Watershed Transformation Research Articles

Comparing Deep Learning and MCWST Approaches for Individual Tree Crown Segmentation

Abstract. Accurate segmentation of individual tree crowns (ITC) segmentation is essential for investigating tree-level based growth trends and assessing tree vitality. ITC segmentation using remote sensing data faces challenges due to crown heterogeneity, overlapping crowns and data quality. Currently, both classical and deep learning methods have been employed for crown detection and segmentation. However, the effectiveness of deep learning based approaches is limited by the need for high-quality annotated datasets. Benefiting from the BaKIM project, a high-quality annotated dataset can be provided and tested with a Mask Region-based Convolutional Neural Network (Mask R-CNN). In addition, we have used the deep learning based approach to detect the tree locations thus refining the previous Marker controlled Watershed Transformation (MCWST) segmentation approach. The experimental results show that the Mask R-CNN model exhibits better model performance and less time cost compared to the MCWST algorithm for ITC segmentation. In summary, the proposed framework can achieve robust and fast ITC segmentation, which has the potential to support various forest applications such as tree vitality estimation.

Fusion of airborne multimodal point clouds for vegetation parameter correction extraction in burned areas.

Most experimental studies use unimodal data for processing, the RGB image point cloud cannot separate the shrub and tree layers according to the visible vegetation index, and the airborne laser point cloud is difficult to distinguish between the ground and grass ranges, to address the above problems, a multi-band information image fusing the LiDAR point cloud and the RGB image point cloud is constructed. In this study, data collected from UAV platforms, including RGB image point clouds and laser point clouds, were used to construct a fine canopy height model (using laser point cloud data) and high-definition digital orthophotos (using image point cloud data), and the orthophotos were fused with a canopy height model (CHM) by selecting the Difference Enhancement Vegetation Index (DEVI) and Normalised Green-Blue Discrepancy Index (NGBDI) after comparing the accuracy of different indices. Morphological reconstruction of CHM + DEVI/NGBDI fusion image, remove unreasonable values; construct training samples, using classification regression tree algorithm, segmentation of the range of the burned areas and adaptive extraction of vegetation as trees, shrubs and grasslands, tree areas as foreground markers using the local maximum algorithm to detect the tree apexes, the non-tree areas are assigned to be the background markers, and the Watershed Transform is performed to obtain the segmentation contour; the original laser point cloud is divided into chunks according to the segmented single-tree contour, and the highest point is traversed to search for the highest point, and corrected for the height of the single-tree elevations one by one. Accuracy analysis of the vegetation information extracted by the method with the measured data showed that the improved method increased the overall recall rate by 4.1%, the overall precision rate by 3.7%, the overall accuracy F1 score by 3.9%, and the tree height accuracy by 8.8%, 1.4%, 1.7%, 6.4%, 1.8%, and 0.3%, respectively, in the six sampling plots. The effectiveness of the improved method is verified, while the higher the degree of vegetation mixing in the region the better the extraction effect of the improved algorithm.

Robust phase demodulation algorithm for single-frame interferogram

This paper presents a robust phase demodulation algorithm (PDA) for single-frame interferogram based on preprocessing and watershed segmentation transformation. Compared to existing methods, the PDA can be applied to closed fringe or non-closed fringe patterns with low number of fringes indiscriminately. The proposed PDA includes two main parts, one is the preprocessing stage based on the empirical mode decomposition and Hilbert spiral transformation for background and noise removal, as well as normalization, and the other one is the correction of the wrapped phase obtained by arccosine through the watershed segmentation transformation. Through numerical simulations, the robustness of PDA to high noise and different kinds of fringe patterns is proved. The experimental evaluation of the proposed algorithm focuses on surface shape measurement of optical components, and the comparative analysis of the first 36 Zernike coefficients demonstrates that the PDA has high accuracy and the phase is good accordance with that obtained by phase-shifting interferometry.

