Automated tree crown detection and size estimation using multi-scale analysis of high-resolution satellite imagery

We tested an automated multi-scale approach for detecting individual trees and estimating tree crown geometry using high spatial resolution satellite imagery. Individual tree crowns are identified as local extrema points in the Laplacian-of-Gaussian scale-space pyramid that is constructed based on linear scale-space theory. The approach simultaneously detects tree crown centres and estimates tree crown sizes (radiuses). We evaluated our method using two 0.6-m resolution QuickBird images of a forest site that underwent a large shift in tree density between image captures due to drought-associated mortality. The automated multi-scale approach produced tree count estimates with an accuracy of 54% and 73% corresponding to the dense and sparse forests, respectively. Estimated crown diameters were linearly correlated with field-measured crown diameters (r = 0.73–0.86). Tree count accuracies and size estimates were comparable with alternative methods. Future use of the presented approach is merited based on the results of our study, but requires further investigation in a broader range of forest types.     

Tree crown segmentation based on a tree crown density model derived from Airborne Laser Scanning

ABSTRACT

This letter describes a new algorithm for automatic tree crown delineation based on a model of tree crown density, and its validation. The tree crown density model was first used to create a correlation surface, which was then input to a standard watershed segmentation algorithm for delineation of tree crowns. The use of a model in an early step of the algorithm neatly solves the problem of scale selection. In earlier studies, correlation surfaces have been used for tree crown segmentation, involving modelling tree crowns as solid geometric shapes. The new algorithm applies a density model of tree crowns, which improves the model’s suitability for segmentation of Airborne Laser Scanning (ALS) data because laser returns are located inside tree crowns. The algorithm was validated using data acquired for 36 circular (40 m radius) field plots in southern Sweden. The algorithm detected high proportions of field-measured trees (40–97% of live trees in the 36 field plots: 85% on average). The average proportion of detected basal area (cross-sectional area of tree stems, 1.3 m above ground) was 93% (range: 84–99%). The algorithm was used with discrete return ALS point data, but the computation principle also allows delineation of tree crowns in ALS waveform data.     

Evaluating wood volume estimates derived from Quickbird imagery with GEOBIA for Pinus nigra trees in the Pentalofo forest,northern Greece

The objective of this research was to evaluate wood volume estimates of Pinus nigra trees in forest stands, which were derived utilizing Geographic Object-Based Image Analysis. Information on forest parameters such as wood volume and number of trees is useful for forest management facilitating forest sustainability. Most of the existing approaches used to estimate wood volume of forest trees require field measurements, which are laboursome. In this study, the collected field data were utilized only in order to investigate the results. Wood volume was estimated based on an individual tree crown approach and using monoscopic satellite images in combination with allometric data. The study area is the Pentalofo forest, which is located in Kozani prefecture in western Macedonia, Northern Greece. About 1 plot surface of 0.1143 ha was utilized. During the preprocessing, a pansharpened image was produced from two Quickbird satellite images (one multispectral image of 2.4 m spatial resolution and one panchromatic image of 0.6 m spatial resolution). Bands of this image were utilized single or in combination in order to delineate the tree crowns individually. The allometric equation served in order to calculate the tree Diameter at Breast Height (DBH) utilizing the detected tree crowns. The evaluation was conducted on three levels: (i) number of trees, (ii) DBH class distribution and (iii) wood volume. On the third level, the evaluation procedure was conducted twice; once using field height and once without. The difference between the results and the field data for the wood volume reached a maximum of approximately 30%. The total number of trees was exactly the same as counted in the field and the DBH distribution showed a tendency for the trees to move to a higher DBH class, resulting in an overestimation of the wood volume.     

