Search

Traditional white blood cell detection usually requires artificial extraction of cell features, which have higher resolution and contain more detailed information. However, due to the complex background of blood cell images and the large individual differences between cells, artificial feature extraction methods have poor generalization for different image data sets. Convolutional neural networks can extract deep features with stronger semantic information through their powerful self-learning capabilities. However, the pooling layer will cause the loss of some detailed information, while also ignoring the relationship between the whole and the part. Therefore, this paper innovatively proposes a white blood cell classification algorithm that combines deep learning features with artificial features. This algorithm not only uses artificial features, but also combines the self-learning capabilities of Inception V3 to make full use of the feature information of the image. At the same time, this paper introduces the transfer learning method to solve the problem of the dataset limitation. The final experimental results show that by fusing deep learning features with artificial features, the classification accuracy of white blood cell images reaches more than 99%. The proposed method has both low complexity and high accuracy, which makes it of great reference value for other medical image detection problems.