Research Of Medical Image Object Detection And Sematic Segmentation Based On Deep Learning

Object detection and semantic segmentation in medical images is one of the important research contents in the field of computer vision and image processing.Its main purpose is to detect and identify specific targets in images to assist doctors in clinical diagnosis and surgical treatment of patients.Because the doctor's diagnosis depends heavily on his subjective experience and consciousness,and the imaging quality of medical images,the complex background environment will also have a greater impact on the doctor's judgment.Accurate detection and segmentation of medical images is of great help to lesion identification and quantitative analysis,and lays an important foundation for clinical applications such as image-guided surgery and treatment evaluation.Therefore,it is of great significance to accurately segment and identify specific targets in medical images.Traditional algorithms for image detection and segmentation need to design features manually,but it is very difficult to design robust features due to the diversity of target shapes and complex background environments.And deep learning technology can learn rich features through a large number of data samples.Compared with traditional image processing methods,the accuracy of deep learning algorithms has huge performance advantages in the fields of object detection and semantic segmentation.Therefore,object detection and semantic segmentation algorithms based on deep learning are widely used in the medical image field.The target of this thesis is to apply deep learning technology to detect and segment object in medical images.Compared with classification tasks,detection and segmentation tasks require accurate location information of the target.In this thesis feature fusion is used to capture the rich spatial location information of target objects And advanced semantic information,thereby improving the accuracy of detection and segmentation in medical images.The main contributions and work of this article are as follows:Firstly,in order to solve the problem of insufficient deep spatial position information in deep convolutional networks,an attention-based object detection method was proposed.This method uses the attention mechanism to enable the target detection network to adaptively fuse features between different layers,strengthen useful features and suppress useless features.Experimental results prove that the method can effectively improve the recognition accuracy of the object detection network.Secondly,to solve the problem that the segmentation network of the encoder-decoder structure has a poor segmentation results on the contours of targets in medical images,a method based on feature fusion is proposed.The final segmentation results are obtained by fusing the feature maps of different layers in the decoder.The experimental results prove that this method can effectively improve the segmentation accuracy.Finally,for the problem of poor segmentation result caused by the change in the shape and scale of the segmentation target in medical images,a semantic segmentation method based on multi-scale is proposed.This method uses different scale of convolution kernel and pooling module to extract multi-scale features and global text features in the encoder.Experimental results show that this method can effectively improve the segmentation accuracy of semantic segmentation networks.