Watershed Algorithm Research Articles

Microtopography-Guided precision restoration of sandy lands through UAV: A case study in Hunshandake Sandy Land, China

Increasing desertification rates have adversely affected biodiversity and ecosystem functioning in sandy lands. Ecological restoration is an effective way to combat desertification. Specific microtopographic characteristics may facilitate vegetation growth, enhancing the success of restoration efforts. However, limited research to date has explored how microtopography may guide the “precision restoration” of sandy lands by supporting spatially continuous patterns of vegetation growth. Here, high-resolution unmanned aerial vehicle (UAV) multispectral data were used to identify individual species and extract microtopographic variables, and the relationships between vegetation growth and microtopography were characterized for the Hunshandake Sandy Land in China. The distribution of three dominant shrub species and grasses was investigated by comparing the performance of five popular machine-learning methods. An auto-marking watershed algorithm was then developed to discriminate individual semi-shrubs (Artemisia desertorum). Finally, a new vegetation growth index (VGI), calculated from the UAV-derived crown area, normalized difference vegetation index (NDVI), and canopy height, was used to characterize the relationships between vegetation growth and several microtopographic variables (aspect, slope, and a topographic wetness index [TWI]). With the highest species classification accuracy (94.36%) and individual discrimination rate (81%), areas with high humidity (TWI), gentle and shady slopes were found to most strongly support vegetation growth; for grasses, VGI was lower in artificially-restored regions than naturally-developed regions with similar microtopographic characteristics. These findings provide valuable guidance on the use of UAV to support diverse ecological restoration solutions in sandy lands with a “precision restoration” strategy, thereby improving the survival of key vegetation for sand restoration, especially in remote areas.

Colorectal cancer histopathology image analysis: A comparative study of prognostic values of automatically extracted morphometric nuclear features in multispectral and red-blue-green imagery.

Multispectral imaging (MSI) has been utilized to predict the prognosis of colorectal cancer (CRC) patients, however, our understanding of the prognostic value of nuclear morphological parameters of bright-field MSI in CRC is still limited. This study was designed to compare the efficiency of MSI and standard red-green-blue (RGB) images in predicting the prognosis of CRC. We compared the efficiency of MS and conventional RGB images on the quantitative assessment of hematoxylin-eosin (HE) stained histopathology images. A pipeline was developed using a pixel-wise support vector machine (SVM) classifier for gland-stroma segmentation, and a marker-controlled watershed algorithm was used for nuclei segmentation. The correlation between extracted morphological parameters and the five-year disease-free survival (5-DFS) was analyzed. Forty-seven nuclear morphological parameters were extracted in total. Based on Kaplan-Meier analysis, eight features derived from MS images and seven featured derived from RGB images were significantly associated with 5-DFS, respectively. Compared with RGB images, MSI showed higher accuracy, precision, and Dice index in nuclei segmentation. Multivariate analysis indicated that both integrated parameters 1 (factors negatively correlated with CRC prognosis including nuclear number, circularity, eccentricity, major axis length) and 2 (factors positively correlated with CRC prognosis including nuclear average area, area perimeter, total area/total perimeter ratio, average area/perimeter ratio) in MS images were independent prognostic factors of 5-DFS, in contrast with only integrated parameter 1 (P<0.001) in RGB images. More importantly, the quantification of HE-stained MS images displayed higher accuracy in predicting 5-DFS compared with RGB images (76.9% vs 70.9%). Quantitative evaluation of HE-stained MS images could yield more information and better predictive performance for CRC prognosis than conventional RGB images, thereby contributing to precision oncology.