DERIVATION OF BUILDING STRUCTURES FROM NOISY DIGITAL SURFACE MODELS

Abstract. In this work we present a novel approach for segementation of a noisy DSM to building structures and other non-building structures – normally trees – and the modeling of them. Mostly Digital Surface Models (DSMs) from only a few aerial images or only from one pair of satellite images tend to be very noisy and lack good quality especially in shadow areas. Since actual methods for deriving roofs rely on a valid height information by joining areas of same slope to a roof-plane these fail regularly with such noisy DSMs. In our presented approach we use a slope map of the DSM only to detect flat regions. Since those regions on top of roofs are mostly good illuminated we can derive the ridges of roofs and flat roofs and also ground areas. All narrow, flat, elevated areas are ridges and may occur on roofs or on trees. After connecting ridges in ridge-directions there remain two types of ridges: long, straight ridges of roofs and mixed short ridges in many directions for the trees. Fitting symmetric planes through the roof-ridge-lines gives finally the roof-planes reducing the effects of noise on shadowed parts of the roof. Taking the other tree-ridges as seeds for a watershed transformation will give the trees. Finally the proposed method is applied to a noisy DSM and the results will be discussed.

Trials and tribulations: Developing an artificial intelligence for screening malaria parasite from peripheral blood smears

BackgroundDetection of malaria parasite from blood smears remains the gold standard for confirmation of diagnosis. Screening blood smears for malaria parasite has a sensitivity of 75 %, and requires intensive training of the laboratory technician. In the present study, we have attempted to develop an artificial intelligence to automate the process of malaria parasite detection. MethodsWe acquired 352 images of Leishman–Giemsa-stained peripheral blood smears, containing either normal red blood cells (RBCs) or parasitised RBCs. With a trial and error approach, we developed five deep learning models: (A) Naive deep convolutional neural network (DCNN) for trophozoites, (B) Modified Inception V3 pretrained neural network (C) Combination of model A and B, (D) Segmentation of cells from the images through Watershed Transform and naive tri-class DCNN (normal RBCs, parasitised RBCs, WBC/platelets), and (E) A naive DCNN model to detect ring forms. The images were randomly split into training and test sets and training was imparted on all the models. After completion of training, performance of each model was assessed on the test set. ResultsOverall, the best combination of sensitivity and specificity was seen in model D (85 % and 94 %, respectively) in detecting parasites; in addition to trophozoites, model D could also detect ring forms. The performance of model A, B & C suffered from lack of either sensitivity or specificity. ConclusionThe present study represents the first step towards development of a complete module for screening malaria parasites from automated microphotography/whole slide images.

Analysis of Early Stroke Diagnosis Based on Brain Magnetic Resonance Imaging using Machine Learning

Stroke causes paralysis resulting from a hemorrhage in the brain or blockage of blood flow to the brain. It is third leading cause of death in Malaysia, with at least 32 deaths per day, and poses a major challenge to Malaysia's health services. A recent study showed that he could save a patient's life if he received treatment within six hours of a stroke. Unfortunately, Malaysia is facing a shortage of neuroradiologists, hampering efforts to treat its growing number of stroke patients. In this research, used Magnetic Resonance Imaging which is better compare to CT scan and CBCT because MRI will produce more detailed images of soft tissues, ligaments and organs. So that, advanced imaging using magnetic resonance imaging (MRI) has gained more attention than conventional angiography in the diagnosis of acute stroke due to its high spatial resolution and fast scan times. Traditionally, diagnosis was made manually by neuroradiologists during a highly subjective and time-consuming task. Detecting collaterals from MRI images is a challenging task due to the presence of noise and artifacts, small size, and heterogeneous structure of vessels. By the way, this paper is mainly about the early diagnosis of stroke based on brain magnetic resonance imaging using machine learning. Based on the results, can see that the Fuzzy c-means (FCM) and Watershed Transformation (WT) segmentation of brain infarcts which are original image, output of guided filter, gradient magnitude image, output of watershed transform and final detected infarct with morphological operation.

