Design And Implementation Of Object Classification System Based On Deep Learning

In recent years,with the development of artificial intelligence,more and more intelligent products based on deep learning are constantly into our daily life.Mobile intelligent software is undoubtedly an important development direction of artificial intelligence industrialization.The object classification system designed in this thesis is an application software which can run offline on the mobile terminal,recognize the real objects in the real scene,and feed back the corresponding English names of the objects in real time.This thesis introduces the development process of the object classification system through the description of the system demand analysis,design and implementation of mobile terminal system software,system testing and analysis.In the system requirements analysis process,firstly,the business process of the whole system according to the requirements is introduced.Then,according to the business process,the system is divided into different business functions,and use case analysis is performed for each business function.In addition,non-functional requirements analysis,data analysis and process modeling are also carried out.For the construction of object classification system,this thesis firstly analyzes and designs the overall framework of the system.Then the structure and function of image acquisition module,model call module,image classification module,result feedback module and system interface module are designed respectively,the design of image classification model is described in detail.In the design process of image classification model,this thesis selects VGG16 image classification model as the benchmark model,then analyzes and evaluates the benchmark model according to the requirements of the system,combines with the characteristics of the system,by making some improvement based on the benchmark model,the image classification algorithm model meets the requirements of this thesis is designed.According to the design architecture of the system,this thesis implements each module of the physical classification system in the form of Android system,and describes the realization process of the system in detail in the form of combining pictures and texts.In addition,in order to make the system independent from the constraints of network environment and have a faster processing speed,the image classification model needs to be transplanted to the Android system for offline operation.Therefore,this thesis chooses Xiaomi MACE framework to transplant the image classification model,and the image classification model is embedded into the Android system in the form of MACE for offline operation.Finally,the construction process of MACE library is introduced in detail.In the process of testing,the image classification model in this thesis is compared with the VGG16 model.Through the analysis of test results,the image classification model in this thesis is evaluated.Then the function of the whole system and each module are tested in detail,which proves that the object classification system and each functional module in this thesis can operate normally,can identify the real objects in the real scene well,and can feedback the corresponding category labels of the objects in real time.Finally,the performance of the system is tested,and it is proved that the system can meet the requirements in terms of performance.After a series of work in this thesis,the object classification system has been developed.A mobile app which can run off-line in mobile terminal and recognize objects in real scenes is realized.This software can bring certain convenience to users' social life,and satisfy their needs of object classification software in real life.Meanwhile,the image classification technology has been successfully applied in reality.