Blood cell detection and counting in microscopy images using overlapping object recognizing model

Computer vision is one of the most striking areas in data science. Like other fields of data science, the application of this field has also become part of our personal lives. Classifying an overlapping object in cell detection is one of the main challenges faced by researchers who work in object detection and recognition. The white and red blood cells have a significant part in controlling the human body's framework. One of the essential tests for assessing human medical issue is checking the amount of white blood cells and red blood cells, as well as platelets number in the blood. In this project, we used one of the Convolutional Neural Network (CNN) techniques which is Faster Region CNN to recognize, classify, and count the blood cell elements. Our objective is to detect blood cell images from dataset, separate between overlapping cells and other objects, and then count the number of each blood cell. The researchers used 720 images containing 1603 labelled blood cells from 10 categories to train the model. Those images were marked by clinical the Hospital Clinic of Barcelona ‘s pathologists and were obtained from the dissemination of blood drawn from patients during 2015-2019. After training steps were performed on 60 images, and blood cell detection and segmentation were performed in the next steps. The average accuracy was 97%, and the verification loss was 18.4%. By using this our Faster R-CNN framework object detection method can enhance the accuracy and speed of cell detection.