A fast and yet efficient YOLOv3 for blood cell detection

Abstract

These days, blood cell detection in microscopic images plays a vital role in cognition, the health of a patient. Since disease detection based on manual checking of blood cells is mostly time-consuming and full of errors, analysis of blood cells using object detectors can be considered as an effective tool. Hence, in this study, an object detector has been proposed which is used for detecting blood objects such as white blood cells, red blood cells, and platelets. This detector is called FED (Fast and Efficient YOLOv3) and it is a One-Stage detector, which is similar to YOLOv3, performs detection in three scales. For the purpose of increasing efficiency and flexibility, the proposed object detector utilizes the EfficientNet Convolutional Neural Network as the backbone effectiveness. Furthermore, the Dilated Convolution is indeed applied in order to increase receptive view of the backbone. In addition, the Depthwise Separable Convolution method is utilized to minimize the detector’s parametersand the Distance Intersection over Union is further used for bounding box regression. Besides, for increasing the performance, the Swish activation function is employed. The experiments are run on the BCCD dataset that the average precision of platelets, red blood cells, and white blood cells become 90.25%, 80.41%, and 98.92%, respectively. The results of experiments and comparisons demonstrate that the proposed FED detector is more efficient than other existing studies for blood cell detection.