Blood Cell Image Recognition Algorithm based on EfficientNet

Abstract

Diagnosis of blood-related diseases usually requires identification of patient blood samples, and fast and accurate classification of blood cells can help improve the efficiency of doctors’ diagnosis. To address the problems of traditional manual methods that are time-consuming and error-prone, as well as the low accuracy of current automatic blood cell classification algorithms, this paper uses EfficientNet to classify blood cell images. We propose an EfficientNet-based blood cell image classification algorithm, using EfficientNet-B7 as the classification model and introducing CLAHE in data preprocessing to improve the image quality. The experiments were conducted on the BCCD dataset and compared with various classification networks and blood cell classification studies. The experimental results show that our proposed EfficientNet-based blood cell image classification algorithm achieves 99.6% accuracy on the BCCD dataset. The current optimal blood cell classification algorithm has only 94% accuracy on this dataset, and the algorithm in this paper outperforms it by 5.6 percentage points. This study will hopefully be used to replace the manual classification of blood cells by professional physicians.