Vibration harvesting is an effective approach for the mechanized harvesting of Camellia oleifera fruit. Accurate prediction of fruit inertial force is crucial for enhancing harvesting efficiency, yet it remains challenging due to the complex interactions between excitation parameters and tree structural characteristics. In this study, a pendulum dynamics model of the fruit-branch system was established, and a multidimensional dataset was collected through field tests. Six machine learning models were subsequently trained and evaluated using this dataset. Results showed that the Random Forest (RF) model provided the best predictive performance, with mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2) values of 1.05, 1.73, and 0.79, respectively. Subsequently, the SHAP method was employed to interpret the RF model and quantify feature importance. The analysis revealed that excitation parameters were the primary determinants of fruit inertial force, contributing 73.6%, compared with 26.4% from tree structural parameters. Among all features, excitation frequency exhibited the greatest influence, contributing 42.06%. Feature selection further improved the predictive performance of the RF model. On the test set, the MAE, RMSE, and R2 values were 1.01, 1.64, and 0.82, respectively. Field validation tests confirmed a low error rate of 6.22% between the model predictions and the measured inertial forces. These findings provide a theoretical basis for precision vibration harvesting of Camellia oleifera and offer a useful reference for the harvesting of other forest fruits.

Electromagnetic wave attenuation in forests is fundamentally governed by the dielectric properties of the propagation medium. In practice, however, the spatial structure between two forest nodes is extremely difficult to quantify, and forest environments exhibit strong heterogeneity. Forest propagation is therefore often treated as a “black box”, where attenuation is measured empirically between transmitting and receiving nodes and fitted with site-specific models that typically show weak generalization ability. Because microclimate strongly regulates the dielectric constant of air, vegetation, and soil, it provides a physically grounded pathway to explain signal variability beyond purely empirical descriptions. To address the question “Is signal attenuation in forests related to the tree-proximal microclimate?”, we deployed a multi-node LoRa monitoring network in a natural mixed forest and collected more than 52,000 synchronous records of RSSI and tree-, soil-, and air-related microclimate variables across four observation campaigns. An interpretable KNN–SHAP framework was applied to quantify nonlinear effects and threshold behaviors. The results show that RSSI increases with temperature and decreases with humidity and soil moisture, forming a clear “heat-enhanced, moisture-suppressed” propagation pattern. A stable microclimatic window with minimal attenuation emerges under moderate dryness (VPD 1.5–1.9 kPa) and gentle wind (1–4 m s−1), whereas signal degradation intensifies near VPD ≈ 1.0–1.3 kPa and at soil moisture levels of 20–35

Walnut trees vary greatly in shape, which results in large differences in the optimal excitation parameters between trees. By clarifying the correlation between the tree shape characteristic parameters and the optimal excitation parameters, the excitation parameters can be matched according to the tree shape characteristics. The purpose of this is to improve picking efficiency and simultaneously reduce damage to the tree. In this study, a field experiment was conducted using a low-frequency shaker (Trunk shakers) and the vibration response under different excitation parameters was analysed using statistical methods. Finally, the range of excitation parameters based on the quantitative matching of tree feature parameters was obtained through linear regression and a comparison of feature importance. It was found that the tree height and crown projection area were the two most important factors affecting the acceleration of the tree vibration response, and they were significantly negatively correlated with each other. The optimal intervals of the excitation frequency and amplitude were successfully predicted from the tree height and canopy projection area as (omega 0 eta min, omega 0 eta max) and (A0 eta min/beta(omega min), Amax eta max/ beta(omega max)), respectively. Validated tests showed a prediction accuracy of 93 % for this range. This study provides a reliable range of excitation parameters for future walnut harvesting operations, which can guide the development of low-frequency intelligent shakers, in addition to field operations.

