Research On Multispectral Image Segmentation And Blood Vessel Detection Technology

With the development of digital imaging technology and medical imaging equipment,image identification has gradually become an important auxiliary treatment in the medical field.Traditional image segmentation algorithms are often difficult to achieve accurate segmentation because of the characteristics of medical images such as high amount of noise,high degree of blur and high complexity of structure and morphology.In addition,the difficulty of obtaining valid data is also a big obstacle that limits the application of image segmentation in the medical field.In view of the above problems and in order to improve the segmentation accuracy,multi-spectral imaging technology is introduced in this paper to utilize multi-band information.The main research work is as follows:(1)In the absence of samples and labels,the multi-spectral image segmentation model based on super-pixel clustering in frequency domain is proposed for the problem that traditional image segmentation algorithm has low segmentation accuracy on medical images.The algorithm consists super-pixel pre-segmentation module and unsupervised clustering in frequency domain module.The super-pixel pre-segmentation module introduces the uniform random initialization model and the multi-dimensional feature computing model.The unsupervised clustering learning module introduces the LBP normalized computing model,the frequency domain channel splicing model and the sample vector dimension reduction model.The algorithm is optimized for multi-spectral images to make full use of the information in multiple dimensions.Experiments show that compared with other algorithms,this model achieves higher segmentation accuracy in medical images.(2)With limited samples and labels,the multi-spectral image segmentation model based on dense convolutional network is proposed for the problem that most of current image segmentation algorithms based on deep learning have poor performance when foreground-background class is unbalanced.The algorithm introduces the neighbor layer feature reconstruction module,the cross-layer feature reconstruction module and the loss balance calculation module to complete the network,which strengthen the transmission and utilization of valid information,and the input of network is optimized for multi-spectral images.Experiments show that compared with other algorithms and the super-pixel clustering in frequency domain segmentation model established in this paper,this model achieves higher segmentation accuracy in medical images.(3)To solve the problem of less imaging band and weak detection ability of existing vascular detection technology,this paper proposes a non-invasive high-precision multi-spectral vascular detection technology based on the two multi-spectral image segmentation models established in this paper,and the prototype system is build.The system consists of a band optimization module,an image preprocessing module,an image fusion module and an image segmentation module.Experiments show that compared with other similar methods,the multi-spectral vascular detection technology with the multi-spectral image segmentation model established in this paper has better detection performance.