Unsupervised Clustering Framework Research Articles

Evaluating the Patterns of Maize Development in the Hetao Irrigation Region Using the Sentinel-1 GRD SAR Bipolar Descriptor

Assessing maize yield is critical, as it is directly influenced by the crop’s growth conditions. Therefore, real-time monitoring of maize growth is necessary. Regular monitoring of maize growth indicators is essential for optimizing irrigation management and evaluating agricultural yield. However, quantifying the physical aspects of regional crop development using time-series data is a challenging task. This research was conducted at the Dengkou Experimental Station in the Hetao irrigation area, Northwest China, to develop a monitoring tool for regional maize growth parameters. The tool aimed to establish a correlation between satellite-based physical data and actual crop growth on the ground. This study utilized dual-polarization Sentinel-1A GRD SAR data, accessible via the Google Earth Engine (GEE) cloud platform. Three polarization descriptors were introduced: θc (pseudo-scattering type parameter), Hc (pseudo-scattering entropy parameter), and mc (co-polar purity parameter). Using an unsupervised clustering framework, the maize-growing area was classified into several scattering mechanism groups, and the growth characteristics of the maize crop were analyzed. The results showed that throughout the maize development cycle, the parameters θc, Hc, and mc varied within the ranges of 26.82° to 42.13°, 0.48 to 0.89, and 0.32 to 0.85, respectively. During the leaf development stage, approximately 80% of the maize sampling points were concentrated in the low-to-moderate entropy scattering zone. As the plants reached the big trumpet stage, the entire cluster shifted to the high-entropy vegetation scattering zone. Finally, at maturity, over 60% of the sampling points were located in the high-entropy distribution scattering zone. This study presents an advanced analytical tool for crop management and yield estimation by utilizing precise and high-resolution spatial and temporal data on crop growth dynamics. The tool enhances the accuracy of crop growth management across different spatial and temporal conditions.

Rarity: discovering rare cell populations from single-cell imaging data.

Cell type identification plays an important role in the analysis and interpretation of single-cell data and can be carried out via supervised or unsupervised clustering approaches. Supervised methods are best suited where we can list all cell types and their respective marker genes a priori, while unsupervised clustering algorithms look for groups of cells with similar expression properties. This property permits the identification of both known and unknown cell populations, making unsupervised methods suitable for discovery. Success is dependent on the relative strength of the expression signature of each group as well as the number of cells. Rare cell types therefore present a particular challenge that is magnified when they are defined by differentially expressing a small number of genes. Typical unsupervised approaches fail to identify such rare subpopulations, and these cells tend to be absorbed into more prevalent cell types. In order to balance these competing demands, we have developed a novel statistical framework for unsupervised clustering, named Rarity, that enables the discovery process for rare cell types to be more robust, consistent, and interpretable. We achieve this by devising a novel clustering method based on a Bayesian latent variable model in which we assign cells to inferred latent binary on/off expression profiles. This lets us achieve increased sensitivity to rare cell populations while also allowing us to control and interpret potential false positive discoveries. We systematically study the challenges associated with rare cell type identification and demonstrate the utility of Rarity on various IMC datasets. Implementation of Rarity together with examples is available from the Github repository (https://github.com/kasparmartens/rarity).

Vegetation descriptors from Sentinel-1 SAR data for crop growth monitoring

Synthetic aperture radar (SAR) remote sensing technology has the advantage of all-weather observation and can acquire time-series images with crop growth period, which has great potential for applications such as crop phenology analysis. However, available studies primarily focus on conducting statistical and crop growth analyses based on the polarization or backscatter intensities of SAR images, and the exploration of polarization scattering information in SAR images is not sufficient. To comprehensively reflect the polarization characteristics and scattering mechanisms of crop at different growth stages, we propose a new method for extracting vegetation descriptors from Sentinel-1 dual-polarimetric SAR data. The method combines the backscattering intensity and polarization decomposition information to construct a normalized index q, which is used to generate three vegetation descriptors: the co-pol purity parameter (mcp), the pseudo-scattering angle (θcp), and the pseudo-scattering entropy (Hcp). Further, a novel unsupervised clustering framework, founded on Hcp and θcp, has been proposed. This framework establishes six zones (named as Z1 to Z6) representing distinct physical scattering mechanisms, and by statistically sampling point data, it can determine the growth stage of the crops For validating the performance of the proposed vegetation descriptors and clustering framework, we conducted a three-year experiment using four crops from two publicly available datasets, namely wheat and canola from the Carman in Canada (Test site-1), corn and soybeans from Iowa in the United States (Test site-2). The experimental results indicate that mcp,θcp, and Hcp exhibit regular changes at different growth stages of crops from planting to maturity, with mcp and θcp gradually decreasing while Hcp gradually increasing. Within the entire phenology window, θcp changes by approximately 42°, while both mcp and θcp varies by about 0.9, and the sampling points shift from the Z2 to the Z5 zone. The vegetation descriptors are highly sensitive to the growth status of crops, and the clustering framework can also effectively respond to different growth stages of vegetation. Furthermore, the vegetation descriptors and clustering framework proposed in this study have the potential for extended application to different crop types and other polarimetric SAR data sources.

