Few-shot sketch recognition for plotting system

Abstract

Plotting system refers to a software system that can draw a series of vector symbols, typesetting, saving, reading and so on. The traditional plotting system uses key and mouse operation to draw. With the development of electronic equipment touch screen in recent years, the traditional menu selection is not convenient enough, but the way of using hand-painted sketch and recognized by the system is more natural and fast. Traditional image recognition methods need huge amount of data for training, but the vector symbol set in the plotting system is not a common symbol set, so it is difficult to obtain a large amount of data. Therefore, this paper proposes a β-TCVAE-Attention hand-drawn plot recognition method which can obtain better recognition accuracy in the case of small samples. The model first uses the Image dataset to pre-train βTCVAE, and then uses the small sample data of the plot sketch to input the β-TCVAE encoder to obtain the sketch features. After the sketch features are enhanced with the attention module, they are input into the Softmax classifier based on cosine similarity. The neural network is fine-tuned according to the classification results of the classifier. Finally, when using network prediction, input the support set and query graph into the network for classification. In this paper, a 20-way-5shot plotting data set is collected for experiment, and the experiment proves that the model has certain accuracy.