Pneumonia Detection And Identification Research Based On Improved SSD

Pneumonia is a common disease in the chest,and its treatment depends to a large extent on the doctor's accurate interpretation of the chest image.But even the best radiologists can easily misdiagnose,because the diagnosis of multiple diseases based on chest imaging is a very difficult problem.In order to assist physicians in accurately interpreting information on chest images,domestic and foreign scholars have used computer image recognition technology to conduct in-depth research in this field.At present,the best method for detecting chest image recognition is based on deep learning pneumonia recognition.In view of the difficulty in detecting and identifying pneumonia,this paper proposes an improved scheme based on the SSD(Single Shot MultiBox Detector)pneumonia detection model.First,by analyzing the data set,it was found that there were about 10% of small diseased areas in the chest X-ray image.In order to let the SSD network model detect these small diseased areas,the output of the last convolutional layer of the third network layer of the original SSD network model is reduced by the pooling method and then used for 1×1 convolution.The convolution operation then fuses the convolution operation result with the input features of the sixth feature layer of the SSD network.In addition,the output of the fifth pooled layer of the SSD network model is convolved by the anti-pooling and deconvolution techniques and then merged with the seventh feature layer of the SSD network model.By using these two fusion schemes,the SSD target detection network is able to detect small targets in lung disease images.Then,this paper selects the correct optimization scheme through experimental analysis for the improved network model,so that the network model can achieve convergence effect faster and improve the detection accuracy.Finally,in order to test the detection and recognition performance of the improved SSD network model for pneumonia,this paper used the lung disease image dataset to perform experiments on Faster RCNN,Yolo,original SSD network model and improved SSD network model.Through the analysis of the experimental results,it is found that the improved SSD model has a detection accuracy of 89.32% for pneumonia diseases,which is greater than the original SSD network model with an accuracy of 76.1%.At the same time,the improved detection accuracy is greater than the accuracy of the Faster RCNN and Yolo methods.From the experimental results:(1)The effectiveness and feasibility of the improved SSD network model for the detection performance of pneumonia diseases.(2)The improved SSD model in this paper can have some guiding significance for the detection of lung diseases.(3)Researchers in the medical field can use this improved SSD network model to help overcome the inherent limitations of human cognition and bias and reduce the rate of misdiagnosis.