Research On Recognition Detection Technology And Application Based On WebAR

In recent years,WebAR technology has been widely used on mobile devices,including in fields such as museum exhibitions,enhanced tourism,campus navigation,and education and training.However,in practical applications,WebAR still faces some challenges in target recognition and detection.Firstly,the computing power of mobile devices is insufficient to meet real-time requirements.Secondly,the problem of low detection efficiency arises due to target occlusion and the appearance of multiple targets in the real environment.In order to solve these problems,this paper uses a deep learning-based method to study target recognition and detection in WebAR applications.The paper mainly completes the following work:(1)A deep learning-based VD-Mobile Net image classification network model algorithm is proposed.This algorithm introduces dilated convolution with an expansion rate of 2 into the deep convolutional layer of the Mobile Net network model without increasing the number of parameters,and applies it to the first few deep convolutional layers to improve the feature information and classification accuracy of the images.In addition,the Leaky Re LU activation function is optimized to preserve negative features and avoid information loss.This paper conducts experiments using a garbage classification dataset,and the results show that the model performs well in terms of classification accuracy,while significantly reducing computational complexity,effectively addressing the problem of limited computing power of mobile devices.(2)A WebAR object detection solution with lightweight multi-scale feature fusion is proposed.Firstly,this paper replaces the original feature extraction network in the SSD framework with the Mobile Net network,resulting in a significant reduction in model parameters and computational complexity,ensuring a lightweight network and fast detection speed.Secondly,a multi-scale feature fusion module is introduced,utilizing Kronecker convolution to effectively capture more feature information.Additionally,a feature pyramid network is introduced to merge the two,better integrating detail and semantic information,reducing unnecessary information loss,and further improving precision and accuracy.Finally,experimental results demonstrate that the improved model in this paper achieves good detection performance on the PASCAL VOC2007 and VOC2012 test datasets.(3)Combining the VD-Mobile Net and multi-scale feature fusion lightweight recognition and detection model proposed in this paper,a WebAR system based on deep learning is built,which can accurately recognize and detect target objects.Experimental results and user evaluations show that the system proposed in this paper has significant application value.