Research On Breast Tumor Target Detection Technology Based On Deep Learning

Breast cancer,also known as breast cancer,is a common disease.In recent years,its morbidity and mortality are on the rise.It is almost impossible to recover in the late stage.Early detection helps to improve the survival rate of patients.This paper mainly studies the application of target detection network in depth learning in the localization and classification of breast cancer lesions,mainly through the following aspects:Firstly,the background of mammary gland image is removed,and the coordinates and classification information of diseased area are obtained according to the doctor's label information.Through these information,the diseased area image is cut down,and the number of diseased area images is increased by image fusion technology.Random clipping technology is used to make the number of diseased grade images equal,and the data set of diseased area is fused;secondly,the residual network is combined with the residual network.Third,the Single Shot MultiBox Detector(SSD)network was selected and improved to improve the SSD network,and then R-MobileNet network was used to replace the benchmark network to improve the SSD network to obtain the SSD-II network.Fourthly,SSD-II network is obtained by fusing image of lesion area.R-MobileNet is pre-trained,and then R-MobileNet after pre-trainingis used as the base network of SSD-II network.Fifth,background image is loaded into SSD-II network for training,and several other target detection networks are used as comparative experiments.The final results show that the proposed network is the best,which shows that the proposed benchmark network is effective.At the same time,it proves that the use of pre-training is helpful to improve the network performance.The model in this paper is helpful to the doctor's assistant diagnosis.