Current research on walnuts (Juglans regia L.) harvesting mainly focuses on improving harvesting efficiency and reducing tree damage, while the impact of the sensor's own data collection accuracy has been less considered. Therefore, this paper proposes a modular, separable, and implantable fruit acceleration measurement method based on Yunnan walnuts. A sensor system is designed by separating the data processing module from the acceleration acquisition module, and the key factor affecting the accuracy of fruit data acquisition-the sensor's self-mass-is analysed. Meanwhile, the optimal sampling frequency is calculated. Based on the modular approach, this study further proposes an implantable fruit acceleration sensor measurement method, which reduces the effect of the sensor's mass on measurement results by replacing part of the fruit's mass with the measurement module. The results show that (1) compared with the modular measurement method, the implanted measurement method under standard vibration mode (walnut mass 65 g, vibration frequency 5 Hz) has an acceleration error of 0.1 gacc, with the error reduced by 79.2 %; (2) The optimal sampling frequency is 6-8 times the Nyquist frequency. At this frequency, the cosine similarity compared to the traditional method is 0.84, significantly reducing data transmission load and computational complexity without losing data accuracy. This method can provide potential technical support for more accurate measurements of walnuts and other fruits in the future.

Mulberry leaves and sea buckthorn are common forest fruits and vegetables, which are medicinal and edible plants. After harvesting, farmers mainly process them by drying. Consumers tend to purchase products with better color after drying. The cover of the direct solar dryer affects the color of the dried product. Although UV blocking coatings have been developed, little is known about the effects of other solar spectral components besides UV light on the color and main pigment contents of dried agricultural products. Therefore, the aim of this study was to investigate the effects of different colored films and their transmission spectra on the drying characteristics and color of dried product. Six different colored films were used for experiment and L*, a*, b*, total color change (Delta E), browning index (BI) and pigment content of the samples were measured. Results showed that the dried mulberry leaves covered by red film have the lowest Delta E value (6.97), mainly due to the low transmittance of the red film (52.31%) and its ability to block UV rays. In addition, red light can reduce the loss of leaf pigment during the drying process. For sea buckthorn, covered by dark green film was closest to its natural color, with a Delta E value of 21.79. During the drying process, incident some blue light could increase the carotenoid content of dried sea buckthorn. This study provides a theoretical basis for development of light quality regulation and drying technology in the drying process of agricultural products.

Ball milling (BM) was a commonly used pretreatment method for pyrolysis. However, the lack of standardized BM intensity parameters made quantitative comparisons between BM pretreatment regulated pyrolysis studies difficult, which limited the practical application of BM pretreatment research findings in industrial production. In this study, the intensity of BM pretreatment was quantitatively evaluated by combining high-speed imaging measurements with discrete element method (DEM) analysis, where the kinetic energy dose was introduced as a measurement parameter. The pyrolysis reaction process and product distribution of Pennisetum giganteum (PG) samples were analyzed in combination with the targeted action sites of mechanical energy during the pretreatment process, and the regulatory mechanism of mechanical energy on the pyrolysis process and outcomes of lignocellulose was investigated. The kinetic energy dose-benefit calculation formula was utilized to analyze the benefits of consuming a unit amount of mechanical energy on the pyrolysis at different kinetic energy dose levels. The results showed that during the BM process, changes in certain physicochemical properties (Crystallinity, C-O and C-C bond content, and aromatic groups) occurred only after the kinetic energy dose reached a threshold. For most properties, unit kinetic energy dose benefits decreased with increasing kinetic energy dose. Under the pretreatment conditions specified in this study, a kinetic energy dose (D) of 200-400 kJ/g maximizes the benefit for pyrolysis oil production from PG, thereby providing a quantitative framework for optimizing BM pretreatment.

Clarifying the impact of excitation and tree structural parameters on the vibration transmission characteristics of trees is crucial for enhancing the efficiency of vibration harvesting. In this study, the mass distributions of the fruits, buds, and leaves on the canopy branches were used to construct a vibration model, and the low-order natural frequency range of the branches was determined to be 1.39 similar to 11.58 Hz. Vibration transmission tests revealed that frequency, amplitude, excitation distance, and crown diameter positively influence crown acceleration, while tree height and trunk diameter have a negative influence. Vibration frequency, tree height, crown diameter, and trunk diameter positively influence the energy decay rate, whereas vibration amplitude and excitation distance have a negative influence. Canopy acceleration and energy decay rate are most strongly affected by frequency, amplitude and excitation distance. With a vibration frequency of 8.96 Hz, an amplitude of 50 mm, and an excitation distance of 60 cm, the canopy acceleration was maximized at 67.12 m/s(2), while minimum energy decay rate was 68.43%. The optimum operating parameters of the harvesting device were reached at a frequency of 8.7 similar to 9.6 Hz, an amplitude of 47 similar to 50 mm, and an excitation distance of 53 similar to 60 cm. These research findings provide an important reference for the development and field application of Camellia oleifera vibration harvesting devices.

