Vegetable Recognition and Classification Based on Improved VGG Deep Learning Network Model - Computer Vision Research Group（计算机视觉实验室）

To improve the accuracy of automatic recognition and classification of vegetables, this paper presents a method of recognition and classification of vegetable image based on deep learning, using the open source deep learning framework of Caffe, the improved VGG network model was used to train the vegetable image data set. We propose to combine the output feature of the first two fully connected layers (VGG-M). The Batch Normalization layers are added to the VGG-M network to improve the convergence speed and accuracy of the network (VGG-M-BN). The experimental verification, this paper method in the test data set on the classification of recognition accuracy rate as high as 96.5%, compared with VGG network (92.1%) and AlexNet network (86.3%), the accuracy rate has been greatly improved. At the same time, increasing the Batch Normalization layers make the network convergence speed nearly tripled. Improve the generalization ability of the model by expanding the scale of the training data set. Using VGG-M-BN network to train different number of vegetable image data sets, the experimental results show that the recognition accuracy decreases as the number of data sets decreases. By contrasting the activation functions, it is verified that the Rectified Linear Unit (ReLU) activation function is better than the traditional Sigmoid and Tanh functions in VGG-M-BN networks. The paper also verifies that the classification accuracy of VGG-M-BN network is improved due to the increase of batch_size.