Forest Survey Research Articles

An Automated Pipeline for Extracting Forest Structural Parameters by Integrating UAV and Ground-Based LiDAR Point Clouds

In recent times, airborne and terrestrial laser scanning have been utilized to collect point cloud data for forest resource surveys, aiding in the estimation of tree and stand attributes over hectare-scale plots. In this study, an automated approach was devised to estimate the diameter at breast height (DBH) and tree height across the entire sample area, utilizing information acquired from terrestrial laser scanning (TLS) and airborne laser scanning (ULS). Centered around a meticulously managed artificial forest in Northern China, where Mongolian oak and Chinese Scots pine are the predominant species, both TLS and ULS operations were conducted concurrently on each plot. Subsequent to data collection, a detailed processing of the point cloud data was carried out, introducing an innovative algorithm to facilitate the matching of individual tree point clouds from ULS and TLS sources. To enhance the accuracy of DBH estimation, a weighted regression correction equation based on TLS data was introduced. The estimations obtained for the Chinese Scots pine plots showed a correlation of R2 = 0.789 and a root mean square error (RMSE) of 3.2 cm, while for the Mongolian oak plots, an improved correlation of R2 = 0.761 and a RMSE of 3.1 cm was observed between predicted and measured values. This research significantly augments the potential for non-destructive estimations of tree structural parameters on a hectare scale by integrating TLS and ULS technologies. The advancements hold paramount importance in the domain of large-scale forest surveys, particularly in the calibration and validation of aboveground biomass (AGB) estimations.

Realistic potential increases in carbon storage via timber rotation extensions: an analysis of the Pacific Northwest

Extensions of harvest rotation length are a commonly proposed method to increase carbon sequestration in forests that are managed for timber. However, several limitations constrain realistic storage potential in intensively managed forests. We present an analysis of the realistic potential for additional carbon sequestration via rotation extension across the Pacific Northwest of the United States, an important timber-producing region, taking into account specific limitations. We first assess the limitations on rotation length imposed by the stand age at which wood production would decline over the long term, and then incorporate the age at which trunk diameter surpasses a reasonable threshold for logging. Using publicly available forest survey data, we empirically model growth parameters across this region for use in this analysis. Despite uncertainties, we find some opportunities for rotation length extension in western Washington with variation by sub-region and timber species, emphasizing the importance of geography- and species-specific growth parameters for forest carbon management even within a general region. However, the total realistic potential for sequestration under this improved forest management scenario is small relative to gross emissions: the estimated cumulative additional sequestration in aboveground live biomass would offset one year of gross emissions in the case of Washington state, while a decadal-scale rotation extension implemented gradually over the landscape to avoid a total pause on commercial timber production would take on the scale of a century to achieve. Overall, practical considerations greatly limit the realistic potential of this carbon sequestration strategy.

Coupling of Forest Carbon Densities with Landscape Patterns and Climate Change in the Lesser Khingan Mountains, Northeast China

This research investigated the effects of the forest landscape composition and spatial distribution and local climate change’s lag effects on the carbon density of stands and provides a reference for optimizing the stand structure and sustainable management of forest resources in Xinqing District, Yichun City, Heilongjiang Province, China. Using second-class forest resource survey data of the Xinqing Forestry Bureau for 2007 and 2017, the forest carbon density, landscape pattern index and local forest climate were quantified by ArcGIS10.7, Fragstats4.2 and SPSS25, and a coupling coordination degree model was constructed to reflect their correlations. The overall broadleaved mixed forest area was larger in the new green area, and the overall forest productivity had improved in the past ten years. Forest management gradually improved from nonforest to forest land, resulting in a high degree of fragmentation in the surrounding landscape. The coupling research on the forest carbon density and the forest landscape pattern index and local climate index showed that, overall, the landscape pattern and the impact of climate change on the forest carbon density had a positive interaction; adjustments and improvements can be made to the forest carbon density in the poor-condition area by combining specific situations of the landscape pattern and climate change.