Tree-climbing robots are primarily utilized for pruning and harvesting in tall trees; however, limited structural degrees of freedom (DoFs) reduce their flexibility in complex environments. To improve the flexibility and environmental adaptability of the robots, this study proposes a novel three-armed claw-type tree-climbing robot inspired by gibbons. A 14 DoFs prototype with a total mass of approximately 2.52 kg was developed, comprising three manipulator arms and independently actuated claws. Kinematic models were separately established for the series-connected arms and the parallel-connected moving platform, with accuracy verified through numerical simulations. Based on these models, a control system was implemented, and a physical prototype was tested in field climbing experiments. Grasping tests on surfaces of varying roughness, including moist tree trunks, artificial wood, and smooth steel plates, demonstrated the adaptability of the claw to diverse materials. The robot successfully climbed trunks inclined at 52-90°, supporting a maximum payload of 1.81 kg; each full gait cycle averaged approximately 4 min. These results indicate that the robot can successfully imitate the movements of gibbons during climbing, thereby verifying the feasibility and practical application value of this bionic design in real-world forestry environments.

Direct solar drying has been widely used as the simple structure,easy manufacturing,and portability.The surface of the material is prone to overheating due to direct exposure to strong light radiation.Mechanical ventilation has been used to improve the drying quality.Shade mesh can be installed to reduce air temperature and excessive light on materials.However,the available solar energy can be reduced at the same time.In addition,the ventilation equipment is typically powered by photovoltaic cells.Most solar radiation energy is converted into heat energy and directly dissipated due to the limitation of the bandgap width of semiconductor materials.Solar energy cannot be fully utilized for drying.Overall,there is a low value in the comprehensive utilization rate of solar energy in the drying systems.It is often required to improve the utilization rate of solar energy for high-quality drying materials.In this study,spectral splitting technology was applied for direct solar drying.The solar spectrum was split into two components at 640 nm.Specifically,the 280-640nm band was reflected into the photovoltaic cell for the direct power generation,while the 640-2 500 nm band was transmitted into the interior of the drying chamber to raise the temperature of the air.A systematic investigation was implemented to explore the effects of direct solar drying,solar shade drying,and spectral splitting solar drying on the drying time,drying efficiency,color difference,nutritional composition,and microstructure of mulberry leaves.Additionally,the electrical performance of the photovoltaic module was also evaluated to determine the comprehensive utilization rate of solar energy.The experimental results show that the drying efficiency of direct solar drying(7.93％)was close to that of spectral splitting solar drying(7.84％),both of which were higher than that of solar shade drying(3.79％);In view of the exposure to intense light and high temperatures,the worst quality of mulberry leaves was found after direct solar drying,with a color difference of 24.39,a large loss of nutrients,and serious shrinkage and deformation of epidermal cells;The color differences of mulberry leaves were 12.01 and 11.33 after solar shade drying and spectral splitting solar drying,respectively.There were high retention rates of nutrients and bioactive compounds,along with minimal shrinkage in microstructure.Therefore,both solar shade drying and spectral splitting solar drying enhanced the mulberry leaf quality.The biological effects of the red light on the mulberry leaves after spectral splitting solar drying also outperformed those after solar shade drying in the quality metrics,such as the color difference and chlorophyll contents;The spectral splitter photovoltaics received less solar energy,with the 11.47％increase of photoelectric conversion efficiency.While the photoelectric conversion efficiency of conventional photovoltaics was only 5.67％,due to the high temperature and light-induced degradation.There was the best quality of mulberry leaves after spectral splitting solar drying,with a solar energy utilization rate of 10.61％.This finding can provide the technical support to improve the quality of solar drying products and the utilization rate of solar energy.

