Detection Of Funds Blocky Abnormal Information

With the development of fundus image acquisition technology,clear fundus images can be easily obtained.And with the development of the times,the scenes in which people obtain information through visual systems in modern urban life have become more diverse,so the health of the visual system has become particularly important.The development of technology and the importance of visual system health make the detection of fundus block anomalies based on machine vision of great significance for ophthalmologists to assist in diagnosis.Due to poor quality of fundus images,uneven lighting,and the presence of various tissue structures,such as blood vessels.The above factors can interfere with the extraction of fundus blocky abnormal information.In order to eliminate interference and effectively detect eyeground blocky abnormal information,this paper analyzes the characteristics of eyeground blocky abnormal information in the gray distribution curve,and combines the direction space theory to carry out the following parts of research:(1)In terms of feature analysis of blocky abnormal information,the grayscale features of G-channel fundus images were analyzed.The specific steps for proposing a detection scheme based on the characteristics of blocky abnormal information and fundus interference are as follows: 1.Fit the fundus background curve;2.Boundary positioning of convex line segments;3.Extraction of bright fundus targets;4.Extraction of fundus blocky abnormal information.(2)In terms of fitting the fundus background curve,a method based on the least squares method has been proposed.(3)In terms of boundary localization of convex line segments,a method based on quadratic combination line segments has been proposed for boundary localization of convex line segments,and a method for boundary localization of convex line segments in filtered images by combining line segments has been proposed.(4)In the aspect of target extraction,a method of extracting bright targets by fitting curve difference threshold is proposed,and based on a method of extracting blocky abnormal information according to the difference between blocky abnormal information and the fluctuation of eyeground background surface,the blocky abnormal information is extracted.Finally,the algorithm was tested using fundus images from the DIARETDB1 dataset,as well as corresponding testing protocols and diagnostic standards.The test results show that the diagnostic sensitivity of the algorithm is 81.80%,the specificity is 74.07%,and the average detection time for one fundus image is 1.24 seconds.The results indicate that this algorithm can have high sensitivity and specificity under testing on standard datasets.Compared with deep learning,this algorithm has strong interpretability,fast operation speed,and does not rely on datasets.Compared with traditional algorithms,this algorithm has advantages such as high detection sensitivity,specificity,and strong robustness.