Progressive Adjacent-Layer coordination symmetric cascade network for semantic segmentation of Multimodal remote sensing images

Semantic segmentation of remote sensing images is a fundamental task in computer vision, with significant applications in forest and farmland cover surveys, geological disaster monitoring, and other related fields. The inclusion of digital surface models can enhance the segmentation performance compared to using unimodal imaging alone. However, most existing methods simply combine the features from both modalities without considering their differences and complementarity, leading to a loss of spatial details. In order to address this issue and improve segmentation accuracy, we propose a novel network called Progressive Adjacent-Layer Coordination Symmetric Cascade Network (PACSCNet). This network employs a two-stage fusion symmetric cascade encoder to leverage the similarities and differences between adjacent features for cross-layer fusion, thereby preserving spatial details. Additionally, our network includes a new dual-pyramid symmetric cascade decoder that extracts similarities in multimodal and cross-layer fusion features for merging. Furthermore, a pyramid residual integration module progressively integrates features at four scales to mitigate noise interference during large-scale fusion. Extensive experimental evaluations on the Vaihingen and Potsdam datasets demonstrate that PACSCNet achieves strong semantic segmentation performance, comparable to state-of-the-art approaches, in terms of accuracy and intersection-over-union. The source code and results of our proposed PACSCNet are publicly available at https://github.com/F8AoMn/PACSCNet.

Fire Affects Tree Growth, Water Use Efficiency and Carbon Sequestration Ecosystem Service of Pinus nigra Arnold: A Combined Satellite and Ground-Based Study in Central Italy

The Mediterranean basin is an area particularly exposed to fire risk due to its climate and fire-prone vegetation. In recent decades, the frequency and intensity of wildfires increased, leading to negative effects on forests, such as a decrease in tree growth or an increase in tree mortality, producing a relevant loss of carbon sequestration ecosystem service. This study of the impacts of fires on forests is fundamental for planning adequate forest management strategies aimed at recovering and restoring the affected areas. In this framework, our research delves into the effects of a forest fire that, in 2017, affected a forest of black pine (Pinus nigra Arnold) in Central Italy. Combining satellite and terrestrial analyses, this study evaluated the impact of the fire on tree growth, water use efficiency and carbon sequestration capacity. Our findings highlight the importance of using remote sensing for the accurate identification of fire-affected areas and precise planning of ground-based activities. However, the integration of satellite data with forest surveys and sampling has proven crucial for a detailed understanding of fire’s effects on trees. Dendrochronology and stable isotopes have revealed the post-fire growth decline and altered water usage of defoliated trees. Furthermore, the quantification of CO2 sequestration highlighted a significant reduction in carbon uptake by damaged trees, with severe implications for this ecosystem service.

Heavy browsing pressure by moose (Alces alces) can interfere with the objectives of ecosystem-based forest management

Forest management can enhance the ecosystem services forests provide, but it can also alter the capacity of forests to provide those services in the future. Ecosystem-based management has been widely implemented to mitigate the deterioration of forest ecosystems by reducing dissimilarities between managed and natural forests. Ecosystem-based management objectives generally aim to maintain community composition, stand structure, ecosystem functions and services. Heavy browsing pressure by ungulates at densities unrepresentative of past abundance could interfere with these objectives by chronically and selectively browsing tree regeneration following logging.Using a browsing exclusion experiment, we investigated how the effects of different moose browsing pressures on tree regeneration could interfere with ecosystem management objectives in boreal forests of Quebec, Canada. Management objectives in this region include, among others, reaching targeted coniferous trees densities for commercial harvest, favouring the conservation of birches (Betula spp.) to maintain soil quality and conserving fruit-bearing trees due to their ecological significance for biodiversity. We used path analyses to differentiate direct and indirect effects of moose on seedling density and mixed models to investigate how browsing affected sapling recruitment and whether resulting densities are compatible with ecosystem management objectives.By directly reducing the overall sapling density, moose browsing indirectly increased seedlings densities of all deciduous species. This indirect effect increased with browsing pressure and favoured raspberry cover, which was positively or negatively linked to seedling density, depending on tree species. Moderate and heavier browsing pressures reduced coniferous sapling density below the ecosystem management target. Moose suppressed birch saplings so that the minimal target of 50 birches ha−1 was not met under heavier browsing pressure. Sapling density of fruit-bearing trees was low under all browsing pressures when compared to exclosures.Policy implicationsHeavy moose browsing is preventing the achievement of ecosystem management objectives in our study area. Positive indirect effects on seedling densities suggest that reducing moose density could allow released seedlings to replenish the sapling layer, but we discuss why this would not guarantee reaching ecosystem management objectives. To help forest managers evaluate and respond to challenges posed by high ungulate densities, we suggest incorporating browsing inventories in national forest surveys to monitor this chronic disturbance and modulate decisions for forest ecosystem management.