Effective monitoring of forest microclimates-including tree, soil, and atmospheric conditions-is essential for understanding ecosystem dynamics and advancing forest management. Yet, in dense or primary forests, traditional wireless sensor networks (WSNs) and existing IoT systems are limited by signal attenuation, restricted range, energy demands, and insufficient sensing depth. Integrated systems capable of stable wireless communication and environmental monitoring, together with evaluation of RF signals, remain lacking. Potential coupling between signal behavior and microclimate is also underexplored. This study proposes and evaluates an AI-IoT framework tailored to a closed-canopy forest stand, composed of: (1) multi-source nodes that monitor trunk, soil, and air microclimate variables while recording RSSI for link-quality tracking; (2) a UAV-relayed LoRaWAN system in which a UAV-mounted gateway ascends from the forest floor to above the canopy, enabling height-resolved propagation analysis and modeling of RSSI as a function of distance and elevation; and (3) a cloud-based platform that receives data streams and supports environmental monitoring and short-term forecasting via LSTM models. Field experiments at the study stand showed that elevating the UAV gateway to 50 m improved RSSI by >30 dB and reduced packet loss from over 60 % to below 5 % over similar to 1 km. The LSTM model achieved high predictive fidelity for temperature- and humidity-related variables, with mean absolute percentage errors typically below 5 %, while soil and intermittent meteorological variables exhibited moderate to lower accuracy. By jointly analyzing RSSI and co-located microclimate observations within the above-canopy clearance zone characteristic of the Baicaowa stand, the framework provides preliminary, site-specific evidence of short-term coupling between signal strength and the thermo-hydric state of trees and soils. These relationships remain correlative and specific to the monitored stand and period, and their generality and causal mechanisms will require cross-site, multi-season, and experimental validation.

The global low-carbon transition is driving the use of renewable energy for ecological monitoring. Traditional power supply for forest monitoring sensor equipment is constrained by high wired costs, frequent battery replacement, and the limitations of low light levels and special spectra under forest canopies on photovoltaic (PV) compatibility. Existing research lacks exploration of the correlation between under-forest spectra and PV performance. This study measured the summer understory light spectra of five tree species in Beijing, evaluated the performance of three types of PV cells—monocrystalline silicon, polycrystalline silicon, and amorphous silicon—and designed a low-light energy harvesting circuit. Results indicate that spectral differences under tree canopies are concentrated from 380–680 nm, exhibiting a distinctive forest-specific spectral feature of “high-band enrichment” above 680 nm. Under low-light conditions, polycrystalline silicon photovoltaics demonstrates optimal performance when adapted to this high-band spectrum. The designed circuit can activate at 5 W/m2 irradiance and stably output 4.16 V voltage. This study fills a spectral gap in northern summer tree canopies, providing a comprehensive solution of “material adaptation + circuit customization” for the practical deployment of shaded forest PV systems.

Three-dimensional (3D) reconstruction is important for obtaining morphological information and making intelligent management decisions for fruit trees. Thus, a method for the 3D reconstruction and parameters extraction of branches based on Neural Radiation Fields (NeRF) was proposed for walnut (Juglans regia L.) trees. This approach combined Structure from Motion (SfM) with NeRF and used multi-view images to reconstruct branches. First, a dataset of multi-view images of walnut trees was built and camera poses were obtained using SfM. Second, WalnutNeRF was optimized by incorporating hash encoding, piecewise sampler and appearance embedding features to address challenges associated with complex outdoor environments and accurately reconstruct branches. A scale recovery method using calibration objects was employed to extract branch parameters. The effectiveness of WalnutNeRF was evaluated by analyzing rendering performance, reconstruction efficiency, point cloud quality, and the accuracy of extracted branch parameters. WalnutNeRF outperformed existing methods in terms of the quality of rendered images and the accuracy of estimated depth, as determined using PSNR, SSIM, LPIPS, and other metrics. WalnutNeRF resulted in a branch reconstruction accuracy of 90.94 %, with a training time that was 9-time faster than that of SfM-MVS. Compared with SfM-MVS, WalnutNeRF decreased reconstruction errors for the main branches, lateral branches, and watershoots by 72 %, 67 %, and 57 %, respectively, and decreased the errors in length by 7.09 %, 4.33 %, and 65.07 %, respectively. Accordingly, WalnutNeRF decreased the reconstruction time, while increasing accuracy, providing robust support for the development of intelligent management applications (e.g., intelligent pruning) for walnut trees.