Fuzzy C‐means clustering algorithm based on superpixel merging and multi‐feature adaptive fusion measurement

AbstractThe fuzzy C‐means clustering (FCM) algorithm is widely used in greyscale and colour image segmentation, especially in real colour images. However, in the process of interested regions extraction, it performs barely satisfactory due to the use of single distance metric in traditional FCM. In order to address the issue, a fuzzy C‐means clustering algorithm based on superpixel merging and multi‐feature adaptive fusion measurement (FCM‐SM) are proposed. First of all, since superpixel image adapts well to irregular image boundaries, the FCM algorithm is elevated to the superpixel level. Moreover, the superpixel merging module based on colour similarity is introduced to improve the traditional watershed transformation by incorporating more gradient and intensity information. Finally, the multi‐feature adaptive fusion measurement is designed in the clustering process to incorporate multiple sources of spatial information, which comprehensively measures the information from different fields to enhance the segmentation capability further. Experiments performed on Berkeley Segmentation Dataset and Brain Tumor Segmentation Dataset demonstrate that the proposed algorithm provides better segmentation results than benchmarks during the object extraction.

Evolution of the landscape ecological pattern in arid riparian zones based on the perspective of watershed river-groundwater transformation

In arid and semiarid areas with scarce precipitation, strong evaporation, and fragile ecosystems, rivers and groundwater interact frequently. Dynamic hydrological changes and solute transport jointly determine the evolution of the landscape in riparian zones, resulting in the formation of a landscape ecological pattern in arid areas that is unique and includes a system of rivers and lakes. In recent decades, unsustainable and mono-purpose grey infrastructure has been implemented in many arid areas, and the effect of river-groundwater interactions on the landscape ecological pattern in riparian zones has often been neglected. These processes have triggered variations in or even the reversal of landscape ecological patterns in these arid areas. This study selected the arid and semiarid basin in northwestern China as the study area. Multiple methods were used to study the evolution of the landscape ecological pattern in the riparian zones driven by typical hydrological processes from the perspective of river-groundwater transformation. Three results were obtained from the analysis: (1) The driving force of the evolution was identified, and the spatial geological heterogeneity, meteorological and hydrological conditions, and river-groundwater interactions played dominant roles in the evolution of the landscape ecological pattern in the riparian zones. (2) Human activities and climate change controlled the direction and intensity of landscape evolution. (3) Pathways for maintaining a healthy landscape ecological pattern in riparian zones and suggestions for identifying the key issues that should be studied in the future were proposed. This paper primarily contributes to a better understanding of landscape ecological patterns in the riparian zone in arid areas from the watershed river-groundwater transformation perspective, thereby promoting integrated watershed management.

Deep learning enhanced Watershed for microstructural analysis using a boundary class semantic segmentation

The mechanical properties of the materials are determined by the size and morphology of fine microscopic features. Quantitative microstructural analysis is a key factor to establish the correlation between the mechanical properties and the thermomechanical treatment under which material condition has been achieved. As such, microstructural analysis is a very important and complex task within the manufacturing sector. Published standards are used for metallographic analysis but typically involve extensive manual interpretation of grain boundaries, resulting in measurements that are slow to produce, difficult to repeat and highly subjective. Computer vision and the evolution of artificial intelligence in the past decade can offer solutions to such problems. Deep learning and digital image processing techniques allow digital microstructural analysis to be automated using a fast and repeatable method. This paper proposes a novel boundary class semantic segmentation approach (BCSS) to identify each phase of the microstructure and additionally estimate the location of the grain boundaries. The BCSS is then combined with more traditional segmentation techniques based on the Watershed Transform to improve the identification and measurement of each feature within the microstructure using a new, hybrid automated digital microstructure analysis approach. The new method is validated on a published dataset of two-phase titanium alloy microstructure pictures captured using a scanning electron microscope. Measurements match the level of accuracy of accepted manual standards, and the method is demonstrated to be more reliable than other automated approaches. The influence of the subjective nature of manual labelling, required to train the proposed network, is also evaluated.

