Research And Application Of Cell Counting Method Based On U-Net++ Neural Network Model

Cell counting is a common laboratory technique that can confirm the number of cells in a biological sample and has a wide range of applications in important scientific problems in biology,medicine,agriculture and environmental sciences.Nowadays,the main counting methods are traditional manual counting methods and automatic counting methods based on computer images.The traditional cell counting method requires manual microscope observation,which is not only time-consuming and laborious,but also easy to be interfered by human factors.The results are often subjective and prone to large errors.Compared with manual counting,the cell counting method based on image processing technology can automatically recognize and count cells through digital image processing algorithm,improve the efficiency and accuracy of counting,and also ensure the objectivity and repeatability of data.In recent years,with the development of artificial intelligence technology,deep learning networks are gradually applied to cell counting tasks to provide more accurate and reliable cell counting results.However,they are more suitable for counting scenes with clear cell outline and no overlap,and the counting effect is not ideal for intensive cell images with many adhesive overlaps and not obvious layers.It is impossible to segment adhesive cells for accurate counting,resulting in a high error rate.To solve the above problems,this paper proposes a cell counting method based on the improved U-Net ++ neural network model,verifies the efficiency of the method by comparing with the existing methods,and finally develops a cell counting system based on the proposed method.The main contents are as follows:(1)This thesis proposes a cell counting method based on the improved U-Net ++neural network model.In this thesis,the U-Net ++ network model integrated with the self-attention mechanism is used as the backbone network to extract the high-dimensional features of cell images,and then the fully-connected neural network is used to train the high-dimensional features extracted to realize the counting.The classification problem is transformed into a regression problem,which can detect and segment cells more effectively and calculate the number of cells in the image more accurately.The proposed method was tested on two microscope image data sets and compared with six other methods.The results show that the proposed cell image counting method based on U-Net++ network model is superior to the existing methods in both accuracy and speed,and its accuracy can reach 97.4%.In this paper,we find that Laplacian operator is better than Roberts operator,Prewitt operator and Sobel operator in the intersection of image segmentation and the final count.It is also found that U-Net++ segmentation of overlapping cells is more clear.The intersection set of FCN network model segmentation is 70.8%,U-Net network segmentation is 78.3%,and U-Net++ segmentation is 86.1%.(2)A cell counting system based on the proposed method is developed in this paper.In order to realize the desktop application of the algorithm in this paper,this paper chooses the development mode of separating the front end from the front end.The frontend project runs in the browser kernel,and then the browser kernel is embedded into the back-end project.The front-end calls the back-end interface through Http request for data interaction and processing.The back end stores the processed data in the SQLite database and sends it to the front end by calling Vue directly,which greatly speeds up the development schedule and reduces the cost of learning.The cell counting system includes six functional modules: basic function of image,microscopic image processing,target measurement/particle statistics,drawing and labeling,and report printing.Support the direct import of cell image,denoising,adjustment,contrast enhancement and other preprocessing operations;Using the counting algorithm in this paper,the system automatically locates the cell position,analyzes the cell size and shape,and finally realizes the cell classification and counting.The cell counting system developed in this paper has comprehensive functions,accurate counting and high degree of automation,which is more in line with the actual needs of researchers.