Multi-label Classification Model Of Human Cell Protein Cell Atlas Based On Convolutional Neural Network

Protein is an important component of all cells and tissues of the human body,and all important parts of the body need the participation of proteins.High-power microscopes can be used to visualize human cellular proteins.These im-ages are often used in biomedical research to help humans learn more about the complexity of cells and various diseases.With the development of high-power mi-croscope imaging technology,The acquisition speed of cellular protein images has increased,forming several large sample datasets represented by human protein cell atlas images dataset.In recent years,With the development of deep learning,the accuracy of the model uesd for images classification has rapidly been improved because of the deep convolutional neural networks.and the new models are usu-ally more robust than before.Using the big data methods to analyze the human protein cell atlas,it has become a hot spot to explore all possible rules in the dataset.Among them,the human protein image classification is an important direction.In the past,cell protein image classification task focused on the discus-sion of multitask classification models.For a certain protein,each classifier can recognize only one protein.In reality,there are different proteins in a single cell,and different proteins may be related.This paper hopes to use multi-label learn-ing algorithms to model the human protein cell atlas image dataset and identify it based on a set of different cells.The different proteins contained in the classifier can simultaneously identify all protein types that may be present in a cell.First of all,several data processing skills are used to avoid the influence of inblanced data,such as data augmentation,oversampling,and standardization.Secondly,this article show the difference between several traditional convolutional neural networks,including AlxetNet,VggNet,InceptionNet,ResNet,DenseNet.The convolutional neural network structure illustrates the characteristics and advan-tages of each convolutional neural network model.After that,this paper talks about the definition of multi-label learning,metrics,common algorithms,existing problems,and our method that is suitable for the human protein cell atlas image dataset.In the practical part,this paper uses data augmentation,oversampling and other methods to preprocess the human protein cell atlas image dataset,and uses some data visulization skills to analyze the dataset.Loss function,op-timization method,and metrics are choosed according to the characteristics of the dataset,and hyperparameter tuning sklls are uesed to find a relatively stable and suitable learningrate.We compare the impact of different neural networks on classification performance.This article innovatively models the human pro-tein cell atlas image dataset.The model has strong generalization ability,high accuracy and macro F1 score,and low model complexity.