A New Bubble Image Model Based on the Recognition of Bubble Flow

AbstractIn this study, a new ellipse‐fitting algorithm is proposed to achieve the reconstruction of bubble shapes in bubbly flow captured by a high‐speed camera in the gas–liquid two‐phase column reactor. Bubble flow patterns and geometric parameters in the experimental images are recognized and identified successfully, represented by means of the topological parameters. Three logical steps are carried out in detail. First, the area threshold and the circularity factors are established to identify the bubbles whether belonging to a single bubble or not. The overlapping bubbles in images can be separated from single bubbles based on a watershed segmentation algorithm. Second, a single bubble image and an overlapping bubble image are combined into one image. After that, statistical analysis for the size distributions and ellipse area bubbles is performed for further analysis and discussion. The advantage of this algorithm is that it can make use of a set of major and minor axes of an ellipse to capture the ellipse parameters more effectively. Simulation results are well agreed with experimental measurements. Moreover, it can be used to detect many ellipse‐like bubbles that are dispersed in high‐speed camera images, indicating that it is a better strategy for the recognition and identification of bubbly turbulent flow accurately.

UNSEG: unsupervised segmentation of cells and their nuclei in complex tissue samples

Multiplexed imaging technologies have made it possible to interrogate complex tissue microenvironments at sub-cellular resolution within their native spatial context. However, proper quantification of this complexity requires the ability to easily and accurately segment cells into their sub-cellular compartments. Within the supervised learning paradigm, deep learning-based segmentation methods demonstrating human level performance have emerged. However, limited work has been done in developing such generalist methods within the unsupervised context. Here we present an easy-to-use unsupervised segmentation (UNSEG) method that achieves deep learning level performance without requiring any training data via leveraging a Bayesian-like framework, and nucleus and cell membrane markers. We show that UNSEG is internally consistent and better at generalizing to the complexity of tissue morphology than current deep learning methods, allowing it to unambiguously identify the cytoplasmic compartment of a cell, and localize molecules to their correct sub-cellular compartment. We also introduce a perturbed watershed algorithm for stably and automatically segmenting a cluster of cell nuclei into individual nuclei that increases the accuracy of classical watershed. Finally, we demonstrate the efficacy of UNSEG on a high-quality annotated gastrointestinal tissue dataset we have generated, on publicly available datasets, and in a range of practical scenarios.

Hyperspectral imaging system for pre- and post-harvest defect detection in paprika fruit

External defects in paprika fruit (Capsicum annuum, L. var. angulosum) directly impact their storage and marketability, potentially leading to internal decay and altering fruit appearance. However, automatically detecting various paprika defects, including cracks, scars, diseases, and bruises, poses a significant challenge due to variations in defect size and visual similarities among different types of defects. This study proposed a comprehensive, accurate, and rapid defect detection method utilizing hyperspectral imaging (HSI) alongside multivariate analysis and imaging technologies. The reflective surface of fruit poses a primary limitation to efficient defect detection using imaging technologies. Therefore, a polarized HSI visible–near-infrared (VIS-NIR, 397–1000 nm) system was developed, and over 500 paprika fruit with various types of defects were imaged using the developed system. Additionally, fruit were manually bruised in different sizes to ensure spectral data variability and measured every 12 hours for two days. Spectral data (over 4580 spectral data) from the regions of interest (ROIs) of the paprika fruit samples, including manually bruised fruit, were extracted. Subsequently, a partial least squares discrimination analysis (PLS-DA) model was developed to distinguish between normal and defective fruit. Remarkably, the developed model achieved an accuracy of over 99 % in predicting normal and defective fruit. Moreover, the chemical images generated enhance the inspection process. The successive projection algorithm (SPA) and variable importance in projection (VIP) were further utilized for critical band selection, yielding an accuracy of 93 % and 98 % in the prediction set, respectively. Furthermore, a classification algorithm based on analysis of variance (ANOVA) was employed to determine the optimal wavelength band ratios (F-values). This approach yielded an impressive accuracy rate of over 88.8 % and further enhanced the performance of the band ratio by an improved watershed segmentation algorithm (IWSA) to reach an accuracy of 91.3 % in multi-defect detection. These findings offer a practical and efficient solution for the external quality inspection of paprika fruit, particularly in cases with naturally occurring defects. By integrating hyperspectral imaging, PLS-DA, band selection methods, ANOVA-based classification, and IWSA, this comprehensive analysis provides a reliable and practical means for identifying and categorizing defects in paprika fruit, ultimately enhancing product quality and consumer satisfaction.