Precise aboveground biomass estimation of plantation forest trees using the novel allometric model and UAV-borne LiDAR

IntroductionPlantation forest is an important component of global forest resources. The accurate estimation of tree aboveground biomass (AGB) in plantation forest is of great significance for evaluating the carbon sequestration capacity. In recent years, UAV-borne LiDAR has been increasingly applied to forest survey, but the traditional allometric model for AGB estimation cannot be directly used without the diameter at breast height (DBH) of individual trees. Therefore, it is practicable to construct a novel allometric model incorporating the crown structure parameters, which can be precisely extracted from UAV LiDAR data. Additionally, the reduction effect of adjacent trees on crown area (Ac) should be taken into account.MethodsIn this study, we proposed an allometric model depending on the predictor variables of Ac and trunk height (H). The UAV-borne LiDAR was utilized to scan the sample plot of dawn redwood (DR) trees in the test site. The raw point cloud was first normalized and segmented into individual trees, whose Acs and Hs were sequentially extracted. To mitigate the effects of adjacent trees, the initial Acs were corrected to refer to the potential maximum Acs under undisturbed growth conditions. Finally, the corrected Acs (Acc) and Hs were input into the constructed allometric model to achieve the AGBs of DR trees.Results and discussionAccording to accuracy assessment, coefficient of determination (R2) and root mean square error (RMSE) of extracted Hs were 0.9688 and 0.51 m; R2 and RMSE of calculated AGBs were 0.9432 and 10.91 kg. The unrestricted growth parts of the tree crowns at the edge of a plantation forest could be used to derive the potential maximum Ac. Compared with the allometric models for AGB estimation relying only on trunk H or on initial Ac and H, the novel allometric model demonstrated superior performance in estimating the AGBs of trees in a plantation forest.

Stable functional structure despite high taxonomic variability across fungal communities in soils of old-growth montane forests

BackgroundMajor advances over the past decade in molecular ecology are providing access to soil fungal diversity in forest ecosystems worldwide, but the diverse functions and metabolic capabilities of this microbial community remain largely elusive. We conducted a field survey in montane old-growth broadleaved and conifer forests, to investigate the relationship between soil fungal diversity and functional genetic traits. To assess the extent to which variation in community composition was associated with dominant tree species (oak, spruce, and fir) and environmental variations in the old-growth forests in the Jade Dragon Snow Mountain in Yunnan Province, we applied rDNA metabarcoding. We also assessed fungal gene expression in soil using mRNA sequencing and specifically assessed the expression of genes related to organic matter decomposition and nutrient acquisition in ectomycorrhizal and saprotrophic fungi.ResultsOur taxonomic profiling revealed striking shifts in the composition of the saprotrophic and ectomycorrhizal guilds among the oak-, fir-, and spruce-dominated forests. The core fungal microbiome comprised only ~ 20% of the total OTUs across all soil samples, although the overlap between conifer-associated communities was substantial. In contrast, seasonality and soil layer explained only a small proportion of the variation in community structure. However, despite their highly variable taxonomic composition, fungal guilds exhibited remarkably similar functional traits for growth-related and core metabolic pathways across forest associations, suggesting ecological redundancy. However, we found that the expression profiles of genes related to polysaccharide and protein degradation and nutrient transport notably varied between and within the fungal guilds, suggesting niche adaptation.ConclusionsOverall, our metatranscriptomic analyses revealed the functional potential of soil fungal communities in montane old-growth forests, including a suite of specialized genes and taxa involved in organic matter decomposition. By linking genes to ecological traits, this study provides insights into fungal adaptation strategies to biotic and environmental factors, and sheds light on the importance of understanding functional gene expression patterns in predicting ecosystem functioning.BiaDVnGNFcTVJp_FejwVZqVideo

Tree species identification and health assessment of forest sample plots based on UAV hyperspectral remote sensing technology

National forest resource information is an important part of national basic information in many countries, sample plot survey is the main means to obtain this information. With the progress of science and technology and social development, the demand for cost saving and scientific and technological innovation has led to the wide application of UAV remote sensing in forest resource survey. This study introduced the practice of UAV hyperspectral remote sensing survey in Yicheng, Hubei Province, described in detail how to realize tree species identification and health assessment in forest sample plots by means of unmanned aerial vehicles. The results showed that the classification results of tree species identification method based on UAV hyperspectral remote sensing technology had high accuracy, and the results of the tree health assessment method used had certain reliability. This method greatly improved the safety of work in hazardous areas and reduced the cost.

