Litcius/Paper detail

An Improved EIoU-Yolov5 Algorithm for Blood Cell Detection and Counting

Zhen Zhang, Zijun Deng, Zhipan Wu, Guoming Lai

20222022 5th International Conference on Pattern Recognition and Artificial Intelligence (PRAI)15 citationsDOI

Abstract

At present, detecting blood cells commonly depends on artificial count by observing microscope. Artificial observation takes a lot of manpower. In the object detection algorithm for real-time detection, the representative algorithm of the real-time object detection algorithm is Yolov5 algorithm. However, the accuracy of Yolov5 algorithm is low. In order to improve the accuracy of the algorithm, realize the automation of blood detection and reduce the workload of manual count, this paper proposes an improved algorithm EIoU-YOLOV5 based on Yolov5, which improves the Loss function of prediction box by replacing CIoU Loss with EIoU Loss. Experimental results on the common data set BCCD show that, compared with the original Yolov5 algorithm, the Recall of EIou-YOLOV5 algorithm increased from 0.855 to 0.917, which increased by 6.2 percentage points. Therefore, it reduced the missed rate effectively. The result of [email protected] is increased from 0.899 to 0.922, which is raised by 2.3 percentage points. Among them, the detection of platelet increased the most. It increased from 0.858 to 0.92, which increased by 6.2 percentage points. Therefore, the improved EIoU-yolov5 algorithm can better assist clinical diagnosis.

Topics & Concepts

AlgorithmComputer scienceObject detectionWorkloadAutomationData setSet (abstract data type)Artificial intelligencePattern recognition (psychology)EngineeringOperating systemMechanical engineeringProgramming languageDigital Imaging for Blood DiseasesRetinal Imaging and AnalysisCOVID-19 diagnosis using AI
An Improved EIoU-Yolov5 Algorithm for Blood Cell Detection and Counting | Litcius