The genetic and environmental etiology of novel frequency-driven regional parcellations of abnormal white matter.

The prevalence of white matter disease increases with age and is associated with cerebrovascular disease, cognitive decline, and risk for dementia. MRI measures of abnormal signal in the white matter (AWM) provide estimates of damage, however, regional patterns of AWM may be differentially influenced by genetic or environmental factors. With our data-driven regional parcellation approach, we created a probability distribution atlas using Vietnam Era Twin Study of Aging (VETSA) data (n = 475, mean age 67.6 years) and applied a watershed algorithm to define separate regional parcellations. We report biometrical twin modeling for five anatomically distinct regions: (1) Posterior, (2) Superior frontal and parietal, (3) Anterior and inferior frontal with deep areas, (4) Occipital, and (5) Anterior periventricular. We tested competing multivariate hypotheses to identify unique influences and to explain sources of covariance among the parcellations. Family aggregation could be entirely explained by additive genetic influences, with additive genetic variance (heritability) ranging from 0.69 to 0.79. Most genetic correlations between parcellations ranged from moderate to high (rg = 0.57-0.85), although two were small (rg = 0.35-0.39), consistent with varying degrees of unique genetic influences. This proof-of-principle investigation demonstrated the value of our novel, data-driven parcellations, with identifiable genetic and environmental differences, for future exploration.

A semantic segmentation-based algorithm for fast flotation bubble size distribution measurement

The visual characteristics of froths on flotation cell surfaces are crucial for the optimal control of the flotation process. However, previous research has primarily focused on optimizing markers to prevent the over-segmentation of froth images, often overlooking computational efficiency. This work proposes a deep learning-based froth segmentation algorithm that prioritizes both segmentation speed and quality. We propose a dataset construction strategy based on pseudo-labeling, which utilizes traditional methods to extract marker maps from froth images to build the dataset for deep learning. Subsequently, a semantic segmentation model is trained to derive marker maps from raw froth images. These marker maps are then employed in a marker-based watershed algorithm, enabling end-to-end segmentation of froth images. The proposed method is validated using froth images from a gold and nickel flotation plant, demonstrating the ability to accurately and efficiently identify different bubbles in real froth images without requiring complex pre-processing or post-processing steps. Industrial experiments show a tenfold increase in segmentation efficiency with virtually no loss of accuracy.

Tomato Plant Disease Identification via Deep Learning Technique

Tomatoes are edible berries that are utilized extensively around the world. Their scientific name is solanum lycopersicum. Many tomato varietals are primarily produced in temperate climates. Even though tomatoes are used as ingredients in food, they are botanically categorized as berries. Bacterial spot, Black mold, Gray spot, Late blight, and powdery mildew (PM) are the primary diseases that affect tomato plants. The most challenging process in this decade is also the early diagnosis of these disorders. In this study, the suggested method is applied to the assessment of disease severity and disease diagnosis in tomato plant images. There are various stages to the procedure. The tomato plant image is initially pre-processed using the Wavelet technique, and then the modified Watershed algorithm is used to segment the image using improved Double Sigmoid and Erosion (WSA-IDS&amp;E). The segmented image is then processed in the feature extraction stage in order to extract features. The Gray Level Co-occurrence Matrix (GLCM), histogram, illness area, and pixel-based retrieved characteristics are fed into the CNN model, a deep learning method for identifying diseases. In addition, a novel training method, WUDHOA, which adjusts the CNN’s weight, is introduced in this study to improve the performance of deep learning techniques. In addition, the severity of the tomato plant illness is measured in order to improve disease identification outcomes, where this severity assessment is taken into account using a fresh evaluation. Finally, the performance of the detection outcome is compared with other existing works to prove its efficiency. The conventional techniques have the lowest specificity ratings, such as [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text], whereas the WUDHOA achieved a specificity of 0.9384.

