Research On Algorithms Of Human Sperm Head Segmentation Based On Convolutional Neural Network

As an important step of sperm morphology analysis,sperm head segmentation is an indispensable part of sperm quality analysis.Traditional segmentation methods often use manual features such as gray and texture to segment sperm head,which lacks sufficient discrimination and robustness.With the development of convolutional neural network and deep learning technology,more and more image segmentation tasks can be well solved by convolutional neural network.Compared with the traditional methods,the convolution neural network method can automatically learn the features in the image through a large amount of data,thus avoiding the manual design of features,and can achieve more accurate segmentation results.Based on convolution neural network,this paper studies the segmentation algorithm of human sperm head:1)This paper proposes a sperm head segmentation network based on feature extraction module and self attention module.In order to solve the problem that the detail information of sperm head image is lost in the process of down sampling,this paper proposes a feature extraction module,which can extract the multi-scale information of the image without reducing the resolution of the image,and better restore the detail information of the image,especially for the sperm head.At the same time,in order to enhance the representation ability of the model,the self attention module is introduced into the decoder.In addition,in order to further improve the segmentation accuracy,the network also uses a multi-scale input module to enhance the segmentation effect of sperm heads with different scales.In this paper,a large number of experiments on human sperm head segmentation data set show that the proposed method is better than the traditional segmentation method and other segmentation methods based on convolution neural network.2)In view of the excellent performance of generative adversary network in many computer vision tasks,we use generative adversary network to achieve sperm head segmentation.Considering the small size and dense distribution of sperm head,we used a high-resolution generation network.The network is composed of four parallel branches with different resolutions.The information in each branch is aggregated by multi-scale information fusion module,which can extract rich context information while keeping the image resolution unchanged.In addition,we use a patch discriminant network.The traditional discriminant network can only discriminate the whole image,which is easy to produce local information ambiguity.Patch discriminant network can discriminate the local information of the image,so as to overcome this problem.A large number of experiments show that our model is better than other segmentation methods based on deep learning.At the same time,the network has good effect on human eye blood vessel segmentation data set,and has good robustness.