改进VGG模型在苹果外观分类中的应用

Application of Improved VGG Model in Apple Appearance Classification

为了将采摘后的苹果进行外观分类,提出了一种基于卷积神经网络的方法,通过改进VGG卷积神经网络完成对外观正常苹果、病斑苹果和腐烂苹果的分类。在VGG-16网络的基础上,加入批归一化层、采用全局平均池化和联合损失函数的方法对其进行结构优化。在经过数据增广的数据集上,与其他分类方法进行对比,结果表明:改进后的VGG网络对外观正常苹果、病斑苹果和腐烂苹果的识别精度分别为99.61%、98.89%和99.26%,均高于未改进VGGNet、AlexNet和GoogLeNet算法,证明此网络能够很好地完成对苹果外观的分类识别,可为采摘后的苹果实现智能分类提供技术支持。

In order to realize the classification of apples after picking, a method based on convolutional neural network was proposed. Based on the improved VGG convolutional neural network, the normal apples, diseased apples and rotten apples were classified. On the basis of the VGG-16 network, the batch normalization layer was added, and the global average pooling and joint loss function were used to optimize the structure. It was compared with other classification methods on data-enhanced data sets. The results show that the improved VGG network’s recognition accuracy for normal apple, diseased apple and rotten apple is 99.61%, 98.89% and 99.26%, respectively, which are higher than that of the unmodified VGGNet, AlexNet and GoogLeNet algorithms. It proves that the proposed network can complete the classification and recognition of the appearance of apples. It can provide technical support for the intelligent classification of apple after picking.