Application of 3D Laser Scanning Data for Forest Survey and Timber Volume Estimation

Application of 3D Laser Scanning Data for Forest Survey and Timber Volume Estimation

Features of natural renewal in pine-linden and larch-linden forest stands in Moscow

Problems of natural renewal of tree species (pine-linden and larch-linden stands) were studied in the territory of Forest experimental station, Moscow Timiryazev Agricultural Academy. Natural regeneration of tree species is one of the most urgent problems of forestry and forest park management in urban areas. Reforestation refers to complex natural processes that affect all components of biogeocenoses. The purpose of the research was to study natural regeneration in mixed stands in Moscow. Methods and results of field surveys of forest stands in permanent trial plots were described. The forest stands of permanent trial plots are mature and overmature, therefore at present there is a loss of large-sized pine and larch trees. Due to greater durability, larch falls off more slowly compared to pine. Further growth of pine-linden and larch-linden plantations depends on quantity and quality of undergrowth, its viability and growing conditions. In Forest experimental station, natural renewal prevails, represented mainly by the following species: maple and linden. Saplings of pine, larch, oak, and birch appear on some permanent test plots, but soon die due to unfavorable conditions. Oak undergrowth is also unreliable due to powdery mildew infection. The greatest loss of growth occurs in unfavorable growing conditions due to increased recreational loads.

Ensemble algorithms for modeling forest live fuel loads and multivariate probability proportional to size sampling in Kunming, Yunnan, China

Forest live fuel load (LFL) models of tree, shrub, and herbaceous layers were established using linear function (LF), power function (PF), K-nearest neighbor (KNN), gradient boosting regression tree (GBRT), and random forest (RF) algorithms to improve the fit and prediction accuracy and reduce the bias of forest LFL models. Multivariate probability proportional to size sampling (MPPS), probability proportional to size sampling (PPS), and probability proportional to frequency sampling (PPF) were used to obtain mean estimates of the forest LFLs, and the sampling accuracies were assessed. We utilized forest stock and biomass survey data (forest fire risk data and forest resource planning and design survey data) from Kunming, Yunnan, China. The results indicated that the RF model had the highest fit and prediction accuracy. The R2 values of the fitted models for the LFLs of Pinus yunnanensis, Quercus spp., Pinus armandi, Eucalyptus spp., other soft broadleaf species, other hard broadleaf species, other conifer species, shrubs, and herbaceous species were 0.79, 0.89, 0.86, 0.88, 0.91, 0.83, 0.88 0.88, and 0.92, respectively. The F-values were 4.652, 2.582, 2.565, 2.551, 2.738, 3.158, 16.156, 3.479, and 8.468, respectively. The sampling results of forest LFLs were much better for MPPS than for PPF and PPS. Moreover, the two-stage MPPS was the most efficient method for estimating tree, shrub, and herbaceous LFLs. The mean estimates of tree fuel loads, shrub fuel loads, and herbaceous fuel loads were 144.7760, 3.8094, and 0.6442 t/hm2, respectively. There were 555 overlapping subunits for the three types of forest LFLs. The proposed method can significantly reduce the cost of LFL surveys and improve estimation accuracy.

Belarusian Oak Forests Afflicted with Armillaria Root Rot during Their Mass Desiccation

During the period of the oak forests’ mass desiccation, Armillaria root rot became widespread, exacerbating the state of depression. The infection of stands by white sapwood rot caused by fungi of the genus Armillaria was more often detected in Bug-Polesia and Berezina-Pre-Polesia forest growth areas. The incidence of the disease in the oak forests of Belarus increases with the raising of the stands age and the decrease in their density. More often, the affliction with Armillaria root rot occurs in floodplain oak forests, as well as in the dry lands forest types with the most fertile soils. The oak’s proportion in the stand’s composition does not significantly affect the occurrence of the infection. Under the conditions of massive weakening of the Belarussian oak forests, facultative parasites from the Armillaria genus can occur as dangerous secondary pathogens that accelerate the death of weakened oak trees. Their pathogenicity persists in the north of the republic, where the negative role of other pathological factors in oak forests decreases. After the end of the depression period, the transition of Armillaria from a parasitic to a predominantly saprotrophic strategy was recorded. The affliction of oak forests by Armillaria root rot can act as one of the oak formation’s condition indicators, marking the depression phase in oak stands. Therefore, the spread of foci of Armillaria root rot should be monitored while conducting forest pathology surveys and monitoring the condition of oak forests.