A worldwide coastal analysis of the climate wave systems

Wind generated waves of a sea state are generally the result of the superposition of wind sea and swells, making the frequency-direction wave energy distribution crucial for comprehending this behavior. Wave spectral partitioning methods provide groups of waves with similar characteristics, thus they have been usually applied to identify wind sea and swell. In addition, several swells can coexist in a sea state. This study develops a method to estimate the wave systems and analyze their characteristics over the coast worldwide using 32year (1989-2020) historical information and more than 10.000 locations. A wave system is considered as the long-term climate conditions prevailing over a frequency-direction wave energy area collecting similar environmental and physical characteristics. The method is applied for the hourly time series of the directional wave spectra. First, the watershed clustering algorithm is used and the partitions found are classified as wind sea or swells based on a wave age criterion. The information obtained from the swell spectral partitions is then used to estimate the probability of their occurrence within specific frequency-direction bins and the clustering algorithm is applied anew to this population in order to identify the number of significant long-term climate wave systems locally and their characteristics. Outcomes reveal that on average swells coexist with wind sea in approximately 70% of the global coast, whereas about 25% is predominantly dominated by pure swells and the wind sea dominates only in the 5%. Only the 2% of the global coast line presents one swell wave system. About 50% of the global coastal locations exhibit three and four, whereas the 15% presents two swell wave systems. The analysis shows that about 30% of the coastal locations present at least five swell wave systems, mostly on Pacific islands and enclosed seas.

New deterministic model for calculating mesh stiffness and damping of rough-surface gears considering elastic–plastic contact and energy-dissipation mechanism

For a long time, meshing parameter calculations and dynamic modeling of gear systems have assumed smooth tooth surfaces, disregarding the impact of actual three-dimensional (3D) microscopic topography. Based on elastic–plastic contact and energy-dissipation mechanisms, this work developed a comprehensive deterministic model that integrates 3D rough surface modeling, reconstruction, and contact analysis, for calculating the time-varying mesh stiffness (TVMS) and mesh damping (TVMD) of deterministic rough-surface gears. Distinct from fractal and statistical models, the deterministic model analyzes deterministic, real 3D rough surfaces rather than simple surfaces defined by a few parameters. Utilizing stochastic process theory, different-topography surfaces are obtained using height and spatial features and reconstructed employing watershed algorithm and ellipsoidal asperity fitting. Subsequently, a deterministic elastic–plastic contact model based on equivalent rough curved-surface contact is constructed, emphasizing local contact states and energy dissipation for calculating contact stiffness (CS) and damping (CD) of gears. Accordingly, a meshing characteristics analysis model is established for calculating TVMS, TVMD, and load sharing ratio (LSR). Experimental validation and comparisons with several statistical and gear contact models demonstrate the effectiveness, reliability, and superiority of the proposed deterministic model. Multi-parameter impact analysis reveals small Sq (meaning low roughness Sa) and Ssk and high load correlate with increased CS and TVMS, while CD and TVMD depend on their interaction and hardness. Overall, by bridging the micro-surface topography and macro-meshing parameters, our model provides essential insights for the design and manufacturing of high-performance gear to reduce vibration and noise.

Effect of particle size distribution variability on the permeability of hydrate-bearing sediments: A CFD study

Natural gas hydrates (NGH) are considered as a future clean energy in the era of carbon neutrality. The seepage behaviour of hydrate-bearing sediment (HBS) and its controlling factors are significant for developing effective production strategies. In this study, we aim to elucidate how the heterogeneous spatial distribution of clayey-silty sediment particles, particularly the non-uniform particle size distribution (PSD) affects the permeability of HBS. An algorithm was developed to generate the heterogeneous spatial distribution of sediment particles based on the PSD of hydrate-bearing cores extracted from Japan Nankai Trough (D50 = 122 μm). A series simulation of fluid flow in porous media based on CFD was conducted for the HBS with varying hydrate saturations (SH). The watershed segmentation algorithm was employed for the identification of pore structure, including pore space and pore throat with a detailed analysis on the pore structure evolution with SH. Host sediments with higher uniformity result in initial high permeability but the permeability reduction with increasing SH is more significant compared with that of lower uniformity. Pore throat width reduced to less than 5 μm with SH above 50% compared with its original width of ∼30 μm. Tortuosity exhibits a linear increase with SH, while permeability decreases exponentially with increasing tortuosity. We propose an improved permeability model (KCpro) based on the classical Kozeny-Carman model accounting for the change in specific surface area for non-uniform PSD in HBS with good agreement. This study confirmed the applicability of CFD modelling as a promising tool to study the seepage behaviour of HBS complementary to classical permeability experimental tests. Results of this study provide fundamental understanding on how permeability evolves in HBS with significant PSD variations in natural gas hydrate reservoirs in nature.

