The real-time detection of pine cones in Korean pine forests is not only the data basis for the mechanized picking of pine cones, but also one of the important methods for evaluating the yield of Korean pine forests. In recent years, there has been a certain number of detection accuracy for image processing of fruits in trees using deep-learning methods, but the overall performance of these methods has not been satisfactory, and they have never been used in the detection of pine cones. In this paper, a pine cone detection method based on Boundary Equilibrium Generative Adversarial Networks (BEGAN) and You Only Look Once (YOLO) v3 mode is proposed to solve the problems of insufficient data set, inaccurate detection result and slow detection speed. First, we use traditional image augmentation technology and generative adversarial network BEGAN to implement data augmentation. Second, we introduced a densely connected network (DenseNet) structure in the backbone network of YOLOv3. Third, we expanded the detection scale of YOLOv3, and optimized the loss function of YOLOv3 using the Distance-IoU (DIoU) algorithm. Finally, we conducted a comparative experiment. The experimental results show that the performance of the model can be effectively improved by using BEGAN for data augmentation. Under same conditions, the improved YOLOv3 model is better than the Single Shot MultiBox Detector (SSD), the faster-regions with convolutional neural network (Faster R-CNN) and the original YOLOv3 model. The detection accuracy reaches 95.3%, and the detection efficiency is 37.8% higher than that of the original YOLOv3.

This study used near-infrared (NIR) spectroscopy to predict mechanical properties of wood. NIR spectra were collected in wavelengths 900–1700 nm, and spectra averaged by radial and tangential surface spectra were used to establish a partial least square (PLS) model based on correlation local embedding (CLE). Mongolian oak (Quercus mongolica Fisch. ex Ledeb.) was used to test the effectiveness of the model. The cross-validation method was used to verify the robustness of the CLE–PLS model. Ninety samples were tested as the calibration set and forty-five as the validation set. The results show that the prediction coefficient of determination ( $$R_{p}^{2}$$Rp2 ) is 0.80 for MOR, and 0.78 for MOE. The ratio of performance to deviation is 2.23 for MOR and 2.15 for MOE.

The development of new approaches to tree-level parameter extraction for forest resource inventory and management is an important area of ongoing research, which puts forward high requirements for the capabilities of single-tree segmentation and detection methods. Conventional methods implement segmenting routine with same resolution threshold for overstory and understory, ignoring that their lidar point densities are different, which leads to over-segmentation of the understory trees. To improve the segmentation accuracy of understory trees, this paper presents a multi-threshold segmentation approach for tree-level parameter extraction using small-footprint airborne LiDAR (Light Detection And Ranging) data. First, the point clouds are pre-processed and encoded to canopy layers according to the lidar return number, and multi-threshold segmentation using DSM-based (Digital Surface Model) method is implemented for each layer; tree segments are then combined across layers by merging criteria. Finally, individual trees are delineated, and tree parameters are extracted. The novelty of this method lies in its application of multi-resolution threshold segmentation strategy according to the variation of LiDAR point density in different canopy layers. We applied this approach to 271 permanent sample plots of the University of Kentucky’s Robinson Forest, a deciduous canopy-closed forest with complex terrain and vegetation conditions. Experimental results show that a combination of multi-resolution threshold segmentation based on stratification and cross-layer tree segments merging method can provide a significant performance improvement in individual tree-level forest measurement. Compared with DSM-based method, the proposed multi-threshold segmentation approach strongly improved the average detection rate (from 52.3% to 73.4%) and average overall accuracy (from 65.2% to 76.9%) for understory trees. The overall accuracy increased from 75.1% to 82.6% for all trees, with an increase of the coefficient of determination R² by 20 percentage points. The improvement of tree detection method brings the estimation of structural parameters for single trees up to an accuracy level: For tree height, R² increased by 5.0 percentage points from 90% to 95%; and for tree location, the mean difference decreased by 23 cm from 105 cm to 82 cm.