Study On Application Of Images Classification And Detection Based On Multi-feature Fusion

Images classification and detection is a computer technology that can effectively detect and identify specific target objects by processing,analysing,and understanding images.There are two foundational questions in this field that how to extract effective information from images and what kind of classification method should be adopted.Besides,whether the organic combination of multi-feature can essentially improve the accuracy of image classification and detection is another problem should be discussed in-depth.In the widely accepted image classification system,single texture feature extraction methods such as Local Binary Patterns,Gray-Level Co-occurrence Matrix,Gabor kernel are all frequently mentioned.Besides,the extraction of texture feature and the selection of classification methods have direct impacts on the classification performance.Hence the selection of the feature extraction and classification methods reveal their certain limitations as the image classification in the previous period was generally engaging with a specific classification task.This paper presents an image classification method of multi-feature weighted fusion which extracted different texture features for weighted fusion from grey images,then conducted a classification experiment on the texture database,aims to carry on a comparative study between results of the classification of the fusion features and the single feature classification results,gets the accuracy of the former is higher.Meanwhile,the Gradient Boosting Decison Tree method is adopted to classify the face data sets,and the practicability of the method is proved.In colour texture images,desired experimental results are achieved when comparing a new feature fusion method based on HSV space,classified by eXtreme Gradient Boosting based on colour texture database,with the experimental results displayed in the official website of the dataset.Furthermore,the objects detection is a heated research topic in the deep learning field.The Single Shot MultiBox Detector network shows better on speed and accuracy.This SSD-network-based study,respectively taking Resnet network and VGG network as pre-networks,under the convolutional feature fusion,conducting target detections for convolutional features extracted from different convolution layers.In this paper,we compare the performance differences between two deep networks be taken as the front network through practical applications,and apply objects detection method to food calorie detection.