A Lightweight Compressed Model of Breast Cancer Classification Based on Convolutional Neural Network

Breast cancer is a severe disease that is quite frequent in women and Convolutional Neural Networks are popular to deal with breast cancer classification. To reach a high accuracy, most CNNs are designed with a large number of parameters which requires a lot of storage capacity. In this paper, a model with few parameters but high accuracy is proposed. The model is fine-tuned by the weight pruning method after preprocessing and initial training. Moreover, the model is also compressed by the method of weight clustering and obtain a model called VGG16_Sp4_Cl8, whose performance of accuracy reaches 91.95% and the size of the model is reduced by 9×compared to the baseline model.