Obtaining highly intact walnut kernels after shell-breaking is a key step in the deep processing of walnuts. Juglans sigillata D., characterized by tightly bonded internal septa and irregular shell geometry, poses significant challenges for achieving kernel integrity. In this study, a three-dimensional walnut model was developed using 120 walnut samples. Four types of compression heads-flat, concave groove-shaped, inverted conical, and protruding stud-shaped-were employed to perform static loading tests at a constant rate of 10 mm/min from three directions: the head, belly, and umbilical regions. The shell-breaking characteristics and crack formation patterns were systematically investigated. Results showed that the shell-breaking process involves an initial elastic deformation followed by plastic failure. The initial shell-breaking force ranged from 172.3 N to 739.7 N, significantly influenced by both compression head type and loading direction. The highest probability of obtaining intact kernels (over 70%) was achieved by applying a belly-directional load using the concave groove-shaped compression head. Initial crack formation patterns were categorized into six types across three main models, with longitudinal cracks perpendicular to the suture being most favorable for intact kernel acquisition. A finite element-based predictive model for crack initiation demonstrated an accuracy of 85.3%, showing high consistency with observed stress concentration zones and actual crack propagation paths. This study elucidates the mechanical behavior and crack formation patterns during walnut shell-breaking and highlights the relationship between crack formation and kernel integrity, offering valuable insights for the development of efficient walnut shelling equipment.

The finite element method can effectively simulate the pruning process to investigate the pruning mechanism. To ensure the reliability of the simulation results, it is essential to measure and calibrate the model parameters. In this study, a finite element model was established to simulate the pruning process executed by pruning robots. The biomechanical properties of Populus tomentosa branches were determined using mechanical tests. The finite element model parameters were calibrated using the Plackett-Burman and Box-Behnken methods, and the reliability of these calibrated parameters was subsequently validated through field testing in a forest environment. Finally, the calibrated finite element model was used to investigate the impact-cutting pruning mechanism of Populus tomentosa branches to further solve the problems of poor pruning quality and serious blade wear caused by unreasonable working parameters of the blade in the pruning process. The results indicate that the calibrated finite element model of the biomechanical properties of Populus tomentosa branches accurately simulates the pruning process. Moreover, the simulation results show that reducing the cutting speed and increasing the blade wedge angle led to a decrease in the peak stress and thus blade wear while increasing the cutting speed and reducing the blade wedge angle led to an improvement in the pruning quality. At cutting speeds of more than 6 m center dot s-1 and with a blade wedge angle of less than 35 degrees, branches with a diameter of 25 mm were cut with a cutting share of nearly 100 %. This study provides a robust model for the simulation and optimization of impact-cutting pruning by robots, which is of considerable significance for the development of high-quality products in the enhancement of forestry pruning practices.

Intelligent pruning technology is significant in reducing management costs and improving operational efficiency. In this study, a branch recognition and pruning point localization method was proposed for dormant walnut (Juglans regia L.) trees. First, 3D point clouds of walnut trees were reconstructed from multi-view images using Neural Radiance Fields (NeRFs). Second, Walnut-PointNet was improved to segment the walnut tree into Trunk, Branch, and Calibration categories. Next, individual pruning branches were extracted by cluster analysis and pruning rules were adjusted by classifying branches based on length. Finally, Principal Component Analysis (PCA) was used for length extraction, and pruning points were determined based on pruning rules. Walnut-PointNet achieved an OA of 93.39%, an ACC of 95.29%, and an mIoU of 0.912 on the walnut tree dataset. The mean absolute errors in length extraction for the short-growing branch group and the water sprout were 28.04 mm and 50.11 mm, respectively. The average success rate of pruning point recognition reached 89.33%, and the total time for pruning branch recognition and pruning point localization for the entire tree was approximately 16 s. This study provides support for the development of intelligent pruning for walnut trees.

This study aims to develop a method for predicting walnut (Juglans regia L.) yield based on the walnut orchard point cloud model, addressing issues such as low efficiency, insufficient accuracy, and high costs in traditional methods. The walnut orchard point cloud is reconstructed using unmanned aerial vehicle (UAV) images, and the semantic segmentation technique is applied to extract the individual walnut tree point cloud model. Furthermore, the tree height, canopy projection area, and volume of each walnut tree are calculated. By combining these morphological features with statistical models and machine learning methods, a prediction model between tree morphology and yield is established, achieving prediction accuracy with a mean absolute error (MAE) of 2.04 kg, a mean absolute percentage error (MAPE) of 17.24%, a root mean square error (RMSE) of 2.81 kg, and a coefficient of determination (R2) of 0.83. This method provides an efficient, accurate, and economically feasible solution for walnut yield prediction, overcoming the limitations of existing technologies.