Detecting metro service disruptions via large-scale vehicle location data

Urban metro systems are often affected by disruptions such as infrastructure malfunctions, rolling stock breakdowns and accidents. The crucial prerequisite of any disruption analytics is to have accurate information about the location, occurrence time, duration and propagation of disruptions. To pursue this goal, we detect the abnormal deviations in trains’ headway relative to their regular services by using Gaussian mixture models. Our method is a unique contribution in the sense that it proposes a novel, probabilistic, unsupervised clustering framework and it can effectively detect any type of service interruptions, including minor delays of just a few minutes. In contrast to traditional manual inspections and other detection methods based on social media data or smart card data, which suffer from human errors, limited monitoring coverage, and potential bias, our approach uses information on train trajectories derived from automated vehicle location (train movement) data. As an important research output, this paper delivers innovative analyses of the propagation progress of disruptions along metro lines, which enables us to distinguish primary and secondary disruptions as well as effective recovery interventions performed by operators.

Unsupervised Driving Style Analysis Based on Driving Maneuver Intensity

This paper proposes a novel unsupervised clustering framework to identify driving style not in terms of the discrete features of driving behavior data, but rather the time-varying patterns of driving maneuver intensity. This framework can describe the dynamic decision-making process of driving behavior and the continuity of driving data, and driving maneuver intensity is the basic of this paper. Therefore, detection, feature analysis and clustered on intensity of driving maneuvers are carried out using a threshold-based approach, hierarchical feature extraction, and k-means clustering. Then, to analyze fine-grained driving style, dynamic time windows are determined according to road alignment. In dynamic time windows, this paper constructs time-varying patterns based on driving maneuver intensity, which consider the intensity and frequency of driving behavior and preserve the time-varying characteristics of time-series data. However, not all dynamic time windows are equal in maneuvers’ duration and number, which means the time-varying patterns of driving maneuver intensity are curves with various lengths. So that, for clustering time-varying patterns, this paper proposes a novel curve clustering algorithm named Similarity-Based Clustering with Dynamic Time Warping (SBC-DTW) that can cluster curves with various lengths. The empirical results based on real driving data demonstrate that the proposed framework can classify driving style more accurately than the classical method. Moreover, according to this framework, we can have an in-depth understanding of dynamic driving behavior and the composition of drivers’ long-term driving styles.

Dual-polarimetric descriptors from Sentinel-1 GRD SAR data for crop growth assessment

Accurate and high-resolution spatio-temporal information about crop phenology obtained from Synthetic Aperture Radar (SAR) data is an essential component for crop management and yield estimation at a local scale. Crop growth monitoring studies seldom exploit complete polarimetric information contained in dual-pol GRD SAR data. In this study, we propose three polarimetric descriptors: the pseudo scattering-type parameter (θc), the pseudo scattering entropy parameter (Hc), and the co-pol purity parameter (mc) from dual-pol S1 GRD SAR data. We also introduce a novel unsupervised clustering framework using Hc and θc with six clustering zones to represent various scattering mechanisms. We implemented the proposed algorithm on the cloud-based Google Earth Engine (GEE) platform for Sentinel-1 SAR data. We have shown the sensitivity of these descriptors over a time series of data for wheat and canola crops at a test site in Canada. From the leaf development stage to the flowering stage for both crops, the pseudo scattering-type parameter θc changes by approximately 17°. Moreover, within the entire phenology window, both mc and Hc varies by about 0.6. The effectiveness of θc and Hc to cluster the phenological stages for the two crops is also evident from the clustering plot. During the leaf development stage, about 90% of the sampling points were clustered into the low to medium entropy scattering zone for both the crops. Throughout the flowering stage, the entire cluster shifted into the high entropy vegetation scattering zone. Finally, during the ripening stage, the clusters of sample points were split between the high entropy vegetation scattering zone and the high entropy distributed scattering zone, with >55% of the sampling points in the high entropy distributed scattering zone. This innovative clustering framework will facilitate the operational use of S1 GRD SAR data for agricultural applications.