A refined method for estimating capsule crops in individual jarrah (Eucalyptus marginata) crowns

An accurate measure of the number of capsules in the crowns of jarrah (Eucalyptus marginata) trees is needed to assess the potential for seedling regeneration prior to silvicultural treatment in jarrah forests. The current method of estimating capsule crops on jarrah trees uses stem diameter and estimates of capsule density in the crown, but has not been fully validated. In this study, we sought to develop an accurate and practical method of assessing capsule crops in the crowns of individual jarrah trees. We did this by measuring a number of tree characteristics prior to felling them. A total of 24 trees were selected, spanning a range of sizes and crown conditions, and the total number of capsules was counted for each tree. Multiple linear regression was used to model capsule number against various combinations of eight different tree/crown variables, with model fit compared using an adjusted coefficient of determination (adjR²). The final model recommended for field use included three easily measured variables (stem diameter, subjective assessment of capsule density, and subjective assessment of capsule clump distribution in the crown) and had a high degree of predictability (adjR² = 0.83), which was the same as that of the full model. This method substantially improved estimates of crown capsule numbers compared with the method currently used (adjR² increased from 0.29 to 0.83), which tended to underestimate canopy capsule numbers.     

Canopy surface reconstruction from a LiDAR point cloud using Hough transform

Tree height and canopy volume are critical forestry parameters that are used to derive estimates of growth, carbon sequestration, standing timber volume, and biomass. Through the use of light detection and ranging, these attributes can be estimated rapidly over large areas. At the stand level, estimates of these attributes have been derived successfully from canopy height models. However, a number of challenges identified in using canopy height models remain, such as correcting for height underestimation and canopy surface irregularities, such as data pits and holes that may result from acquisition and/or post-processing, and consistent delineation of tree crowns – all of which can limit the accurate retrieval of individual tree and crown attributes. In this letter, a novel canopy model is proposed in which individual tree crowns are represented as objects for which delineations can be derived through geometric operations. The technique is based on fitting simple geometric shapes to the raw light detection and ranging point cloud and thereby compensates for this underestimation, reduces data size, and allows effective and efficient modelling at the individual tree level.     

Tree extraction from multi-scale UAV images using Mask R-CNN with FPN

ABSTRACT

Tree detection and counting have been performed using conventional methods or high costly remote sensing data. In the past few years, deep learning techniques have gained significant progress in the remote sensing area. Namely, convolutional neural networks (CNNs) have been recognized as one of the most successful and widely used deep learning approaches and they have been used for object detection. In this paper, we employed a Mask R-CNN model and feature pyramid network (FPN) for tree extraction from high-resolution RGB unmanned aerial vehicle (UAV) data. The main aim of this paper is to explore the employed method in images with different scales and tree contents. For this purpose, UAV images from two different areas were acquired and three big-scale test images were created for experimental analysis and accuracy assessment. According to the accuracy analyses, despite the scale and the content changes, the proposed model maintains its detection accuracy to a large extent. To our knowledge, this is the first time a Mask R-CNN model with FPN has been used with UAV data for tree extraction.     

A tissue-specific adaptive texture filter for medical ultrasound images

The interpretation of ultrasound images remains a difficult task and the opinion of different doctors is generally not unequivocal. Therefore, there is a growing interest in the field of computer-aided diagnosis. In the field of medical image processing, computer-aided diagnosis includes image enhancement to facilitate visual interpretation, automatic indication of affected areas, organs and other regions of medical interest, the performance of automatic measurements and image registration. In this article, we introduce a new algorithm for ultrasound image enhancement that employs a multivariate texture classifier based on the co-occurrence matrix, which, in combination with an adaptive texture smoothing filter, is used to enhance the visual difference between and improve boundary detection between healthy neonatal brain tissue and tissue affected by periventricular leukomalacia. For a quantitative comparison, we delineate the periventricular leukomalacia-affected regions with two different active contours before and after processing 10 images with the proposed technique and several speckle filters from the literature. The semi-automatic delineations thus obtained are compared with the manual delineations of a neonatologist. In all cases, the average delineation achieved with the proposed technique is closer to that of the manual expert delineation than when the images are processed with the other techniques.     

