Research on Blood Cell Detection Algorithm Based on Improved YOLOv7

In the biomedical field, the detection of blood cells in microscopic images is crucial for assisting physicians in diagnosing blood-related diseases and plays a significant role in promoting the development of medicine towards more precise and efficient treatments. Traditional manual detection methods are time-consuming and prone to errors, and existing blood cell detection technologies face significant challenges in meeting the requirements of high precision and real-time performance. In light of this, this paper, from the perspective of image recognition and with the aid of deep learning, proposes an efficient and rapid detection model based on YOLOv7. Firstly, in order to further extract global features, this paper selects the advanced Vision Transformer module to be added to the backbone network. Then, the convolutional layer of the SPPCSPC layer in the YOLOv7 backbone network is replaced with a parameterized convolution, thus avoiding the shortcoming of traditional static convolutions where all samples share one convolution kernel. To directly add the global feature information learned in the small-scale feature layer to the maximum-scale feature layer, this paper adds an upsampling module between the minimum-scale feature layer and the maximum-scale feature layer. By using the SIoU loss function instead of the CIoU loss function, the convergence speed is further accelerated, and the precision is improved. Secondly, the experimental results validate the effectiveness of the improved YOLOv7 model. Compared to the results of YOLOv7, the mAP@0.5 value of this paper’s model has increased by 3.4% compared to the original YOLOv7, by 0.7% compared to the YOLOv5x with a larger number of parameters, and by 0.3% compared to the currently best-performing open-source model in this dataset-CST-YOLO.