Fault Prediction in Electrical Assets Using Infrared Thermography

An infrared thermography technique is discussed in this paper to locate and diagnose the electrical equipment issue. Overheating is among major problems for an electrical device and entire power system. These overheated hotspots must be detected so that the fault can be predicted before it occurs. We used a thermal image of an electrical device taken from a infrared camera to apply our method to detect the hot region and find the maximum temperature and minimum temperature in the image. Image de-noising is done with the help of morphological reconstruction followed by Watershed transformation for image segmentation to identify and separate the region having high pixel intensity as we know that area with high pixel intensity will also have high temperature. Our method achieved near to exact maximum and minimum temperature of electrical device.

Analysis of density-dependent bead and cell structure of expanded polypropylene bead foams from X-ray computed tomography of different resolution

Closed-cell bead foams with their hierarchical geometrical structure are a challenge for statistical reconstruction and finite element modelling. For the purpose of providing the fundamental micro- and meso-structural descriptors - wall thickness, cell as well as bead volume and sphericity - of expanded polypropylene bead foams of different density, 3D-images from X-ray computed tomography are analyzed. A detailed description of development and application of an image analysis methodology for the determination of feature distributions from CT-scans of different level of detail is provided. The methods are based on off-the-shelf algorithms provided by the open-source package distribution FIJI. It should be highlighted, that beside essential methods such as thresholding, euclidean distance and watershed transformation here the Trainable WEKA segmentation is applied for separating material phases in the images. Although the methods elaborated are generally very case sensitive, the reader benefits from the validation strategies applied, so that development of individual methods into the direction of reliability, repeatability and automation is supported.

A Novel Segmentation Recognition Algorithm of Agaricus bisporus Based on Morphology and Iterative Marker-Controlled Watershed Transform

Accurate recognition of Agaricus bisporus is a prerequisite for precise automatic harvesting in a factory environment. Aimed at segmenting mushrooms adhering together from the complex background, this paper proposes a watershed-based segmentation recognition algorithm for A. bisporus. First, the foreground of A. bisporus is extracted via Otsu threshold segmentation and morphological operations. Then, a preliminary segmentation algorithm and a novel iterative marker generation method are proposed to prepare watershed markers. On this basis, a marker-controlled watershed algorithm is adopted to segment and recognize A. bisporus individuals. All the algorithms are implemented based on OpenCV (Open Source Computer Vision) libraries. Tests on images of A. bisporus collected at the cultivation bed show that the average correct recognition rate of the proposed algorithm is 95.7%, the average diameter measurement error is 1.15%, and the average coordinate deviation rate is 1.43%. The average processing time is 705.7 ms per single image, satisfying the real-time constraints based on 1 image/s. The proposed algorithm performed better than the current Circle Hough Transform (OpenCV’s implementation). It is convenient and easy to operate, providing a sound basis for subsequent research on mechanized harvesting equipment for A. bisporus.

Research on spatiotemporal characteristics and trends of urban housing sites based on remote sensing analysis technology

In order to improve the analysis effect of the spatiotemporal characteristics of urban homesteads, this paper combines remote sensing analysis technology to study and analyze the spatiotemporal characteristics and trends of urban homesteads, segment the spatiotemporal characteristics of urban homesteads, and use the watershed transformation based on multi-scale morphological reconstruction to generate superpixels. The watershed transformation based on multi-scale morphological gradient reconstruction can obtain superpixel segmentation with good local information with less time consumption and a smaller number of superpixels. The research shows that the spatiotemporal feature analysis model of urban homestead based on remote sensing analysis technology proposed in this paper can effectively improve the analysis effect of urban homestead spatiotemporal feature, and can predict the spatiotemporal feature trend of urban homestead on this basis.

