Red Blood Cell Detection Using Improved Mask R-CNN

Automatic segmentation of microscopy images is an important task in medical image processing and analysis. Red blood cell detection is an important example of this task. Manual detection is not only labor-intensive, but also prone to misdirection and omission. In order to enhance the speed and accuracy, an improved mask regional convolution neural network (Mask R-CNN) is proposed in this paper. The algorithm utilizes Split-Attention Networks (ResNeSt) as a feature extraction network. ResNeSt combines channel attention with multi-path representation, and feature extraction is performed in combination with Feature Pyramid Network (FPN). The experimental results show that the improved Mask R-CNN has an average precision increase of 2.55%, and improves the efficiency of red blood cell detection.