Pore structure characteristics of coral reef limestone: a combined polarizing microscope and CT scanning study

Coral reef limestone is a special class of geological medium formed through long-term deposition following the death of reef-building coral groups. Because it retains the skeletal structure of marine organisms during its formation, its pore structure is hyper-developed and complex. Deciphering the pore structure of the coral reef limestone is important because it is closely related to its macroscopic physical and mechanical properties. This study conducted a comprehensive analysis of the pore structure features of two types of coral reef limestone collected from the construction site of a nuclear power station located in the South China Sea using a combination of polarizing microscopy and CT scanning technologies. The fractal dimension of the pore structure of the treated reef limestone image was calculated, and the pore structure characteristics were statistically analyzed by considering several parameters including porosity, pore size, pore equivalent radius, shape factor, etc. In addition, the directional feature of the pore structure was explored. The results show that the improved watershed segmentation algorithm can accurately segment the pore structure of reef limestone images; both coral reef limestone specimens are loose of high porosity; the fractal dimension of pore structure lay between 1.58∼1.75, indicative of a high self-similarity; the pore size of the two coral reef limestone specimens is quite different, and the distribution of equivalent pore radius conforms to the normal distribution law; the pore structure of the two samples had obvious directionality, which can be quantified using a directional tensor. This study sheds light on future investigations linking the microscopic structure and macroscopic properties of coral reef limestones.

An image segmentation of adhesive droplets based approach to assess the quality of pesticide spray

Pesticide spray is a widely used chemical control method to minimize biological disasters in the agriculture industry. It is important to evaluate the efficacy and the quality of a pesticide sprayer, however, it cannot be conveniently achieved due to the lack of accessibility to the sprayed droplets and the intensity of the associated work. This paper proposes a novel method, based on an image processing-based approach, to assess the spray quality. The proposed method uses a combination of algorithms and criteria functions; such as, the maximum between-cluster variance algorithm, area threshold criteria, roundness factor, mathematical morphology operations, and the optimized watershed algorithm, to segment and assess the dark blue adhesive droplet images on a water-sensitive paper, placed in the crop field. In this work, three kinds of evaluation experiments are considered: (i) the manual analysis via droplet counting and processing, (ii) the automatic analysis by the commercially available droplet analyzer by Shenzhen DJI Co. Ltd., and (iii) the novel image processing based method introduced in this paper. The experimental results show a better consistency between the introduced novel method and the manual method. The former, however, provides a convenient and rapid way to assess the spray quality. It comes with an assessment algorithm along with an embedded device that provides a hardware view of the spray quality.

Examining the Use of the Watershed Algorithm for Segmenting Crown Closure on a Dry Land Forest

This paper uses a watershed algorithm to detect canopy cover in dryland forests. The study at to determine the best parameters of the watershed segmentation algorithm to obtain information on crown closure from filtered and unfiltered high and very high-resolution images. The best performance of each parameter combination of tolerance value (T), mean value (M), and variance value (V), which is written as C:[T]-[M]-[V], is determined based on the level of accuracy. This study uses Pleiades-1B and SPOT-6 images as primary digital data. The results showed that the low-pass filtered Pleiades-1B image showed the best performance with a combination of parameters C6-MF:[10]-[0.7]-[0.3], had an overall accuracy (OA) of 91.0% and an accuracy Kappa (KA) by 83.2%. While the low-pass filtered Spot-6 image shows the combination of parameters C7-MF:[10]-[0.8]-[0.2], which has an accuracy of 90.6% OA and 65.4% KA. This study concludes that the filtered image with a low-pass filter always gives more accurate results than the original data (without filter), both for Pleiades-1B and SPOT-6 images. The very high spatial resolution provides better accuracy than the high spatial resolution

