Study On Recognition Of Body Fluid Cells Microscopic Image Based On Convolutional Neural Network

For the moment,the traditional method of cell image recognition is to artificially design features at first,and then use the classification algorithm of machine learning to conduct identification.For each specific project,it is necessary to design special algorithms for preprocessing,segmentation,feature extraction.In the processes of algorithm design,it is quite easy to cause errors,and hard to achieve the ideal recognition precision,which makes cell image classification a hot and difficult issue in the research.The cell image classification method based on the convolutional neural network avoids the complex feature engineering and preprocessing in traditional methods,and has higher accuracy and robustness.This paper applies the convolutional neural network method to the automatic identification of body fluid cells,which has the advantages of simple use and high accuracy.The main tasks of this paper are as follows:(1)This paper first introduces the composition of the data set in this paper,and then performs data expansion,normalization,and image enhancement preprocessing steps for the problems of non-uniform cell distribution and poor image clarity of the original data set.,laying the foundation for the following experimental validity at the data level;(2)The process of artificial neural network evolution from perceptron model to convolutional neural network is introduced.The basic theoretical knowledge is described,which provides a theoretical basis for the establishment and improvement of the following model.(3)Introduce the structure and principle of classical LeNet-5 model and AlexNet model.Based on the classic LeNet-5 model and AlexNet model,construct several improved cell image classification models respectively,and compare each model by simulation experiments.The effect of classification on the network performance of the convolution kernel and the number of neurons in the network layer was initially explored;(4)To summarize and analyze the shortcomings of the improved network model based on LeNet-5 and the improved AlexNet-based network model,and then construct a network structure with the best effect on the recognition of humoral microscopic cells.Based on AlexNet network,this paper uses the fractional maximum pooling layer instead of the maximum pooling layer to reduce the speed of feature downsampling by the largest pooled layer,to ensure that the expression of the final abstracted feature is not greatly lost;then in the network In the first convolutional layer,the 3 × 3 and 5?5 size convolutional images are simultaneously used to perform feature extraction more comprehensively.The results of simulation experiments show that the improved convolutional neural network finally achieves a 96.7%recognition rate for microscopic images of body fluid cells,which is superior to the traditional artificial features and other cell classification networks and has a certain use value.