Design Of Microbial Holographic Microscopic Image Classification System Based On Convolutional Neural Network

As the protection of microbial environment has been increasingly valued,water ecological pollution caused by microbial imbalances has become a universal phenomenon.In water quality detection and sewage treatment,microbial identification is of great significance to the society.Traditional methods are inefficient and require large-scale instruments and manual intervention.In view of the above problems,a microbial digital holographic microscopic image classification system based on convolutional neural networks is designed in this paper.In this paper,a digital holographic microscope is used to collect microbial images,and a convolutional neural network is introduced for classification calculation.Tengine architecture is used to deploy neural network algorithms on the domestic RK3399 embedded platform.The simplified Goog Le Net-Lite model is constructed and implemented to achieve classification on the Microbial holographic image dataset.Compared with traditional methods,this system has advantages in recognition accuracy and speed,and has the characteristics of portability and low cost to meet people's needs.The thesis first analyzes the basic theoretical knowledge of holographic microscopy principles,holographic imaging light paths,and holographic simulation algorithms,and introduces convolutional neural network algorithms and embedded related principles and technologies.Then the hardware platform,software architecture and convolutional neural network classification algorithm of the system are designed.Then,this paper designs the system hardware platform and implements the convolutional neural network classification algorithm of microbial holographic images on the embedded platform.Finally,the system's overall operating indicators were tested and analyzed.The main work and results of this article are as follows:(1)Design of holographic microscopic image classification system: First,introduce the basic working process of holographic microscopic image classification system.Then design the system hardware platform,including the system structure,optical path and image acquisition,computing processing core and user information interaction.Then design the system software architecture and explain its software operating rules.Finally,the core algorithm of the system,the classification algorithm of convolutional neural network,is analyzed,including the establishment of data sets,the holographic calculation of optical microscopic images,the comparative analysis of classic network models,and the Goog Le Net network clipping optimization.(2)Implementation of holographic microscopic image classification system: First,design the overall operating framework of the system.Then build the system hardware platform,including the holographic microscope light path and the main controller.The detailed description of the embedded system software platform construction,including system firmware programming,system environment configuration,embedded platform Tengine network framework construction,and ACL acceleration library transplantation.Then train the convolutional neural network algorithm on the PC platform.Finally,a specific method of transplanting convolutional neural network algorithms for embedded platforms is designed.(3)System index test and result analysis: It mainly tests the algorithm operation index in different environments,the embedded platform algorithm operation efficiency analysis,and the system self-collection data set test.Under the test of different environments,the Goog Le Net-Lite network model has obtained higher accuracy and computation speed.On the embedded platform,Goog Le Net-Lite achieves an accuracy rate of 94.15% while ensuring a high frame rate and meeting system requirements.Finally,compared with traditional methods,this system has superiority in recognition speed,portability and cost control.