Detection Of Saccharomyces Cerevisiae Cells Based On Neural Network

Microbial infection can be divided into bacterial infection,viral infection and fungal infection.Diseases caused by microbial infection are the most common diseases in human beings,and also the most infectious diseases in human beings.Invasive saccharomyces cerevisiae(S.cerevisiae)infection is a very typical type of invasive fungal infection,such as unexplained fever,pneumonia,fungemia,liver abscess,peritonitis,vaginitis,urinary tract infection or septic shock after taking yeast preparation.The operation process of common clinical detection methods for pathogenic bacteria is not only tedious and time-consuming and requires expensive high-precision instruments,but also may contaminating test samples during operation.Therefore,the establishment of a new rapid detection method of S.cerevisiae is of great significance for timely prevention and rapid diagnosis of the disease.In this paper,the detection of invasive S.cerevisiae infection diseases is taken as the specific research object,and the deep learning method is used to detect Saccharomyces cerevisiae cells in blood microscopic images,so as to realize the rapid and accurate detection of S.cerevisiae cells in microscopic images.The specific contents are as follows:(1)The basic structure of convolutional neural network(CNN)is briefly introduced,including a basic introduction to convolutional layers,pooling layers,activation functions,fully connected layers,and loss functions,followed by a description of the structure of two convolutional neural network models,Res Net and Ghost Net,and their subsequent use.(2)In order to simulate the real environment in which S.cerevisiae cells invaded into the blood,the culture and activation of S.cerevisiae cells was completed in the laboratory,and the S.cerevisiae infection samples were prepared with blood donated by volunteers.The sample image was collected by microscope,and the S.cerevisiae cells in the image were labeled by labeling software,and the S.cerevisiae cell infection data set was built.The feasibility of the detection method is verified by using the SSD algorithm to detect microscopic images of blood cells mixed with S.cerevisiae cells.In order to improve the network performance,first the feature extraction network of the SSD algorithm was replaced by using Res Net50 convolutional neural network as the feature extraction network of the SSD algorithm,and then a batch normalization layer was added after each convolutional layer on five additional feature layers.A higher accuracy S.cerevisiae cell detection algorithm was constructed by the above method,with an average precision AP(Average Precision)of97.70% and a detection speed of 0.31 s for a single image on CPU,which is 1.88% higher than the average precision of the original SSD algorithm,with no significant difference in detection speed.(3)The YOLOv3 algorithm is improved to improve the detection accuracy and detection speed and reduce the number of model parameters.First,in order to reduce the number of model parameters and improve the detection speed,the feature extraction network of YOLOv3 algorithm is replaced by using Ghost Net,a lightweight convolutional neural network,to replace Dark Net53 as the feature extraction network,and some convolutional layers of the YOLOv3 network are modified to depthwise separable convolution,which reduces the number of model parameters to about one-fifth of the original YOLOv3 and improves the detection speed.To improve the detection of small targets,the receptive field of Ghost Net is increased by using dilated convolution;and the Leaky Re LU activation function is replaced by the Hard Swish activation function to gain better performance.In addition,to locate S.cerevisiae cells more accurately,the CIo U loss function is applied to the bounding box regression,which can directly minimize the distance between two bounding boxes to obtain faster convergence and better regression results.Finally,by adding the attention mechanism CBAM to the network structure,it enables the network to better focus on the key features.The microscopic images were verified by ablation experiments,and the experimental results showed that the improved YOLOv3 improved the detection accuracy and detection speed compared with the original YOLOv3,with an average accuracy AP of 97.96% and a detection time of 0.021 s for a single image on GPU.In addition,compared with classical object detection methods such as Faster-R-CNN,SSD,Retina Net,Efficient Det and the Res Net50-based SSD algorithm used in the third chapter,the detection accuracy of the improved YOLOv3 is higher and its detection speed is faster,which further reflects the superior performance of the improved YOLOv3.To sum up,in view of the shortcomings of common pathogenic bacteria detection methods,such as tedious and time-consuming,this paper puts forward a detection method of invasive Saccharomyces cerevisiae infection diseases based on neural network,which realizes the rapid and accurate detection of Saccharomyces cerevisiae cells in blood microscopic images,provides a new and potential method for detecting invasive Saccharomyces cerevisiae infection and other related diseases caused by microorganisms,and plays a positive role in developing and perfecting the application of deep learning technology in the field of cell detection.