A novel algorithm for the objective detection of tropical cyclone centres using infrared satellite images

Tropical cyclones (TCs) are one of nature’s most destructive phenomena, and a key element in forecasting TC tracks is the ability to accurately detect TC centres. In this paper, a novel algorithm has been proposed to objectively detect TC centres using infrared satellite images. Pyramid representation and optical flow technique are utilized to construct the cloud motion wind (CMW) field of each cyclone, and thereafter the centre is determined by analysing the constructed CMW field. Ten TCs formed in the Northwestern Pacific Ocean in 2014 have been tested to evaluate the performance of the present method, and TC Halong and Rammasun were analysed in detail as instances. Experimental results comparing with forecast track derived from Unisys Weather show that the proposed method provides an accurate detection of TC centres. The present algorithm has a potential to be employed to assist forecasters to detect TC centres.     

Change detection in multispectral images based on multiband structural information

Change vector analysis (CVA) and spectral angle mapper (SAM) are usually used to generate difference image in change detection of multispectral images. Although CVA and SAM can describe the difference between multispectral images, they are defined mathematically and lack support of human visual system (HVS) theory. Advanced structural similarity (ASSIM) complies with the pattern that human perceives the changes occurred in an objective scene. Nevertheless, ASSIM was designed for single band images and cannot be used for extracting multiband structural information from multispectral images directly. Therefore, we first propose two strategies to extract multiband structural information from multiband images. Then, we propose the approaches based on multiband structural information for change detection in multispectral images. Experimental results from one semisynthetic data set and two real data sets acquired by Sentinel-2A and QuickBird satellites validate the effectiveness of the proposed approaches.     

Estimation of forest stand diameter class using airborne lidar and field data

In this study, a method for estimating the stand diameter at breast height (DBH) classes in a South Korea forest using airborne lidar and field data was proposed. First, a digital surface model (DSM) and digital terrain model (DTM) were generated from the lidar data that have a point density of 4.3 points/m², then a tree canopy model (TCM) was created by subtracting the DTM from the DSM. The tree height and crown diameter were estimated from the rasterized TCM using local maximum points, minimum points and a circle fitting algorithm. Individual tree heights and crown diameters were converted into DBH using the allometric equations obtained from the field survey data. We calculated the proportion of the total number of individual trees belonging to each DBH class in each stand to determine the stand DBH class according to the standard guidelines. More than 60% of the stand DBH classes were correctly estimated by the proposed method, and their area occupied over 80% of the total forest area. The proposed method generated more accurate results compared to the digital forest type map provided by the government.     

Automated brain tumour detection and segmentation using superpixel-based extremely randomized trees in FLAIR MRI

Purpose

We propose a fully automated method for detection and segmentation of the abnormal tissue associated with brain tumour (tumour core and oedema) from Fluid- Attenuated Inversion Recovery (FLAIR) Magnetic Resonance Imaging (MRI).

Methods

The method is based on superpixel technique and classification of each superpixel. A number of novel image features including intensity-based, Gabor textons, fractal analysis and curvatures are calculated from each superpixel within the entire brain area in FLAIR MRI to ensure a robust classification. Extremely randomized trees (ERT) classifier is compared with support vector machine (SVM) to classify each superpixel into tumour and non-tumour.

Results

The proposed method is evaluated on two datasets: (1) Our own clinical dataset: 19 MRI FLAIR images of patients with gliomas of grade II to IV, and (2) BRATS 2012 dataset: 30 FLAIR images with 10 low-grade and 20 high-grade gliomas. The experimental results demonstrate the high detection and segmentation performance of the proposed method using ERT classifier. For our own cohort, the average detection sensitivity, balanced error rate and the Dice overlap measure for the segmented tumour against the ground truth are 89.48 %, 6 % and 0.91, respectively, while, for the BRATS dataset, the corresponding evaluation results are 88.09 %, 6 % and 0.88, respectively.

Conclusions

This provides a close match to expert delineation across all grades of glioma, leading to a faster and more reproducible method of brain tumour detection and delineation to aid patient management.