Fast Watershed Segmentation for Breast Cancer Detection

Within engineering and computer specializations, image processing is the most important study topic. It is one of today's fastest-growing technologies, with applications in a variety of biological sectors, including cancer sickness. According to the latest data from throughout the world, breast cancer is the most lethal of all cancer kinds. It is the most frequent cancer in women and the second leading cause of cancer mortality in women. In this study, we advocate using a watershed transformation to create a fast segmentation technique. This allows for the blending of updated information about picture objects, extending the partitioning of the dividing waterline and therefore the standard watershed technique. The method requires a mechanism to express the test picture in terms of the amount of change around every given pixel before it can begin the watershed modification. Each pixel in the greyscale representation of the original picture is subjected to the Sobel operator. According to the form, the tumors identified are round or semicircular and the light of the tumor dims as we travel away from its core. The complement for this previous data may be seen as a local minimum that necessitated the start of the watershed process. As a result, each tumor picture may be represented as a lake, with the center in the complement tumor picture being the least value. The identification of tumor percentage gets more reliable after using the approach. As a consequence, our computer-aided diagnostic method for mammographic breast cancer detection has improved significantly thanks to the novel methodology. The method was written in MATLAB and tested on a Windows computer. The strategy was put to the test using photos from MIAS (Mammogram Image Analysis Society, UK), which offers a consistent categorization system for mammographic examinations. In this study, we advocate using a watershed transformation to create a rapid segmentation technique. This allows for the blending of data about picture objects, extending the partitioning of the dividing waterline and therefore the standard watershed technique.

Ventricular segmentation and modeling using topological watershed transformation and harmonic state model

This paper proposes an adapted ventricular segmentation method based on topological watershed transform. Segmentation will allow spatio-temporal modeling of trajectories of the different points belonging to the borders of the ventricle using a harmonic motion model that is able to describe such motion over the entire cardiac cycle. In addition, extraction of the adopted canonical state vector and the corresponding state equations guarantees an optimal efficacy and a gradual transition from order n to order n + 1. To validate the proposed approach, an intern-image base was used. Our results show a promising ability to discern whether subjects are healthy or pathological with an 80% success rate.

Sand Soil Image Processing Using the Watershed Transform and Otsu Thresholding Based on Gaussian Noise

Image processing technology is one of the technologies that can help facilitate and speed up human work, especially in the process of determining the grain size distribution of soil in a civil building plan. Its utilization has been widely used to study, analyze, and understand the structure and framework of the soil. Image analysis is carried out as an initial or fundamental step in image processing to discover and comprehend information. With so many image segmentation methods, it is necessary to conduct research to determine which method is best for sandy soil image segmentation based on one of the image segmentation quality criteria, namely gaussian image noise. By testing the watershed transform method and the Otsu thresholding method as two of the area-based methods that are considered suitable for segmenting sandy soil images before and after distorted Gaussian noise based on the calculation of the mean square error (MSE) value.The results showed that the watershed transform method is better for segmenting sandy soil images when compared to the Otsu thresholding method. This is indicated by the average squared error (mse) of 3.08 for the watershed transform method and 4.09 for the Otsu Thresholding method. In addition to the comparison of quality tests of sandy soil based on gaussian noise with standard deviation values of normal distribution and noise intensities of 10, 20, and 30, it proves that the watershed transform method is still better at segmenting noise-distorted sandy soil images than the Otsu thresholding method. However, in terms of processing time, the Otsu Thresholding method is faster or better than the Watershed Transform method. of the results or conclusions brief. There are no citations, tables or figures in abstract.

LUNG SCANS SEGMENTATION USING MARKER-CONTROLLED WATERSHED TRANSFORMATION

Image segmentation is the process of partitioning a digital image into multiple segments knows as set of pixels. Image segmentation is typically used to locate objects and boundaries (lines, curves, etc.) in images. It is the process of assigning a label to every pixel in an image such that pixels with the same label share certain visual characteristics. The Watershed Transform consists of an image partitioning into its constitutive regions. This transform is easily adapted to be used in different types of images and it allows distinguishing complex objects. The marker watershed transformation is an improvement one on the basis of the watershed transform, which is set a tag in the image. The tag can be a point, a line or an area, the importance is its location instead of the marked shape. Each tag represents the image of a final partition and selecting a tag is a key factor in the decision segmentation.

