Research On Semantic Segmentation Of Medical Images Based On Improved Marine Predator Algorithm Optimized PSPNet

Meta-heuristic optimization algorithm(MOAs)is essentially a category of stochastic optimization in Operations Research.It mainly uses random operators and variables to search the space globally,while avoiding local optimality,which is much easier to understand and implement than deterministic methods.Although MOAs cannot guarantee that the global solution will be obtained in the last iteration,its near-global solution,independence,flexibility and non-gradient characteristics make it widely studied and used.Convolutional Neural Network(CNN)is generally used as the backbone network of image semantic segmentation model based on deep learning.In the training stage,CNN needs a large amount of training data,with many hyperparameters and various types of fixed structures.Meanwhile,CNN is very sensitive to the value of its hyperparameters,so it is difficult to determine the optimal value of these hyperparameters manually.Expert experience is needed to determine the hyperparameter values,non-experts generally use random search or grid search.Random searches and grid searches are better than manually tuned hyperparameters in the same time-intensive situation.Therefore,some recent related papers can consider determining the value of hyperparameters as an optimization problem,MOAs can be used to optimize the hyperparameter of CNN.COVID-19 is a rapidly spreading and deadly coronavirus that now spreads around the world.Which has a high fatality rate when accompanied by underlying diseases.The medical image selected in this paper is the COVID-19 data set,which is only used as an example to verify the effectiveness of the model in this paper.Existing deep learning models for diagnosing COVID-19 are mostly used for classification.In order to facilitate the direct observation and diagnosis of lesions by medical staff,a semantic segmentation model for automatic segmentation of COVID-19 lesions was proposed in this paper.There are not many studies on semantic segmentation combining deep learning and MOAs,so this paper uses MOAs to optimize CNN for COVID-19 image semantic segmentation,which is of research significance.The main work completed is as follows:(1)The basic concepts related to MOAs and the Marine Predator Algorithm(MPA)mainly used in this paper are introduced.The MPA is improved from two aspects of optimization initialization and position update,and the effectiveness of the improved algorithm is verified by experiments step by step.(2)The COVID-19 semantic segmentation dataset of this paper was selected and established from the publicly published dataset of the Chinese Consortium for Chest CT Imaging Investigation(CC-CCII).Introduced the related process of establishing image data set,including data set acquisition,image preprocessing,data set division and data enlargement.(3)Use the improved optimization algorithm in the first step to optimize the hyperparameters of the training backbone network,input them into the PSPNet network,and add the attention mechanism module at the end.Integrate the above to obtain a semantic segmentation network that effectively improves the accuracy of image segmentation.The original network,two common networks and the improved network are compared and experimented.The experiment proves that the accuracy of the improved network is higher,which is 99.53%,and the mean intersection over union(mIOU)is 0.7275,which is 0.1701 higher than the original network.The classification and location of the lesion can be segmented to provide a basis for the patient’s condition and the possibility of other complications.The purpose of this case study is to provide radiologists and other clinicians with a diagnostic aid.Under no circumstances is this pilot study a substitute for medical advice.