Accuracy improvement of Thai food image recognition using deep convolutional neural networks

To improve the performance of the convolutional neural networks, it is normally done by increase the deepness or put more layers to the network. By doing such, the number of parameters is increased. In this paper, NU-InNet, which was developed from GoogLeNet, is modified by adding more layers to the network in order to improve the accuracy of the network while keeping the number of the parameters to be suitable for being used in a smartphone. Testing the proposed model with a database containing 50 well-known kinds of Thai food, it is found that the processing time and size of the parameters of NU-InNet with a depth of 4 are less than those of BN-Inception network by 1.5 and 11 times, respectively. Importantly, the accuracy of NU-InNet with a depth of 4 is higher than that of BN-Inception network by 8.07%.