Design And Implementation Of An Image Classification System Based On Saliency Detection And Neural Network

Over the years,artificial intelligence has been a hot topic for scholars,and computer vision is a very important part of artificial intelligence.It refers to use images or videos as input and convert them into digital signals by computer for further in-depth the study.Computer vision has its wide range of applications,such as face recognition,video surveillance analysis,image recognition analysis,assisted autopilot,three-dimensional images,etc.,which are almost full of all aspects of our lives.Image classification,as part of computer vision,naturally has its necessity for research and has a certain impact on other areas of computer vision.In order to study the accuracy of image classification,the system combines manifold ranking,saliency detection and convolutional neural networks,and mainly uses Python and deep learning libraries such as tensorflow and opencv for development.First,the system's requirements are analyzed.Firstly,the system's demand analysis is carried out,and then the use case diagram is given according to the functional requirements of the system.The whole system is divided into three levels: data layer,logic layer and interface layer.The logic layer is divided into four modules: image acquisition,saliency detection,neural network classification and classification display.The data acquisition includes three methods of local picture input,camera shooting input and folder input.The system is designed in detail according to the functions of each module.The core algorithm is implemented by training on MSRA,CIFAR-100 and other data sets to select the most suitable parameter values,and finally the whole system is realized.Users can read pictures from local albums,or use the camera to take photos,or enter a folder,then click the sort button to sort the pictures,the system will display the pictures and categories and probability to the user,the user can click save to save the results in the folder of the corresponding category of the picture.After testing,the function of the system has been well realized,and it has good adaptability in terms of non-functionality such as running speed,illumination and shape.