Earth-Observation-Based Monitoring of Forests in Germany—Recent Progress and Research Frontiers: A Review

One-third of Germany’s land surface area is covered by forest (around 11.4 million hectares), and thus, it characterizes the landscape. The forest is a habitat for a large number of animal and plant species, a source of raw materials, important for climate protection, and a well-being refuge for people, to name just a few of its many functions. During the annual forest condition surveys, the crown condition of German forests is assessed on the basis of field samples at fixed locations, as the crown condition of forest trees is considered an important indicator of their vitality. Since the start of the surveys in 1984, the mean crown defoliation of all tree species has increased, now averaging about 25% for all tree species. Additionally, it shows a strong rise in the rate of dieback. In 2019, the most significant changes were observed. Due to the drastic changes in recent years, efforts are being made to assess the situation of the forest using different remote sensing methods. There are now a number of freely available products provided to the public, and more will follow as a result of numerous projects in the context of earth-observation (EO)-based monitoring and mapping of the forests in Germany. In 2020, the situation regarding the use of remote sensing for the German forest was already investigated in more detail. However, these results no longer reflect the current situation. The changes of the last 3 years are the content of this publication. For this study, 84 citable research publications were thoroughly analyzed and compared with the situation in 2020. As a major result, we found a shift in the research focus towards disturbance monitoring and a tendency to cover larger areas, including national-scale studies. In addition to the review of the scientific literature, we also reviewed current research projects and related products. In congruence to the recent developments in terms of publications in scientific journals, these projects and products reflect the need for comprehensive, timely, large-area, and complementary EO-based information around forests expressed in multiple political programs. With this review, we provide an update of previous work and link it to current research activities. We conclude that there are still gaps between the information needs of forest managers who usually rely on information from field perspectives and the EO-based information products.

Method and Device for Measuring the Diameter at Breast Height and Location of Trees in Sample Plots

The diameter at breast height (DBH) and location of trees are important factors when surveying forest resources and ecological functions. In this study, a device mainly comprising a self-made DBH-measuring instrument and positioning base station was used. The hardware consisted of two devices to simultaneously measure the DBH and location of trees within a sample plot. Specifically, DBH is acquired by processing angle data with an algorithm, and locations are obtained by a five-sided ranging and positioning algorithm based on the received signal strength indicator and ultra-wideband (UWB) sensor. Data uploading, storage and analysis are performed by an upper computer. The device was used for the actual measurement of eight 24 m × 24 m square plots. The measurements of this device are essentially consistent with those of the DBH tape and calliper, with biases of −0.03 cm and −0.29 cm, respectively. Compared with a compass for location measurement, the device had a mean range bias of 25.41 cm, overall bias along the X-axis of 2.40 cm and overall bias along the Y-axis of 1.99 cm. Therefore, the device is considered to be sufficiently portable and practical and can reduce the heavy workload for surveyors to meet the requirements of accurate and smart measurements in forest resource surveys.

Automatic Tree Height Measurement Based on Three-Dimensional Reconstruction Using Smartphone.

Tree height is a crucial structural parameter in forest inventory as it provides a basis for evaluating stock volume and growth status. In recent years, close-range photogrammetry based on smartphone has attracted attention from researchers due to its low cost and non-destructive characteristics. However, such methods have specific requirements for camera angle and distance during shooting, and pre-shooting operations such as camera calibration and placement of calibration boards are necessary, which could be inconvenient to operate in complex natural environments. We propose a tree height measurement method based on three-dimensional (3D) reconstruction. Firstly, an absolute depth map was obtained by combining ARCore and MidasNet. Secondly, Attention-UNet was improved by adding depth maps as network input to obtain tree mask. Thirdly, the color image and depth map were fused to obtain the 3D point cloud of the scene. Then, the tree point cloud was extracted using the tree mask. Finally, the tree height was measured by extracting the axis-aligned bounding box of the tree point cloud. We built the method into an Android app, demonstrating its efficiency and automation. Our approach achieves an average relative error of 3.20% within a shooting distance range of 2-17 m, meeting the accuracy requirements of forest survey.

Modeling of the Spatial Distribution of Forest Carbon Storage in a Tropical/Subtropical Island with Multiple Ecozones.

