Research On White Blood Cell Classification And Recognition Method Based On Convolution Neural Network

The detection of white blood cells in the peripheral blood is of great value for the diagnosis and treatment of many diseases. Currently, blood cell analyzers and manual microscopy are often used clinically to classify and detect white blood cells. However, this method is less efficient and is greatly affected by the subjective feelings of the microscopy personnel. With the development of medical image processing technology, the use of image processing methods to automatically classify and recognize white blood cells has high practical value. Therefore,this paper proposes a white blood cell classification and recognition method based on convolutional neural network.First of all, in view of the different staining methods in the white blood cell image collection and the influence of hardware equipment, the gray unevenness, weak boundary and noise problems that interfere with the accurate segmentation of white blood cell images are proposed. . The algorithm performs density peak clustering and level set processing on the white blood cell image, takes the obtained white blood cell nuclear contour as the initial contour curve,and uses the symbolic pressure function based on the Johnson-Shannon divergence information to overcome the above-mentioned problems of the white blood cell image and guide the contour curve Complete the precise segmentation of white blood cell images. Secondly,aiming at the problem of the lack of peripheral blood image data, the segmented white blood cell images were expanded, and the expanded images were randomly pasted to synthesize a large number of simulated peripheral blood images to complete the production of a multi-label white blood cell classification and recognition data set. Then, there are few types of white blood cell classification research, large changes in white blood cell scales in peripheral blood images,complex backgrounds, and problems that affect detection performance. This paper proposes a multi-label white blood cell classification and recognition method based on convolutional neural network, which realizes the classification and recognition of 11 types of white blood cells. Based on the YOLOv3 network, the method based on feature map channel information enhancement is integrated into the network feature extraction module, and the feature map fusion method with weight is used to replace the original feature pyramid model, so that the feature extraction and fusion capabilities of the network are enhanced. Finally, an improved model is used to visually analyze the detection results, which confirms that the proposed method has improved the accuracy of white blood cell image positioning and small-size white blood cell classification and recognition.Experimental results show that the proposed white blood cell image segmentation algorithm has better segmentation results for white blood cell images with noise interference,uneven gray scale and weak boundaries. The proposed white blood cell classification and recognition method based on convolutional neural network has an average accuracy of 92.2% in the classification and detection of 11 types of white blood cells, and shows good recognition ability in real peripheral blood images, realizing a true end-to-end Multi-label classification and recognition.