Research On Improved Particle Swarm Optimization Algorithm In Segmentation Of Fundus Haemorrhages

Image segmentation is a very important image preprocessing means,and it is also one of the prerequisites and basic work of computer vision and artificial intelligence.According to statistics,due to the impact of people's lifestyle,the incidence of diabetes is very high,and diabetic retinopathy which occurred in the late stage of diabetes can cause bleeding in the fundus,therefore,through the image segmentation technique to detect the fundus haemorrhages can effectively assist the doctor's treatment and greatly improve the efficiency of doctors.In recent years,there are many researches on image segmentation.However,the algorithm of image segmentation mainly focuses on the whole region segmentation and template matching segmentation.The features of the fundus hemorrhages are weak and the shape and rules are not uniform,therefore,image segmentation algorithm is difficult to achieve satisfactory results in medical applications.However,the current segment method of fundus hemorrhages is mainly depend on the traditional machine learning and deep learning the two aspects,however,the traditional machine learning and deep learning requires a lot of manual marking,feature selection,adjustment parameters,so not only spend a lot of time but also the effect is limited.In this paper,a large number of learning and research on diabetic retinopathy and fundus hemorrhages were studied.The particle swarm optimization algorithm was used to find the clustering center of the haemorrhages and the rectangular areas where the haemorrhages were located.Then,the active contour model segments the contours of the fundus hemorrhages.In this paper,three improvements are proposed.Firstly,the particle swarm optimization algorithm based on Metropolis criterion is proposed for the problem that the particle swarm optimization algorithm is easy to fall into the local optimal solution.Secondly,a new feature similarity coefficient is constructed by combining the edge,region and shape features and introduce fuzzy clustering to construct a new particle swarm fitness function for the clustering of haemorrhages.Thirdly,the localizing region-based active contours based on particle swarm clustering and adding clustering coefficients is proposed to make the initial contour closer to the target edge for the presence of today's active contours model which have the problem of slow segmentation and sensitive to starting contours.Through the Matlab simulation experiments,the improved particle swarm optimization and improved localizing region-based active contours is improved in stability,speed,precision and the ability of anti-noise.In addition,the hemorrhages segmentation algorithm can accurately segment the contours of the haemorrhages and superior to the present machine learning-based fundus hemorrhages detection algorithm in terms of manual consumption,sample requirements,and segmentation accuracy.