by 33.28%.The improved lightweight convolutional neural network not only reduces the model parameters and training time,but also improves the detection accuracy and speed.Therefore,the new detection method can reduce the number of model parameters,as well as ensure the detection accuracy and efficiency of the model,which has strong engineering practicability and provides theoretical reference for apple defect classification.