     

Self-supervised anomaly detection,staging and segmentation for retinal images

Unsupervised anomaly detection (UAD) is to detect anomalies through learning the distribution of normal data without labels and therefore has a wide application in medical images by alleviating the burden of collecting annotated medical data. Current UAD methods mostly learn the normal data by the reconstruction of the original input, but often lack the consideration of any prior information that has semantic meanings. In this paper, we first propose a universal unsupervised anomaly detection framework SSL-AnoVAE, which utilizes a self-supervised learning (SSL) module for providing more fine-grained semantics depending on the to-be detected anomalies in the retinal images. We also explore the relationship between the data transformation adopted in the SSL module and the quality of anomaly detection for retinal images. Moreover, to take full advantage of the proposed SSL-AnoVAE and apply towards clinical usages for computer-aided diagnosis of retinal-related diseases, we further propose to stage and segment the anomalies in retinal images detected by SSL-AnoVAE in an unsupervised manner. Experimental results demonstrate the effectiveness of our proposed method for unsupervised anomaly detection, staging and segmentation on both retinal optical coherence tomography images and color fundus photograph images.     

MLS-assisted validation of WorldView-2 panchromatic image for estimating Pinus sylvestris crown height

High spatial resolution satellite imaging has the advantages of both fine scale and large coverage that indicate the potential for measuring forest morphologies. However, because of the aerial view, imaging has limited capacity of explicitly deriving the under-crown structural parameters. A possible solution is to explore the relationships between this kind of variables such as crown height (CH) and the feature parameters readily derived from the satellite images. However, field sampling of the training data is not a trivial task. To handle this issue, this study attempted the state-of-the-art remote sensing technology of vehicle-based mobile laser scanning (MLS) for collecting the sample data. Evaluation for the case of the Scots pine (Pinus sylvestris) trees has preliminarily validated the plan. That is, MLS mapping enabled the parameter of CH to be estimated from WorldView-2 panchromatic images.     

A novel change detection method combined with registration for SAR images

In this letter, we propose a novel change detection method combined with image registration. Inspired by the mutual beneficial property of change detection and image registration, we model these two tasks into a unified process and represent this process as an optimization problem. Then, we use the alternating direction method (ADM) algorithm with proper initialization to solve the optimization problem. Finally, we use spatial information to refine the changed component obtained by the ADM algorithm. Experimental results accomplished over synthetic data and RADARSAT-2 synthetic aperture radar (SAR) images demonstrate that the proposed method can improve both the accuracy of image registration and change detection.     

Boundary-rendering network for breast lesion segmentation in ultrasound images

Breast Ultrasound (BUS) has proven to be an effective tool for the early detection of cancer in the breast. A lesion segmentation provides identification of the boundary, shape, and location of the target, and serves as a crucial step toward accurate diagnosis. Despite recent efforts in developing machine learning algorithms to automate this process, problems remain due to the blurry or occluded edges and highly irregular nodule shapes. Existing methods often produce over-smooth or inaccurate results, failing the need of identifying detailed boundary structures which are of clinical interest. To overcome these challenges, we propose a novel boundary-rendering framework that explicitly highlights the importance of boundary for automated nodule segmentation in BUS images. It utilizes a boundary selection module to automatically focuses on the ambiguous boundary region and a graph convolutional-based boundary rendering module to exploit global contour information. Furthermore, the proposed framework embeds nodule classification via semantic segmentation and encourages co-learning across tasks. Validation experiments were performed on different BUS datasets to verify the robustness of the proposed method. Results show that the proposed method outperforms states-of-art segmentation approaches (Dice=0.854, IOU=0.919, HD=17.8) in nodule delineation, as well as obtains a higher classification accuracy than classical classification models.     

Improving the estimation of soil-available nutrients at the sub-field scale using time-series UAV observations

ABSTRACT

Soil-available nutrients (SANs)are essential for crop growth and yield formation. Appropriate variable rate fertilization (VRF) can control SAN at a normal level to avoid unnecessary damage to sustainable production capacity. The precondition of optimizing the application of VRF is obtaining the real-time status of SAN. A new method for SAN estimation has been proposed by integrating modified World Food Studies (WOFOST) and time-series satellite remote sensing (RS) data. This method can provide field scale SAN estimations with high accuracy. However, the estimation accuracy at a subfield scale was low for VRF application because of the poor spatial resolution of common satellite imagery. In this letter, the subfield SAN estimations were optimized to ensure the VRF value. Time-series multispectral images acquired by an unmanned aerial vehicle (UAV) were used to replace common satellite data, and the SAN values for haplic phaeozem in selected spring maize plot in Hongxing Farm (48°09? N, 127°03? E) were estimated. Based on the field SAN data, the estimation accuracies using satellite data and UAV data were analyzed. The results show that the UAV data improved SAN estimations at the subfield scale).     