Detection of Mammogram Using Improved Watershed Segmentation Algorithm and Classifying with Feed Forward Neural Network (FNN)

Segmentation of breast tumors with more accuracy using computerized methods is essential for breast cancer monitoring and quantification. Both segmentation and classification of breast tumors using a fully automated or Computer-Aided Diagnosis system poses various problems in terms of imaging properties. In this work, a new hybrid algorithm is proposed for segmentation with a two-step process. Initially, a watershed transformation is applied to separate all basins based on pixel density variation from the mass present in tumors, since it has been quite booming in the presence of tumors in all circumstances. Though this is very perceptive to tiny fluctuations in the size of the image, large numbers of areas are produced unacceptably, and the boundaries after segmentations are also quite hard. The second level set is an effective method of segmenting all types of medical images because; it easily flows with, cavities, folds, splits, and merges. To make the recognition step easier and more accurate, the result of segmentation is considered the beginning position of the curve, and the same will be used at the next step of the level set. This produces a closed, smooth, and accurately placed contour or surface. As a result, the present research uses watershed segmentation to isolate tumor regions and performs classification using Feed Forward Neural Network (FNN) to extract features for classification. Experimental results are evaluated based on performance and quality analysis. In the classification process, the study obtained an accuracy rate of 91.2% in the learning model and 71.8% in a testing model.

Application Of Dynamic Segmentation In Stroke Detection Software With ANN

One way to find out whether there is a stroke is to do a CT scan . But the results of the examination with a new CT scan can be obtained in quite a long time. In addition, sometimes there are differences of opinion between doctors and radiologists regarding what is seen from the results of the examination. This research was conducted to produce a software that can later be integrated with the existing system on the CT Scan tool so that it can immediately be known whether or not stroke is present from the CT Scan results. In this study, a dynamic image segmentation method is implemented, namely the watershed transformation method which will later produce regions as a feature for the stroke detection process carried out with the backpropagation algorithm. From experiments conducted on CT scan images of the brain, this method can detect stroke well. The results obtained are 100% for training data and 90% for test data.

Optimized hybrid machine learning approach for smartphone based diabetic retinopathy detection.

Diabetic Retinopathy (DR) is defined as the Diabetes Mellitus difficulty that harms the blood vessels in the retina. It is also known as a silent disease and cause mild vision issues or no symptoms. In order to enhance the chances of effective treatment, yearly eye tests are vital for premature discovery. Hence, it uses fundus cameras for capturing retinal images, but due to its size and cost, it is a troublesome for extensive screening. Therefore, the smartphones are utilized for scheming low-power, small-sized, and reasonable retinal imaging schemes to activate automated DR detection and DR screening. In this article, the new DIY (do it yourself) smartphone enabled camera is used for smartphone based DR detection. Initially, the preprocessing like green channel transformation and CLAHE (Contrast Limited Adaptive Histogram Equalization) are performed. Further, the segmentation process starts with optic disc segmentation by WT (watershed transform) and abnormality segmentation (Exudates, microaneurysms, haemorrhages, and IRMA) by Triplet half band filter bank (THFB). Then the different features are extracted by Haralick and ADTCWT (Anisotropic Dual Tree Complex Wavelet Transform) methods. Using life choice-based optimizer (LCBO) algorithm, the optimal features are chosen from the mined features. Then the selected features are applied to the optimized hybrid ML (machine learning) classifier with the combination of NN and DCNN (Deep Convolutional Neural Network) in which the SSD (Social Ski-Driver) is utilized for the best weight values of hybrid classifier to categorize the severity level as mild DR, severe DR, normal, moderate DR, and Proliferative DR. The proposed work is simulated in python environment and to test the efficiency of the proposed scheme the datasets like APTOS-2019-Blindness-Detection, and EyePacs are used. The model has been evaluated using different performance metrics. The simulation results verified that the suggested scheme is provides well accuracy for each dataset than other current approaches.