BACNN: Multi-scale feature fusion-based bilinear attention convolutional neural network for wood NIR classification

Effective development and utilization of wood resources is critical.Wood modification research has become an integral dimension of wood science research,however,the similarities between modified wood and original wood render it challenging for accurate identification and classification using conventional image classification techniques.So,the development of efficient and accurate wood classification techniques is inevitable.This paper presents a one-dimensional,convolutional neural network(i.e.,BACNN)that combines near-infrared spectroscopy and deep learning techniques to classify poplar,tung,and balsa woods,and PVA,nano-silica-sol and PVA-nano silica sol modified woods of poplar.The results show that BACNN achieves an accuracy of 99.3%on the test set,higher than the 52.9%of the BP neural network and 98.7%of Support Vector Machine compared with traditional machine learning methods and deep learning based methods;it is also higher than the 97.6%of LeNet,98.7%of AlexNet and 99.1%of VGGNet-11.Therefore,the classification method proposed offers potential applications in wood classification,especially with homogeneous modified wood,and it also provides a basis for subsequent wood properties studies.

Wood defect detection method with PCA feature fusion and compressed sensing

We used principa/component analysis （PCA） and compressed sensing to detect wood defects from wood plate images. PCA makes it possible to reduce data redundancy and feature dimensions and compressed sensing, used as a elas- sifter, improves identification accuracy. We extracted 25 features, including geometry and regional features, gray-scale texture features, and invariant moment features, from wood board images and then integrated them using PCA, and se- lected eight principal components to express defects. After the fusion process, we used the features to construct a data dic- tionary, and realized the classification of defects by computing the optimal solution of the data dictionary in l1 norm using the least square method. We tested 50 Xylosma samples of live knots, dead knots, and cracks. The average detection time with PCA feature fusion and without were 0.2015 and 0.7125 ms, respectively. The original detection accuracy by SOM neural network was 87 %, but after compressed sensing, it was 92 %.

A rapid, automated flaw segmentation method using morphological reconstruction to grade wood flooring

Region-Growing Algorithms （RGAs） are used to grade the quality of manufactured wood flooring. Traditional RGAs are hampered by prob- lems of long segmentation time and low inspection accuracy caused by neighborhood search. We used morphological reconstruction with the R com- ponent to construct a novel flaw segmentation method. We initially designed two template images for low and high thresholds, and these were used for seed optimization and inflation growth, respectively. Then the extraction of the flaw skeleton from the low threshold image was realized by applying the erosion termination rules. The seeds in the flaw skeleton were optimized by the pruning method. The geodesic inflection was applied by the high threshold template to realize rapid growth of the flaw area in the floor plate, and region filling and pruning operations were applied for margin optimization. Experi- ments were conducted on 512×512, 256×256 and 128×128 pixel sizes, re- spectively. The 256×256 pixel size proved superior in time-consumption at 0.06 s with accuracy of 100%. But with the region-growing method the same process took 0.22 s with accuracy of 70%. Compared with RGA, our pro- posed method can realize more accurate segmentation, and the speed and accuracy of segmentation can satisfy the requirements for on-line grading of wood flooring.