Visual data on the geographic distribution of carbon storage help policy makers to formulate countermeasures for global warming. However, Taiwan, as an island showing diversity in climate and topography, had lacked valid visual data on the distribution of forest carbon storage between the last two forest surveys (1993-2015). This study established a model to estimate and illustrate the distribution of forest carbon storage. This model uses land use, stand morphology, and carbon conversion coefficient databases accordingly for 51 types of major forests in Taiwan. An estimation in 2006 was conducted and shows an overall carbon storage of 165.65 Mt C, with forest carbon storage per unit area of 71.56 t C ha-1, where natural forests and plantations respectively contributed 114.15 Mt C (68.9%) and 51.50 Mt C (31.1%). By assuming no change in land use type, the carbon sequestration from 2006 to 2007 by the 51 forest types was estimated to be 5.21 Mt C yr-1 using historical tree growth and mortality rates. The result reflects the reality of the land use status and the events of coverage shifting with time by combining the two forest surveys in Taiwan.

A Tree Point Cloud Simplification Method Based on FPFH Information Entropy

LiDAR technology has been widely used in forest survey and research, but the high-resolution point cloud data generated by LiDAR equipment also pose challenges in storage and computing. To address this problem, we propose a point cloud simplification method for trees, which considers both higher similarity to the original point cloud and the area of the tree point cloud. The method first determines the optimal search neighborhood using the standard deviation of FPFH information entropy. Based on FPFH information entropy and Poisson disc sampling theory, the point cloud is partitioned and sampled. By optimizing the separation thresholds of significant feature points and less significant feature points using a genetic algorithm with the Hausdorff distance and point cloud area as the objective function, the final simplified point cloud is obtained. Validation with two point cloud data sets shows that the proposed method achieves good retention of the area information of the original point cloud while ensuring point cloud quality. The research provides new approaches and techniques for processing large-scale forest LiDAR scan point clouds, reducing storage and computing requirements. This can improve the efficiency of forest surveys and monitoring.

A Spatiotemporal Survey of Tikjda Forest Dynamics Over A 34-Year Period by Aerial Photographs

Abstract Obtaining accurate forest cover information and dynamics of land occupation, through time, such as the spatial extent and pattern of disturbance and recovery is essential knowledge and assistance for forest managers and a crucial basis for the protection and conservation of current forest resources. Because most recent researches have focused on forest field survey and monitoring, a land classification containing information on forest cover dynamics is critically needed. Over the last decades, advances in remote sensing technology have enabled an accurate classification of different land covers from several sensors and remotely sensed data. We presently retained Tikjda forest (Djurdjura southerner, Algeria) as a case study to investigate the possibility of aerial photos classification and to analyze the historical dynamics of the area using a change detection analysis of multi-temporal data. To classify the study area’s main cover types, we used photographs collected over a period of 34 years (i.e., from 1983 to 2017). The results revealed that in 2017, Tikjda forest was composed of forest areas (24.1%), degraded areas (49.7%), and barren areas (26.2%). Throughout the investigated period, the analysis revealed a notable increase in barren areas (+9.8%), and degraded areas (+14.4%), While forest areas experienced a significant decrease (−24.2%). Moreover, the results confirm the potential of aerial photographs for an accurate classification of forests.

Automatic Pear Extraction from High-Resolution Images by a Visual Attention Mechanism Network

At present, forest and fruit resource surveys are mainly based on ground surveys, and the information technology of the characteristic forest and fruit industries is evidently lagging. The automatic extraction of fruit tree information from massive remote sensing data is critical for the healthy development of the forest and fruit industries. However, the complex spatial information and weak spectral information contained in high-resolution images make it difficult to classify fruit trees. In recent years, fully convolutional neural networks (FCNs) have been shown to perform well in the semantic segmentation of remote sensing images because of their end-to-end network structures. In this paper, an end-to-end network model, Multi-Unet, was constructed. As an improved version of the U-Net network structure, this structure adopted multiscale convolution kernels to learn spatial semantic information under different receptive fields. In addition, the “spatial-channel” attention guidance module was introduced to fuse low-level and high-level features to reduce unnecessary semantic features and refine the classification results. The proposed model was tested in a characteristic high-resolution pear tree dataset constructed through field annotation work. The results show that Multi-Unet was the best performer among all models, with classification accuracy, recall, F1, and kappa coefficient of 88.95%, 89.57%, 89.26%, and 88.74%, respectively. This study provides important practical significance for the sustainable development of the characteristic forest fruit industry.