Research on Small Sample Image Recognition Based on Transfer Learning

Abstract

With the improvement of electronic hardware configuration performance, high-speed computing of large and complex data can be realized, and artificial intelligence, which requires a large amount of data calculation has been rapidly developed. Nowadays, artificial intelligence has developed rapidly in image recognition, natural language processing and speech recognition, and also has achieved good results in these areas. Among them, the convolutional neural networks plays an important role in image recognition. Some excellent convolutional neural networks such as VggNet, GoogleNet and ResNet have reached more than 90% in image recognition accuracy. But these convolutional neural network models are based on big data, consume a lot of computing time, which is very costly. Therefore, when data is limited, hardware configuration is low, and time is tight, a research on small sample image recognition based on transfer learning becomes very necessary. In this paper, the inception-V3 model is used to transfer learning. The training nodes of the model are skillfully used to train the model on the data set of ten kinds of fruits. The model is tested by the training model and good accuracy is obtained. The advantage of the experiment is that the training time is short, the required data set is small, and the image recognition accuracy is high