Image Segmentation Of Retinal Blood Vessels Based On U-net

The continuous development and progress of society,heavy work pressure and decline in physical fitness have led to an increasing number of people suffering from diseases caused by retinopathy.Therefore,the analysis and accurate detection of retinal vascular structure is considered to be the primary task for the prevention of fundus diseases in large-scale populations.In the diagnosis of ophthalmic diseases,segmentation of retinal blood vessels is a very effective method.The computer algorithm can quickly and accurately obtain the retinal blood vessel image,and it is objective,rational,and low in cost,which is of great significance to the treatment and observation of ophthalmic diseases.In the use of the method,due to the restriction of the image acquisition equipment,the contrast of the captured retinal blood vessel image is low,and some small blood vessels are too bifurcated and the composition is complicated.In order to solve these problems,the accurate and fast segmentation of retinal blood vessel images has become a key research direction in the current academic circle.In response to these problems,this paper proposes an improved retinal vessel segmentation algorithm based on residual network and attention gate mechanism and an improved retinal vessel segmentation algorithm based on Ladder-net and RAU-net.The main research contents are as follows:(1)Based on the improvement of the U-net network structure proposed by the fully convolutional network,a new retinal vessel segmentation algorithm RAU-net is proposed to solve the problem of retinal vessel image segmentation.In order to solve the problem that the more convolution operations,the deeper the network,and the degradation of network performance,we have introduced a residual module in the U-net network.In order to suppress unnecessary feature information at the decoder stage,we added an attention learning module to improve the accuracy of blood vessel image segmentation.While introducing two kinds of improvement modules,the two are skillfully combined together,so that the two improvement methods are brought to a higher level,and the improvement effect is more obvious and outstanding.(2)Based on RAU-net for further improvement,combined with Ladder Net,a new segmentation algorithm Lad-RAU-net is proposed.On the basis of the RAU-net network architecture,the multi-pair encoder-decoder structure in Ladder Net is introduced,and the two RAU-net structures are connected in parallel to extract more feature values.The attention mechanism gate is used to replace the ordinary summation mechanism to eliminate noise and improve the accuracy of blood vessel image segmentation.By comparing three new modules with different composition content,a new residual module is used to replace the ordinary module,the features are captured more carefully,and the segmentation accuracy is greatly improved.Through the experimental training of these two methods on the three public retinal image data sets of DRIVE,STARE and CHASE＿DB1,the proposed two network model methods based on the improved U-net structure can effectively complete the background and non-retina vascular image.Compared with the traditional unsupervised algorithm and the supervised algorithm,the segmentation effect of the background area is better and the effect is better.