Extracting knowledge of NCI research directions from funding data using language processing.

e13547 Background: In fiscal year (FY) 2019, 42% of the $6 billion NCI budget went towards nearly 5,000 research project grants, of which about 60% are R01 type. Given the enormity of allocated resources, there is a need for the scientific community to have a more rigorous understanding of the cancer research landscape. While the NCI Budget Fact Book publishes statistics based on pre-designated codings, it is unclear if this method yields the best representation of fields within oncology. Open questions include: how many distinguishable areas of cancer research are being funded? Are there differences in growth rate, publication rate and geographic distribution among those areas? Addressing these questions in a systematic manner is a well-suited problem for unsupervised machine learning. Methods: We analyzed 55,362 R-type grants from FY 2010-2021 (up to 1/31/21) from NIH ExPORTER. Preprocessing was done on ‘Project Terms’ to weight their importance using TF*IDF vectorization and principal component analysis. We used minibatch K-means clustering repeated 100 times, with the best iteration chosen by Calinski-Harabasz clustering quality score. Over 100 repetitions, the Adjusted Rand Index was 0.9907±0.0037 (mean ±standard deviation), indicating robustness to initial conditions of K-means. For publication rate analysis, FY 2020 and 2021 grants were excluded, and 2021 grants were excluded for trajectory analysis. Optimal cluster number was determined based on a combination of inertia, Calinski-Halabasz, and silhouette scores. Results: We found the optimal number of 24 clusters to best represent separation of the R-type grant research directions. These 24 clusters clearly represent topics such as immunotherapy, cohort risk-factor studies, and imaging. Notable trends include increased funding of immunotherapy and targeted inhibitor clusters averaging +$9.9M and +$9.2M growth per year respectively over 10 years, and decreased funding of pathway regulation and genetics clusters averaging -$7.8M and -$5.7M per year. These examples suggest a broader trend that funding is shifting from basic to translational science. Further analysis shows that the targeted inhibitor cluster is most geographically skewed, with 30% of grants going to institutions in just three cities. The number of journal articles (per grant) also shows a bias, with development/training and genetics clusters having publication rates of 32.6 and 20.8 and randomized control trials having a publication rate of 6.3. The average publication rate of NCI was 14.8. Conclusions: Using a novel framework for unsupervised clustering of NCI grant key-phrases, we can organize research projects more holistically than keyword searching, and more efficiently than manual categorization. Our model identifies growing and shrinking areas of research and points out biases in location and publication rate across these areas.

Dynamic cattle behavioural classification using supervised ensemble classifiers

In this paper various supervised machine learning techniques were applied to classify cattle behaviour patterns recorded using collar systems with 3-axis accelerometer and magnetometer, fitted to individual dairy cows to infer their physical behaviours. Cattle collar data was collected at the Tasmanian Institute of Agriculture (TIA) Dairy Research Facility in Tasmania. In the first stage of analysis a novel hybrid unsupervised clustering framework, comprised of probabilistic principal component analysis, Fuzzy C Means, and Self Organizing Map network algorithms was developed and used to study the natural structure of the sensor data. Findings from this unsupervised clustering were used to guide the next stage of supervised machine learning. Five major behaviour classes, namely, Grazing, Ruminating, Resting, Walking, and other behaviour were identified for the classification trials. An ensemble of classifiers approach was used to learn models of cow behaviour using sensor data and ground truth behaviour observations acquired from the field. Ensemble classification using bagging, Random Subspace and AdaBoost methods along with conventional supervised classification methods, namely, Binary Tree, Linear Discriminant Analysis classifier, Naive Bayes classifier, k-Nearest Neighbour classifier, and Adaptive Neuro Fuzzy Inference System classifier were compared. The highest average correct classification accuracy of 96% was achieved using the bagging ensemble classification with Tree learner, which had 97% sensitivity, 89% specificity, 89% F1 score and 9% false discovery rate. This study has shown that cattle behaviours can be classified with a high accuracy using supervised machine learning technique. As dairy and beef systems become more intensive, the ability to identify the changes in the behaviours of individual livestock becomes increasingly difficult. Accurate behavioural monitoring through sensors provides a significant potential in providing a mechanism for the early detection and quantitative assessment of animal health issues such a lameness, informing key management events such as the identification of oestrus, or informing changes in supplementary feeding requirements.