In precision forestry and forestry harvester heads, accurately determining tree diameter is essential for precise volume estimation and effective resource management. Controlling the rotation angle of the feed roller mechanism in forestry harvester heads provides a direct method for obtaining reliable diameter information, allowing for more accurate assessment of timber yield. In addition, by leveraging angle control, it becomes possible to navigate the challenges posed by the nonlinear dynamics of hydraulic systems, which are difficult to model and control using conventional methods. To address these needs, this paper presents an innovative control strategy that combines finite time sliding mode control (FTSM) with active disturbance rejection control (ADRC), termed FTSM-ADRC. This approach optimizes the rotational control of the hydraulic cylinder driving the feed rollers, thereby improving both measurement accuracy and system responsiveness. The FTSM-ADRC algorithm incorporates an extended state observer (ESO) to real-time monitor and compensate for disturbances within the hydraulic system, enhancing robustness against internal and external fluctuations. Moreover, the high-order sliding mode control component allows the system to adapt quickly to environmental variations, ensuring stability within a finite time. Experimental results using a forestry harvester head show that FTSM-ADRC surpasses traditional ADRC, with notable reductions in overshoot, shorter settling times, improved integral of absolute error (IAE), and faster convergence rates. Bench tests further validate the effectiveness of FTSM-ADRC in controlling hydraulic cylinder-driven systems, confirming its potential to enhance measurement precision and operational efficiency in forestry machinery. Overall, FTSM-ADRC provides a robust solution for advancing precision forestry, offering significant improvements in diameter measurement and control accuracy.

The capacity of Lithium-ion batteries degrades over the time, making accurate prediction of their Remaining Useful Life (RUL) crucial for maintenance and product lifespan design. However, diverse aging mechanisms, changing working conditions and cell-to-cell variation lead to the inhomogeneous cell lifespan and complicated life prediction. In this work, a data-driven algorithm based on stacked Long Short Term Memory (LSTM) encoder–decoders is proposed for RUL prediction. The encoder and upstream decoder form an autoencoder framework for feature extraction. The encoder and the downstream decoder form the encoder–decoder framework for RUL prediction. To enhance generalization during training, the Maximum Mean Discrepancy (MMD) loss is included in the autoencoder framework. The similarity of aging patterns is analyzed during splitting source and target datasets through k-means and Density-Based Spatial Clustering of Applications with Noise (DBSCAN). The Euclidean metric with accumulated Equivalent Cycle Number (ECN) sequence during aging shows better performance for similarity-based data splitting than the Dynamic Time Wrapping (DTW) distance metric based on capacity fading trajectory. The experimental results indicate that the proposed algorithm can provide accurate RUL prediction using 5% fading data and shows good generalization with R2 score of 0.98.

This paper proposes NERF-F, a three-dimensional (3D) reconstruction method for orchard scenes using unmanned aerial vehicle (UAV) images and neural radiance fields (NERF) theory, to address the challenges of largescale orchard point cloud reconstruction. Multi-view images of orchards are captured using a DJI Phantom 4 UAV, and the camera motion model is established. A method to recover camera poses from UAV position and orientation system (POS) data is introduced, with the images and pose data fed into a multilayer perceptron (MLP) network for rendering. This process generates a NERF that describes the orchard scene's colour and voxel density. The MLP network weights are optimised via a loss function, iterating until convergence. After training, the orchard NERF is used to generate and export the orchard point cloud model. Experimental results showed that NERF-F accurately reconstructed macro structures such as tree crowns, terrain, tree height, and inter-tree spacing. Individual tree models exhibited clear outlines and accurate crown structures. The average reconstruction time was 1.05 h, producing point clouds with tens of millions of points. The reconstruction accuracy of the orchard model was 92.15 %, with centimetre-level scale accuracy. For tree height and canopy area, the mean relative errors were 2.11 % and 8.01 %, respectively. Compared to SFM-MVS and PIX4D, NERF-F outperformed in point cloud density, reconstruction accuracy, and measurement precision, providing a solid foundation for digital orchard development.