Wood density is a pivotal indicator for evaluating wood quality. Therefore, through the accurate prediction of Pinus sylvestris wood density, the utilization efficiency of wood can be effectively enhanced, bearing significant implications for the rational utilization of wood resources and the cultivation of forest trees.

This study focuses on Pinus sylvestris wood and utilizes NIR non-destructive testing technology. It employs feature extraction algorithms including uninformative variable elimination (UVE), competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), as well as grid search algorithm-support vector machine regression (Grid-SVM), genetic algorithm-support vector machine regression (GA-SVM), and particle swarm optimization-support vector machine regression (PSO-SVM). These algorithms are sequentially applied for feature band extraction and SVM modeling. Consequently, an ideal near-infrared prediction model for Pinus sylvestris wood density is developed.

Ninety-seven samples of Pinus sylvestris wood were divided into training and testing sets in a 1∶3 ratio. Grid-SVM, GA-SVM, and PSO-SVM models were established using the full spectrum of near-infrared data. To simplify the near-infrared models, three feature extraction algorithms-uninformative variable elimination (UVE), competitive adaptive reweighted sampling (CARS), and successive projections algorithm (SPA) were sequentially applied to select spectral bands. Subsequently, Grid-SVM, GA-SVM, and PSO-SVM algorithms were employed to model the selected wavelengths. Ultimately, it was found that the GA-SVM model, utilizing feature bands extracted by UVE, exhibited the highest accuracy and best performance for predicting Pinus sylvestris wood density, with an R² value of 0.910 8 and RMSEP of 0.005 9 for the prediction set.

Optimized the SVM model for Pinus sylvestris wood density using near-infrared spectroscopy feature extraction algorithms, thereby achieving rapid and precise prediction of wood density. This approach not only minimizes wastage of wood resources but also maximizes their utility and value. Moreover, it provides a theoretical foundation and technical guidance for enhancing and optimizing the cultivation of superior tree species and forest management practices. These advancements propel the wood industry towards greater efficiency and sustainability.

The spectra of larch wood with different surface roughness were analyzed, and the NIR models suitable for different surface roughness were established, which provided a theoretical basis for improving the accuracy and universality of NIR models in predicting wood density.

Taking larch wood from Xinghuo Forest Farm in Heilongjiang Province as the research object, the near-infrared spectra of unpolished (M0), 150-mesh (M1) and 320-mesh (M2) samples were analyzed and studied. 11-point moving average smoothing, baseline correction and SG smoothing were used for spectral preprocessing to remove redundant spectral information. Manual selection, backward interval partial least squares (BiPLS) and synergy interval partial least squares (SiPLS) were used to complete band optimization. A single prediction model for different surface roughness and a mixed near-infrared model with three surface roughness samples were constructed.

The M0 sample contained more spectral information than the other two samples. Among the three pretreatment methods, the comprehensive evaluation of the modeling effect of SG smoothing pretreatment showed the best. The basic density prediction models of M0, M1 and M2 were established based on the three band optimization methods, and the band selection method of SiPLS had the best effect. For the three surface roughness samples of M0, M1 and M2, the validation set correlation coefficients R and RMSEP were 0.865 9 and 0.022 7, 0.766 0 and 0.021 4, 0.725 6 and 0.027 4, respectively. The prediction ability of the SIPLS mixed prediction model based on the mixture of three different roughness samples was better than that of the single model based on each roughness sample. For the three surface roughness samples of M0, M1 and M2, the RMSEP of the model decreased by 11%, 25% and 5%, respectively.

The NIR models based on the three kinds of surface roughness samples can achieve effective prediction of wood density, and the prediction accuracy of the model is M0 > M1 > M2. The SiPLS band selection method can be used to optimize the influence of surface roughness on the prediction model, and the mixed model established on this basis makes the NIR prediction model more universal. It provides a theoretical basis and technical support for the classification, optimization and fine utilization of wood.

Based on near infrared spectroscopy (NIRS), the moisture prediction models of single poplar and pine sawdust were constructed, and the spectral data sets of the two sawdust samples were mixed to establish the moisture prediction model, so as to predict the moisture of multi-tree sawdust simultaneously.

In this study, partial least squares regression (PLSR) and support vector regression (SVR) were used to establish a pulpwood sawdust moisture prediction model based on NIRS. Particle swarm optimization (PSO) and grey wolf optimizer (GWO) were used to optimize the hyperparameters of the model. Using the sawdust of pulping material as the object of study, the near-infrared spectra of 120 mixed sawdust samples, mainly poplar and pine, were collected. The original spectral data were screened out and preprocessed using the high leverage studentized residual (HLSR) and standard normal variate (SNV) methods, respectively. The uninformative variables elimination (UVE) was used to extract the informative bands. The performance of GWO-SVR, PSO-SVR and PLSR models were compared.

