Breast is an exocrine gland of human which is used to breed offspring,composed of skin,fibrous tissue,breast glands and fat.Breast cancer is a malignant cancer,the incidence of breast cancer in Chinese women increased year by year,it has been a serious threat to the health of Chinese women.Using near-infrared spectroscopy for breast detection,is a relatively new way for breast examination.The interpretation and diagnosis of near-infrared mammography are subjective,and subjective differences in doctors during interpretation are inevitable.Therefore,using information technology to recognize the tumor in near-infrared mammography automatically for reference,will effectively improve the accuracy of breast examination.In this paper,convolution neural network and support vector machines were applied to recognize the tumor in near-infrared mammography automatically.The main work and achievements are as follows:1.Extracted several texture features based on gray-level co-occurrence matrices in different regions of near-infrared mammography,then put these features into support vector machine for training,then predicted every pixels in the pictures which waiting to be indentified,finally got the segmentation of the tumor,this entire process was automatic.2.Designed various shallow Convolution Neural Networks(CNNs)against different input training patches with different sizes.After training,predicted every pixels in the pictures which waiting to be indentified,and got the segmentation automatically.This method does not need to extract the feature manually,and its segmentation effect is better than the SVM segmentation,and the processing speed is greatly improved.3.Proposed automatic segmentation of the near-infrared mammography based on residual networks.The effect of this segmentation is better than that of the shallow convolution neural network,and the velocity is better than that of the support vector machine.4.Took a series of image processing,then used the loaction and size of the tumor area to achieve the automatic identification of the tumor in near-infrered mammography.The experimental results show that the automatic recognition of near infrared breast images based on residual network can effectively identify the tumor area in the image,and has excellent stability and accuracy. |