Vegetation extraction in riparian zones based on UAV visible light images and marked watershed algorithm

The riparian zone is an area where land and water are intertwined, and vegetation is rich and complexly distributed. The zone can be directly involved in ecological regulation. In order to protect the ecological environment of the riparian zone, it is necessary to monitor the distribution of vegetation. However, there are many disturbing factors in extracting riparian vegetation, the most serious of which are water bodies with similar colours to the vegetation. To overcome the influence of water bodies on vegetation extraction from UAV imagery of riparian areas, this paper proposes a novel approach that combines the marked watershed algorithm with vegetation index recognition. First, the image is pre-segmented using edge detection, and the output is further refined with the marked watershed algorithm. Background areas are classified as potential regions for vegetation distribution. Subsequently, the final vegetation distribution is extracted from these potential vegetation areas using the vegetation index. The segmentation threshold for the vegetation index is automatically determined using the OTSU algorithm. The experimental results indicate that our method, when applied to UAV aerial imagery of the riparian zone, achieves an overall accuracy of over 94%, a user accuracy of over 97%, and a producer accuracy of over 93%.

Tree barriers height information extraction under power line based on watershed segmentation and local maximum algorithm

ABSTRACT In view of the problem that the traditional power line inspection cannot accurately obtain the height of the tree barriers under the power line, this study takes a forest area in Guilin City and a power line corridor in Yinchuan City as the research object, and combines watershed segmentation algorithm and local maximum algorithm to study the height of the tree barriers under power line. The results show that the watershed segmentation algorithm and the local maximum algorithm can accurately extract the height of tree barriers in power lines. In the Guilin experiment, there is a good correlation between the extracted height of different trees and the measured height, The correlation coefficients (R2) for osmanthus, eucalyptus, and pine trees are 0.9368, 0.8923, and 0.8227, respectively. In the experiment in Yinchuan, the tree height extracted by unmanned aerial vehicles (UAV) and LiDAR also has a good correlation with the measured height, with correlation coefficients (R2) of 0.926 and 0.952, respectively. These two experiments have verified the feasibility of watershed segmentation algorithm and local maximum algorithm in extracting the height of tree barriers under power lines, which is of great significance for monitoring and warning of tree barriers under power lines.

Estimating the Aboveground Biomass of Robinia pseudoacacia Based on UAV LiDAR Data

Robinia pseudoacacia is widely planted in the Loess Plateau as a major soil and water conservation tree species because of its dense canopy, complex structure, and strong soil and water conservation ability. The precise measurement of small-scale locust forest biomass is crucial to monitoring and evaluating the carbon sequestration functions of soil and water conservation vegetation. This study focuses on an artificial locust forest planted in the early 1990s in Caijiachuan Basin, Jixian County, Shanxi Province. A drone equipped with LiDAR was used to obtain point cloud data and generate a canopy height model. A watershed segmentation algorithm was used to identify tree vertices and extract individual trees. A relationship model between tree height, diameter at breast height, and biomass, combined with sample survey data, was established to explore the spatial distribution of biomass in the artificial locust forest at the level of the entire basin. The results show the following: (1) the structural parameters of locust extracted using UAV point cloud data have a good degree of fit and accuracy, and the recall rate is 72.7%; (2) the average error rate of the extracted maximum tree height value of locust is 7%, that of the minimum tree height value is 14%, and that of the average tree height value is 18%; (3) the average error rate of the extracted maximum diameter at breast height of locust is 15%, that of the minimum diameter at breast height is 37%, and that of the average diameter at breast height is 36%; and (4) the average error rate of the biomass estimation of locust calculated using point cloud data is 16.0%.

