Development And Implementation Of Remote Identification System For Wood Based On Images

Timber smuggling seriously damages the normal economic order and affects the development of China's forestry industry and ecological construction.Correct and rapid identification and diagnosis of timber species is the key to law enforcement supervision by customs.Due to the wide variety of timber species and the small difference in appearance of some timbers,manual identification has problems such as low efficiency and error-prone.At present,the research results of automatic recognition of wood based on images have not been widely used due to the problems of few wood species recognition,low accuracy of recognition model and low robustness.In response to the above problems,this paper researches a deep learning-based wood cross-section macro-construction image recognition method and develops an image-based wood remote identification system(APP and Web terminal),which realizes the information query of 120 common types of wood,automatic wood image recognition,and expert remote identification,providing a fast wood species identification aid for customs timber inspectors and greatly saving economic and time costs.The main research contents and results are as follows.(1)Research on wood image recognition algorithms based on deep learning.In this paper,four common deep convolutional neural network wood recognition models,Caffe Net,Goog Le Net,VGG16 and Res Net101,are trained for 120 common wood cross-section macro-constructed images;test results show that Res Net101 wood recognition model obtains 93.1% recognition rate,and the recognition rate for 24 approximate species of wood is only 82.2%,and the false alarm rate was 15.4%.In order to improve the recognition rate of wood and reduce the false alarm rate,a second layer of bilinear fine-grained image recognition model was added on the basis of the recognition results of the Res Net101 model,and the test results showed that the average accuracy of the Res Net101+DBTNet two-layer recognition model for 120 species of wood recognition was improved to 94.5%,and the false alarm rate was reduced to 9.8%.(2)Design and implementation of client-side APP for image-based wood remote identification system.The client-side APP is developed using Android language,data transmission using Ok Http and image browsing using Glide implementation.The client APP consists of five functional modules: frontier lookout module,knowledge dictionary module,intelligent identification module,online support module and user center module.Users can browse or query timber information and check and issue case information in the client APP.Use the mobile client to shoot images or upload existing images to the server for recognition and can display the recognition results in real time.If the identification result does not match with the expectation,you can request expert remote identification through graphic or audio-video call to improve the accuracy of wood species identification results.(3)Design and implementation of the server side of the image-based wood remote identification system.The server side consists of four parts: web side,database,application server and image recognition server.web uses Vue to realize the separation of front and back ends,the server side development framework uses Spring Boot,and the database uses My SQL.the web side mainly realizes the functions of timber species information maintenance,expert remote identification,and timber image review and storage;the database stores user data,timber information The database stores user data,wood information,identification history and online support data;the application server is used to process user requests,access the database and call the image identification server.