Image Classification Method Based On Deep Learning With Multiple Features And Its Application

Classification method is an important research task in data mining and machine learning,and image classification is a hot research direction of image mining.Image classification usually consists of three parts:feature extraction,feature expression and classifier.However,the performance of the classification system is largely determined by the effectiveness of the extracted features,which can improve the performance of image classification.In practical application,it can rely on the prior knowledge of related fields to achieve effective feature extraction,but it takes a lot of time,and it is difficult to effectively express image information.With the emergence of Deep Learning(DL)method,it has successfully solved the problem of image analysis and processing,and also provided an effective method for image classification.As a deep neural network model,Deep convolution neural network(DCNN)has good performance in image recognition,motion analysis,medical diagnosis and many other fields.However,when DCNN is applied to complex scene image classification,its model performance still depends on whether the existing network can accurately migrate to a specific image.For this reason,this paper proposes a multi feature fusion image classification method based on deep learning.This method is based on DCNN and image texture features,aiming at improving the accuracy of image classification,combined with the image processing classifier to establish a multi feature fusion model.The specific contributions and work of this paper are as follows:(1)The characteristics of DCNN model are analyzed,and the inception network model is used as the basic model to realize DCNN.Using image data to train the whole network model for learning,and evaluate the performance of the classification model.When DCNN is applied to image classification in different fields,especially when the image data is insufficient and the training sample size is too small,the classification performance of the model is not stable.The natural life image and small size of sample image data are used to train the model,the feature extraction part of the front level of the natural life image training model is adopted,and then the image data of a small number of training samples for model learning training is introduced,which help to make use of the existing data and improve the accuracy of the model for small sample image data classification.(2)Image features are extracted by scoring feature,Gray-Level Co-occurrence Matrix and Histogram of Oriented Gradient.Support Vector Machine and Extreme Gradient Boosting classifier are trained for classification.Making full use of the advantages of DCNN in image feature processing,DCNN is selected as the feature extraction tool to grade the classification of image.The feature fusion of scoring feature and image texture feature is used as the input of classifier.(3)The classification performance is evaluated by Accuracy(ACC),Precision(Pre),Recall(Rec),F1-score(F1),and Overall Accuracy(Overall ACC).The results show that the performance of the multi feature fusion image classification method proposed in this paper is the best.Compared with the support vector machine algorithm,the performance of the polar gradient enhanced classifier is better.Its overall accuracy is 92.80%,and the accuracy of a single category is 84%,which shows that this method is helpful to classify a certain category of images more accurately.