| At present,target detection is widely used in face recognition,traffic flow detection and so on.However,the wood counting mainly uses manual counting methods in China.The staff need to count the wood on the spot,the efficiency of manual wood counting is low and Manpower-consuming.Recognizing the number of square wood through the target detection algorithm model,and manually modifying the false detection can undoubtedly save manpower and reduce costs.In this thesis,we obtained the square wood data set by shooting on the spot and enhanced the data to solve the problem of uneven brightness in the data set.This thesis design experiments to select the best data enhancement method.Afterwards,This thesis uses the dilation convolution to optimize the YOLOv4 network structure and proposes three network structures.comparative study of loss function and structures were introduced to obtain the DYOLOD model and the kmeans++ algorithm was used to cluster the anchor frames.The experimental results prove that the DYOLOD model is better than Faster R-CNN and other models in the intelligent recognition of square wood number.This thesis designs and implements an intelligent identification system for wood counting,including square wood counting system and system management system,at the same time apply the DYOLOD model to square wood number recognition。Using Spring framework,Vue framework and B/S architecture and other technologies to implement the intelligent identification system for wood number on the Android side and the system administrator system on the Web side.It has implemented the intelligent identification function of the square wood number,the module selection function,the report generation function,the personal information management function,and the model management function.Finally,the function and performance of the system are tested.Through the test,it is proved that the function and performance of the system meet the basic requirements of the initial operation,and can provide services for the square wood counting user and improve the efficiency of wood counting. |