Research And Implementation Of X-ray Detection Of Pneumonia Based On Deep Learning

Globally,millions of people are diagnosed with lung disease each year,and four million of them die from pneumonia.As the threat of pneumonia to people’s health becomes more and more serious,traditional pneumonia diagnosis is gradually replaced due to low efficiency,misdiagnosis and missed problems,so researchers began to introduce deep learning technology into pneumonia detection to solve the above problems.However,due to the small target,low resolution and insufficient image information in the pneumonia lesion area in the chest X-ray image,the detection of pneumonia is difficult,and in view of these problems,this paper studies the pneumonia X-ray image detection algorithm based on deep learning.The main work contents are as follows:1.In view of the problem of false detection and missed detection in pneumonia detection,several optimization strategies are proposed on the basis of the Faster R-CNN model.Firstly,the feature extraction network VGG16 in the Faster R-CNN model is replaced with ResNet50 to improve the accuracy of pneumonia detection.Secondly,optimize ROI Align to further strengthen feature extraction.Finally,Soft-NMS was introduced to reduce the missed detection rate of the model in the area of pneumonia lesions.Experimental results show that the detection accuracy of the improved model Faster_RAS is 91.4%and the speed is 74FPS,which is better than the original Faster R-CNN model.2.In view of the problem of slow detection speed Faster RAS the above improved model,this paper selects the YOLOv5 model to further improve its performance,and proposes several optimization strategies.Firstly,the CBAM attention mechanism is introduced in the YOLOv5 model to strengthen the extraction of pneumonia characteristic information by the model.Secondly,the idea of bidirectional cross-scale connection of BIFPN network is learned to optimize the feature fusion structure.Finally,it is proposed to introduce the SIOU loss function to optimize the bounding box regression loss to improve the model training speed,so that the improved model can meet the requirements of real-time and accuracy of pneumonia detection,and the improved model is named YOLOv5_CBS.The experimental results show that the detection accuracy of the improved model YOLOv5_CBS is 91.2%,the detection speed is 120FPS,the detection progress is 5.5 percentage points higher than the original YOLOv5 model,the speed is 31FPS faster,and the detection accuracy and speed are improved.3.In order to make the research results operable in practical applications,design and complete a pneumonia detection system to meet the needs of doctors to easily and quickly detect pneumonia lesions and assist in the treatment and diagnosis of pneumonia.