| Transmission multispectral imaging(TMI)has the potential for medical applications,such as early screening for breast cancer.However,due to the strong scattering and absorption characteristic of biological tissue,the acquired blurred image data has very has low signal-to-noise ratio(SNR).The combination of frame accumulation technology and function signal modulation and demodulation technology can improve the gray scale and SNR of the acquired image,making it possible to detect and analyze the heterogeneity.However,for two-dimensional images,the computation required by demodulation image is huge,and the image obtained by demodulation is still very fuzzy,which makes it difficult to perform heterogeneity detection.In view of the above problems,this paper proposes a kind of two-dimensional fast demodulation algorithm and a heterotopic detection method.The main contents are completed as follows:(1)The combination method of frame accumulation and functional signal modulation and demodulation technology can obtain images of higher quality.For twodimensional images,it requires a large amount of computation for demodulated these image data.Moreover,the signal generator and camera that modulates the light sources are not driven by the same clock system,which makes phase fluctuation problem.The influence of this error on image quality increases with the decrease of sampling rate.Aiming at this application environment,an improved fast demodulation algorithm is proposed and applied to two-dimensional images.The effect of phase jitter problem can be reduced by averaging the modulation frequency.Under the premise of guaranteeing image quality,this method can significantly improve image demodulation speed and provide certain reference for fast demodulation of TMI.(2)In order to obtain the position and characteristic information of the heterogeneity,a method for detecting the heterogeneity in TMI is proposed.This method is divided into two steps: the first step is to obtain the general outline of the image,and the second step is to detect the heterogeneity by using spectral analysis.In the part of heterogeneity contour extraction,a complex heterogeneity contour extraction method for fuzzy image is proposed.The method mainly includes sampling under image exponent,Laplace operator and Otsu binarization.In the part of heterogeneity detection,a kind of assumption of invariant features of heterogeneity detection is proposed for heterogeneity classification based on spectral analysis,which reduces the difficulty of heterogeneity detection.(3)Deep learning is a new interdisciplinary subject of image recognition.It is an effective method to combine traditional image processing methods with deep learning to detect heterogeneous bodies.Unet++ were used for semantic segmentation.At the input end,the high-gray-order multi-dimensional image is input through frame accumulation and function signal modulation and demodulation,and the light source information is input,so that the network can learn the semantic information of each order.Finally,98.53%,67.71% and 80.89% accuracy are obtained in the detection,segmentation and classification of multi spectral transmission image heterogeneity.In addition,the architecture of Unet++ network is adjusted to improve the segmentation accuracy.The iteration speed of the image is also improved slightlyTMI has the potential value in medical applications.The proposed fast demodulation algorithm and detection method are beneficial to improve the detection speed and accuracy of transmission multispectral imaging,which provide a reference for the future study of breast tumor detection. |
PDF Full Text Request