Development of a multi-scale object-based shadow detection method for high spatial resolution image

Effective treatment of shadows generated by the obstruction of trees and buildings is an inevitable task for extracting detailed spectral and spatial information from urban high-resolution images. Object-based shadow detection methods can take full advantages of spatial features in the urban very high resolution (VHR) images. However, the effect of different segmentation parameters for detecting shadows has not been well studied. In this study, we proposed an object-based method for shadow detection on urban high-resolution image and addressed quantitative assessment of segmentation. In proposed object-based method, a multi-scale segmentation method, known as fractal net evolution approach (FNEA), was employed to generate primitive objects; then, three spectral properties of shadows were fused based on Dempster–Shafer (D–S) evidence theory to identify shadows. In quantitative assessment, a method for ordering significance of parameters and deriving optimal parameters based on orthogonal experimental design was proposed to evaluate the impact of different segmentation variables on the accuracy of shadow detection. Experimental results indicate that the best overall accuracy (OA) for shadow detection of our method was 89.60% after segmentation parameters’ optimization and scale is the most influential parameter of FNEA segmentation parameters in determining the performance of shadow detection.     

Glaucoma risk index: Automated glaucoma detection from color fundus images

Glaucoma as a neurodegeneration of the optic nerve is one of the most common causes of blindness. Because revitalization of the degenerated nerve fibers of the optic nerve is impossible early detection of the disease is essential. This can be supported by a robust and automated mass-screening. We propose a novel automated glaucoma detection system that operates on inexpensive to acquire and widely used digital color fundus images. After a glaucoma specific preprocessing, different generic feature types are compressed by an appearance-based dimension reduction technique. Subsequently, a probabilistic two-stage classification scheme combines these features types to extract the novel Glaucoma Risk Index (GRI) that shows a reasonable glaucoma detection performance. On a sample set of 575 fundus images a classification accuracy of 80% has been achieved in a 5-fold cross-validation setup. The GRI gains a competitive area under ROC (AUC) of 88% compared to the established topography-based glaucoma probability score of scanning laser tomography with AUC of 87%. The proposed color fundus image-based GRI achieves a competitive and reliable detection performance on a low-priced modality by the statistical analysis of entire images of the optic nerve head.     

An extended maxima transform-based region growing algorithm for convective cell detection on satellite images

Convective cell detection is a critical step for tracking and forecasting of Mesoscale Convective Systems from geostationary satellite infrared (IR) data. Conventional threshold methods for identifying convective cell depend on the choice of threshold. Adjacent convective cells cannot be well distinguished because of the influence of anvil cloud. To address this problem, a new algorithm called extended maxima transform–based region growing (EMTRG) is proposed. First, EMTRG algorithm uses extended maxima transform to generate seed points of convective cell, and applies neighbourhood criterion to cluster adjacent seed points. Second, the merger and split times of pair-wise seed clusters are counted, and a merger criterion is utilized to decide whether pair-wise seed clusters should be merged. The algorithm is applied to various case studies over China. Experimental results on geostationary satellite IR images show that the proposed algorithm distinguishes adjacent convective cells region efficiently.     

Image reconstruction by minimizing second-order moment

Statistical methods have received much attention in image reconstruction, especially when the projection data are not sufficient in number, since they usually give more reliable results in such a case than other methods. In this paper, we propose a new statistical method which minimizes the second-order moment of an object that indicates the variance and randomness in a statistical structure. By minimizing the second-order moment, the resulting image reflects the statistical structure imposed by the available information and is biased toward a flat gray structure in the absence of information. The computer simulation results show a good convergence behavior of the proposed method. This method was successfully applied to ultrasound attenuation CT using a sponge phantom.