Activated Sludge Microorganism Image Classification And Target Based On Deep Learning

With the increase of population,the rapid development of economy and the improvement of people’s living standard,the water consumption of human beings increases rapidly,and the amount of sewage produced also increases rapidly.At present,most sewage treatment plants use activated sludge method to treat sewage.The study shows that the quantity and activity of microorganism population in activated sludge can predict the effluent quality of wastewater treatment process very well.Therefore,it is of great significance to master the species and quantity of microorganism in activated sludge in time for guiding wastewater treatment process.The traditional method of microorganism detection is to identify microorganisms manually by microscopic examination,which is inefficient and requires the skilled professional knowledge of the detection personnel.With the rapid development of artificial intelligence technology,it has shown great potential to use machines to automatically identify microbe microscopic images.The main research work of this paper includes the following three aspects:（1）In view of the current microorganism image data set of activated sludge without open source at home and abroad,this paper obtained the original microorganism image of activated sludge by taking water samples in the sedimentation tank of sewage treatment plant and taking microscope photos.Six protozoa and metazoa such as Rotifers,Euplotes,Peranema trichophonrum,Vorticella,Litonotus,Nematode were identified by professionals.For the classification task,each single image is obtained by intercepting the original image,and the data set is expanded by data enhancement method.For target detection task,data enhancement method was used to expand the number of original images,and Labelimg software was used to label the target microbes in the original image and generate the corresponding XML label file.（2）For the task of microorganism image classification of activated sludge,classical convolutional neural network classification models Alex Net,VGGNet16 and Res Net50 are used to achieve end-to-end classification of microorganism images of activated sludge,and the problems existing in using classical convolutional neural network classification are analyzed.An improved Res Net50 classification model based on transfer learning and attention mechanism is proposed.The experimental results verify that the proposed model has better classification effect by comparing the accuracy rate,accuracy rate,recall rate,and harmonic mean of accuracy rate and recall rate before and after the improvement of the model.（3）In view of the scene with complex image background,multi-target and multi-species microorganisms,this study proposed a target detection model of lightweight YOLOv4 activated sludge microorganism image based on attention mechanism on the basis of single-stage target detection algorithm YOLOv4.The main feature extraction network of YOLOv4 model was replaced by lightweight network Mobile Net,and the 3 × 3 standard convolution of the rest parts of YOLOv4 model was replaced by deep deprivable convolution to reduce model parameters,and the attention mechanism module of the network was added to strengthen important features.The experimental results show that the improved model has smaller memory,faster detection speed and higher detection accuracy,which meets the requirements of lightweight,rapid and accurate detection of activated sludge microorganism image target detection task.