Research On Online Laos's National Characteristics Recognition Model Based On Deep Learning

As an international student from Laos,after studying in China,I deeply felt that the speed of image processing in Laos is relatively slow.Image features are inherent spatial attributes commonly found in objects,and the extraction of texture features,color features,and text features are the prerequisite steps for the processing,analysis and application of the online version of ethnic characteristic images.Therefore,extracting effective and reasonable feature values from the image is the key to accurately identify the image information.Image inherent feature analysis technology is widely used in image recognition,classification,segmentation,synthesis,retrieval and other texture-based image digital processing,so it has always been an active research content in the field of digital image processing.Among them,in the field of ethnic characteristic recognition,although many related excellent technologies and algorithms have been derived,the GLCM(Gray level co-occurrence matrix)algorithm based on the probability and statistical analysis method is used to extract the characteristics of ethnic characteristic images and Analysis is still a research field that has not yet been outdated,and the use of neural network algorithms for model deep learning is also a current hot spot.A visual application platform is a container that carries application functions.Only an application platform with portable access,stable performance and rich functions can be favored by users.Therefore,the design of a network-based humanized image processing and recognition platform with ethnic characteristics is the design of this paper Original intention.The main research content of this paper is based on the C++ language and the visualization application development tool Xojo,the web version of the Lao national characteristic image,the use of image feature extraction which includes headwear,earrings,clothing and characteristics,ethnic characters,ethnic artifacts,and ethnic buildings based on the co-occurrence matrix algorithm and the RBM-BP(Restricted Boltzmann Machine-Back Propagation)neural network algorithm Image feature recognition and analysis,and complete the establishment of a network version of the image recognition platform,that is,the ultimate goal is to deploy the application to INTENET,users can access and operate at any location through the browser software of any smart phone or computer device,The feature recognition function for any grayscale or color texture image.Among them,the texture feature parameter combination optimization based on weight sorting is proposed,so that the 15 extracted image feature parameters can be reasonably and effectively used.The recognition algorithm uses the RBM-BP neural network algorithm.The design research found the shortcomings and shortcomings of the traditional gradient descent RBM algorithm in the image feature recognition process.The RBM algorithm is optimized and the multi-threaded divergence contrast RBM algorithm is used for recognition.The comparative experiment verifies the feasibility and efficiency of the optimization method,and improves the recognition performance of the system.This paper selects the image feature parameters of 37 sample ROIs of 4 types of image features as the training set,and finally tests 4 types of 7 images,the recognition rate of all kinds of sample images is above 97%.The design of this paper has been successfully completed to build a network version of the national characteristic image recognition and processing system.After the experimental test of 4 types of texture images,the system interface and functions have reached the design index requirements.