For establishing the NIR prediction model of poplar sawdust moisture, SNV+Auto+SGS, CARS and PSO-SVR were the best optimal preprocessing method, selecting characteristic wavelengths method and modeling methods, respectively ( $R_P^2$=0.916 4, RMSEP=0.114 8%). For establishing the NIR prediction model of pine sawdust moisture, SNV+Auto+SGS combined with UVE, and PSO-SVR were the best pre-processing and modeling methods, respectively ( $R_P^2$=0.934 3, RMSEP=0.063 7%). For establishing the NIR prediction model of mixed sample sets of poplar and pine sawdust moisture, MSC+Auto+SGS combined with UVE, and PSO-SVR were the best pre-processing and modeling methods, respectively ( $R_P^2$=0.922 1, RMSEP=0.111 1%).

NIRS can be used to predict the sawdust moisture content of one single tree species, and it is feasible to construct a prediction model for the sawdust moisture content of multi-tree species in the meanwhile. Model optimization by comparing and screening combinations of different preprocessing and optimization algorithms can significantly improve the accuracy of sawdust moisture NIR estimation models. It provides a theoretical basis and technical support for detecting the moisture of sawdust in real time.

Wood density is not only related to various wood properties, but also an important indicator to evaluate the quality and value of wood. The NIRS analysis technique can predict wood density quickly and efficiently, avoiding the tedious detection steps in conventional experiments. As a variant of recurrent neural network (RNN), long short-term memory network (LSTM) can not only learn the high-order characteristic information between sequential data, but also handle the problems of long-distance dependencies, gradient explosion and gradient extinction in RNN. By combining LSTM and NIRS, a non-destructive detecting technique that can accurately predict the air-dry density of Pinus sylvestris wood was proposed to provide a theoretical basis for improving the accuracy of the NIRS model in predicting the air-dry density of wood.

In this study, the spectral data of 106 samples of P. sylvestris were obtained by NIRS spectrometer, and the air-dry density of these samples was obtained with NIRS spectrometer under constant temperature(20±2 ℃) and relative humidity(65%±3%) environment. By comparing multiple groups of pretreatment methods and feature selection methods, Savitzky-Golay smoothing (SGS) and other methods were used for pretreatment, competitive adaptive weighting algorithm (CARS) was used for band selection, eliminating high-frequency noise and redundant information in NIRS data, and improving spectral quality, modeling speed and accuracy. In order to verify the predictive ability of LSTM model, it was compared and analysed with the modelling algorithms such as partial least square regression (PSLR), convolutional neural network (CNN). Each of these above-mentioned three modelling methods were applied to establish a near-infrared prediction model for the air-dry density of P. sylvestris wood.

The NIRS models established based on the above-mentioned modelling methods achieved effective prediction of the air-dry density of P. sylvestris wood. And the prediction accuracy and regression fit of LSTM model were better than PLSR and CNN model. The LSTM model treated by SGS+CARS had the highest prediction accuracy, the strongest generalization performance and the best fitting effect (R²=0.959, RMSEP=0.005, RPD=5.033).

Through the collection of spectral data and air-dry density of pine sylvestris wood, a novel method for detecting air-dry density of wood based on NIRS and LSTM was established. Comparing to the traditional regression model, LSTM model has higher prediction accuracy, better regression effect and stronger robustness. The detection method can not only ensure the integrity of the wood, but also improve the prediction accuracy of the air-dry density, which achieves the rapid non-destructive detection of the air-dry density of the Pinus sylvestris wood, and provides a referable model and theoretical basis for the NIRS analysis of wood.

As is widely known, an increasing number of forest areas were managed to preserve and enhance the health of forest ecosystems. However, previous research on forest management has often overlooked the importance of structure-based.

Our objectives were to define the direction of structure-based forest management. Subsequently, we investigated the relationships between forest structure and the regeneration, growth, and mortality of trees under different thinning treatments. Ultimately, the drivers of forest structural change were explored.

On the basis of 92 sites selected from northeastern China, with different recovery time (from 1 to 15 years) and different thinning intensities (0–59.9%) since the last thinning. Principal component analysis (PCA) identified relationships among factors determining forest spatial structure. The structural equation model (SEM) was used to analyze the driving factors behind the changes in forest spatial structure after thinning.

Light thinning (0–20% trees removed) promoted forest regeneration, and heavy thinning (over 35% of trees removed) facilitated forest growth. However, only moderate thinning (20%–35% trees removed) created a reasonable spatial structure. While dead trees were clustered, and they were hardly affected by thinning intensity. Additionally, thinning intensity, recovery time, and altitude indirectly improve the spatial structure of the forest by influencing diameter at breast height (DBH) and canopy area.

Creating larger DBH and canopy area through thinning will promote the formation of complex forest structures, which cultivates healthy and stable forests.