Prediction And Analysis Of Conifer Tracheid Performance Based On Atlas Fusion

Coniferous wood is the main structural wood species.Coniferous wood in the Greater Khingan Mountains forest area is selected as the research object to design a mechanical property analysis method of coniferous wood based on microscopic image and near infrared spectrum fusion,which includes image processing,spectral analysis,feature fusion and deep learning algorithms.Through the collection of microscopic images and surface near-infrared spectral data of coniferous wood,combined with mechanical property analysis of wood as the model characterization,a multi-channel macroscopic mechanical analysis model of wood at the microscopic tracheid level was constructed,which realized the numerical extraction and analysis of tracheid characteristics at the microscopic level,and provided a new method for the performance analysis and practical application of coniferous wood.Understanding the macroscopic mechanical behavior of wood at the microscopic scale is of great significance for the design of cell wall-like composites and pulp paper making.Using the numerical method to extract tracheid traits from larch,and the data set is established using the load intensity data,the gray-level co-occurrence matrix,cell segmentation,spectral dimension reduction,and feature selection.A feature data prepossessing algorithm based on random forest was established to optimize the eigenvalues.The results showed that s,PC3,ASM,PC2,AREA,CON,PERIMETER and CORR were the eight features that had a greater impact on the mechanical properties of wood,and used as the feature input of the network.Based on the principle of finite element and numerical simulation,the five-level classification standard of new larch structural timber was established.In order to predict the mechanical properties of tracheid and analyze their relationship with tracheid features,multi-source data sets of features,microscopic images,and near-infrared spectroscopy were integrated.The existing data features were improved over the baseline fuzzy neural network,the full convolutional neural network,and the adversarial generative neural networks.Further,the three-level feature fusion mechanism and sparse attention mechanism were combined to achieve multi-source feature fusion,and Multi-FF-Net model was designed.Lastly,the training set accuracy of the Multi-FF-Net model achieved 88.37%.Where morphological,texture,and spectral characteristics under the microscope are inputs respectively,the training model has an accuracy of 77.32 %,75.53 %,and 82.08 %,and the spectral characteristics have a more significant influence on the mechanical properties up to a point.Comparisons between experiments provide evidence that the model can improve prediction accuracy.