Monitoring acoustic cavitation effects in tissues under the action of HIFU based on ultrasound images

The cavitation effect plays a key role in the field of High-intensity focused ultrasound treatment. By monitoring cavitation to adjust the irradiation dose, effective control of cavitation effects can be achieved, which is crucial for ensuring the safety and efficacy of HIFU therapy. In this paper, a method for monitoring cavitation effects in tissues based on ultrasound images is proposed. First, support vector machine is used to rapidly and accurately determine whether cavitation has occurred. Then, the improved watershed algorithm is employed to extract the regions where cavitation occurs, and the extracted cavitation region images are reconstructed using the total focusing method based on sparse array elements. Finally, to validate the feasibility and stability of the proposed method and investigate the development patterns of cavitation effects. An experimental system was established to conduct cavitation effect monitoring experiments on fresh bovine liver and a simulated human tissue model. The experimental results showed that, except for the recall rate in the fresh bovine liver test set, all classification metrics are greater than 0.9, demonstrating the practicality of the model established in this paper. The improved watershed algorithm achieved average values of Dice Similarity Coefficient, accuracy and precision at 93.69%, 96.74% and 92.23% respectively, representing a significant improvement compared to the traditional watershed method. The proposed total focusing method based on sparse array elements can effectively improve imaging efficiency while ensuring reconstruction accuracy.

EWSeg: a fast segmentation algorithm for images based on edge linking and watershed constraints

In this paper, we propose a two-stage algorithm, named watershed-constrained image segmentation, for exploring complete edge-closed regions from edges. In the first stage, the input image is pre-processed and the image gradient information is obtained using a gradient operator. Anchors are then obtained from the gradient information. Finally, initial edges are obtained by intelligently connecting the anchors. In the second stage, a marker-based watershed algorithm is adopted to obtain marker points from the gradient information obtained in the first stage. A Gaussian filtered image is then used as the input image to obtain a watershed hyper-segmented edge map. Finally, complete edge-closed regions are obtained by combining the initial edges and the hyper-segmented edge map and searching for weak edges. The image segmentation results are then obtained from the edge-closed regions, demonstrating the excellent performance of our proposed algorithm on various images and videos.

Tree Crown Segmentation and Diameter at Breast Height Prediction Based on BlendMask in Unmanned Aerial Vehicle Imagery

The surveying of forestry resources has recently shifted toward precision and real-time monitoring. This study utilized the BlendMask algorithm for accurately outlining tree crowns and introduced a Bayesian neural network to create a model linking individual tree crown size with diameter at breast height (DBH). BlendMask accurately outlines tree crown shapes and contours, outperforming traditional watershed algorithms in segmentation accuracy while preserving edge details across different scales. Subsequently, the Bayesian neural network constructs a model predicting DBH from the measured crown area, providing essential data for managing forest resources and conducting biodiversity research. Evaluation metrics like precision rate, recall rate, F1-score, and mAP index comprehensively assess the method’s performance regarding tree density. BlendMask demonstrated higher accuracy at 0.893 compared to the traditional watershed algorithm’s 0.721 accuracy based on experimental results. Importantly, BlendMask effectively handles over-segmentation problems while preserving edge details across different scales. Moreover, adjusting parameters during execution allows for flexibility in achieving diverse image segmentation effects. This study addresses image segmentation challenges and builds a model linking crown area to DBH using the BlendMask algorithm and a Bayesian neural network. The average discrepancies between calculated and measured DBH for Ginkgo biloba, Pinus tabuliformis, and Populus nigra varitalica were 0.15 cm, 0.29 cm, and 0.49cm, respectively, all within the acceptable forestry error margin of 1 cm. BlendMask, besides its effectiveness in crown segmentation, proves useful for various vegetation classification tasks like broad-leaved forests, coniferous forests, and grasslands. With abundant training data and ongoing parameter adjustments, BlendMask attains improved classification accuracy. This new approach shows great potential for real-world use, offering crucial data for managing forest resources, biodiversity research, and related fields, aiding decision-making processes.