The 4th International Conference on Electronics, Communications and Control Engineering | 2021
Corn Ear Quality Recognition Based on DCGAN Data Enhancement and Transfer Learning
Abstract
Aiming at the problems of poor feature extraction ability, low accuracy, low recognition efficiency, and over-fitting of convolutional neural networks when the data set (training set) is small, we propose a new corn ear quality recognition model called CornNet. Taking normal, messy kernels, mildew, mottled, and lacking kernels as the research objects, self-made datasets, and made classification labels. First, use the Deep Convolution Generative Adversarial Network (DCGAN) that introduces the Dropout2d method to enhance the sample data to expand the image training set. Then, the DPN-92 network is simplified into a DPN-68 network, a new fully connected layer module is designed, and the convolutional layer trained on the DPN-68 model on ImageNet is transferred to this model, and only new learning and transfer learning are trained the three learning methods of the fully connected layer of the model and all the layers of the transfer learning training model are compared for experiments. Finally, after DCGAN data enhancement, the entire transfer learning training model is used to form the CornNet model. The results show that the average recognition rate of the CornNet model is 98.75%, which is 3.58%∼12.04% higher than the state-of-the-art results.