Research On Intelligent Detection Method And Application Of Defects For Solid Wood Panels Optimal Processing

At present,many wood processing enterprises mainly use rough processing,which causes serious wood waste and low utilization rate.To reverse this situation,this subject takes artificial intelligence as the starting point to carry out research on wood optimal processing technology,and then,While improving the grade and yield of solid wood panels,optimize the processing mode of enterprises and enhance the intelligence level of wood processing industry in China,this has important theoretical significance and high practical value for the transformation and upgrading of enterprises to intelligent development.To solve the problems of low defect detection accuracy,long detection time,large amount of calculation and easy local convergence of intelligent layout in the optimal processing of solid wood plate processing enterprises.In this paper,the intelligent defect detection method and application research for solid wood plate were carried out.The main research contents and innovations were as follows:1.According to the needs of the cooperative enterprises,this subject mainly used red pine and camphor pine as test materials,used CCD camera to collect data,used digital image processing technology,combined with data augmentation method,for the wood defects such as dead knot,live knot,blue stain,brown stain,pitch streak,decay and crack,a classification data set and a segmentation data set were established.The classification data set contains 117,091 defect images,and the segmentation data set contains 8,887 pairs of image data,which prepared relevant data for subsequent defect detection research work.2.Aiming at the problem that it is difficult to reconcile the accuracy and speed of wood surface defect identification,this paper innovatively proposed an AMR-CNN(Multiple kernels residual module convolutional neural network based on ACB optimized)network model.The model was based on the classic classification network VGGNet(Visual Geometry Group Net)by constructing multi-convolution kernel residual modules,reducing network parameter redundancy,enhancing network convolution kernel skeleton,so that it could quickly and accurately extract the features of wood surface defects,reduce half of the inference time,effectively improve the resolution of defects with high similarity,thereby improve the overall recognition accuracy.3.For the problem of the poor positioning accuracy of the solid wood surface defects,this paper innovatively proposed an AMR-FCN-CRF(Multiple kernels residual module fully convolutional network based on ACB and CRF-RNN optimized)network model.This model was based on AMR-CNN model,by constructing deconvolution and feature map fusion module decoder structure,integrating spatial information,while ensured that the classification accuracy met the requirements,the positioning accuracy of the model was greatly improved.Compared this model with classic network models such as FCN-8s(Fully convolutional network-8s),Seg Net(Segmentation network),and U-Net(U network),the former improved the recognition accuracy by more than 1.66%,and the PA and the m Io U by more than 2.57% and 8.93%respectively in less than 0.5s.4.According to the actual needs of the enterprise,in the solid wood plate optimization processing system,it integrated operation functions such as defect detection,intelligent sorting,and optimized layout processing,which included two types of processing modes: grading optimization and optimal layout processing.In addition,for the problems of large amount of calculation and easy local convergence of intelligent layout,this paper innovatively proposed a dendrogram genetic algorithm layout model.