Image Recognition And Analysis Based On XGBoost-CNN Model

The convolution neural network model,called CNN,is a classic deep learning model,and has made remarkable achievements in the field of image recognition.The image data is different from other data,it is difficult to extract features by human understanding.CNN can directly learn from input data through the design of convolution layer and pool layer.It not only avoids data preprocessing,but also uses spatial structure information to reduce parameters and improve training efficiency.So the CNN can be widely used in the field of image recognition.But the CNN also has many shortcomings,such as complex structure,large amount of calculation,high training cost,long training time and so on.The parameters of the CNN model are numerous,and the process of adjusting the parameters is complicated and time-consuming.In practical applications,it takes a lot of time and energy to train a CNN model with high accuracy.However,it is much easier to train a CNN model with accuracy better than a random guess.From the Boosting algorithm,we can know that weak classifiers learn from errors and reduce the probability of mistakes by iteration,so many weak classifiers can be combined into a strong classifier.Therefore,the main idea of this paper is based on the Boosting algorithm,combining multiple weak CNN models to achieve higher accuracy.There are many boosting algorithms,and XGBoost is the shorthand for eXtreme Gradient Boosting which has fast training speed by taking two order Taylor approximation for the loss function.Therefore,this paper proposes XGBoost-CNN model by taking CNN model as basic classifier of XGBoost,and then we explore XGBoost-CNN's classification effect in image recognition.In this paper,we identify and analyze the liver cancer image dataset.In order to improve the training speed,we also use multithreaded parallel,binary file storage,data enhancement,migration learning and other improved methods.It is found that the test accuracy of the XGBoost-CNN model in liver cancer image dataset is 86%,and it has obvious advantages compared with the Random Forest and XGBoost.In conclusion,the XGBoost-CNN model can be effectively applied to the problem of image recognition.