Unsupervised white matter fiber clustering and tract probability map generation: Applications of a Gaussian process framework for white matter fibers

With the increasing importance of fiber tracking in diffusion tensor images for clinical needs, there has been a growing demand for an objective mathematical framework to perform quantitative analysis of white matter fiber bundles incorporating their underlying physical significance.This article presents such a novel mathematical framework that facilitates mathematical operations between tracts using an inner product between fibres. Such inner product operation, based on Gaussian processes, spans a metric space. This metric facilitates combination of fiber tracts, rendering operations like tract membership to a bundle or bundle similarity simple. Based on this framework, we have designed an automated unsupervised atlas-based clustering method that does not require manual initialization nor an a priori knowledge of the number of clusters. Quantitative analysis can now be performed on the clustered tract volumes across subjects, thereby avoiding the need for point parameterization of these fibers, or the use of medial or envelope representations as in previous work. Experiments on synthetic data demonstrate the mathematical operations. Subsequently, the applicability of the unsupervised clustering framework has been demonstrated on a 21-subject dataset.

Visual MRI: Merging information visualization and non-parametric clustering techniques for MRI dataset analysis

This paper presents Visual MRI, an innovative tool for the magnetic resonance imaging (MRI) analysis of tumoral tissues. The main goal of the analysis is to separate each magnetic resonance image in meaningful clusters, highlighting zones which are more probably related with the cancer evolution. Such non-invasive analysis serves to address novel cancer treatments, resulting in a less destabilizing and more effective type of therapy than the chemotherapy-based ones. The advancements brought by Visual MRI are two: first, it is an integration of effective information visualization (IV) techniques into a clustering framework, which separates each MRI image in a set of informative clusters; the second improvement relies in the clustering framework itself, which is derived from a recently re-discovered non-parametric grouping strategy, i.e., the mean shift. The proposed methodology merges visualization methods and data mining techniques, providing a computational framework that allows the physician to move effectively from the MRI image to the images displaying the derived parameter space. An unsupervised non-parametric clustering algorithm, derived from the mean shift paradigm, and called MRI-mean shift, is the novel data mining technique proposed here. The main underlying idea of such approach is that the parameter space is regarded as an empirical probability density function to estimate: the possible separate modes and their attraction basins represent separated clusters. The mean shift algorithm needs sensibility threshold values to be set, which could lead to highly different segmentation results. Usually, these values are set by hands. Here, with the MRI-mean shift algorithm, we propose a strategy based on a structured optimality criterion which faces effectively this issue, resulting in a completely unsupervised clustering framework. A linked brushing visualization technique is then used for representing clusters on the parameter space and on the MRI image, where physicians can observe further anatomical details. In order to allow the physician to easily use all the analysis and visualization tools, a visual interface has been designed and implemented, resulting in a computational framework susceptible of evaluation and testing by physicians. Visual MRI has been adopted by physicians in a real clinical research setting. To describe the main features of the system, some examples of usage on real cases are shown, following step by step all the actions scientists can do on an MRI image. To assess the contribution of Visual MRI given to the research setting, a validation of the clustering results in a medical sense has been carried out. From a general point of view, the two main objectives reached in this paper are: (1) merging information visualization and data mining approaches to support clinical research and (2) proposing an effective and fully automated clustering technique. More particularly, a new application for MRI data analysis, named Visual MRI, is proposed, aiming at improving the support of medical researchers in the context of cancer therapy; moreover, a non-parametric technique for cluster analysis, named MRI-mean shift, has been drawn. The results show the effectiveness and the efficacy of the proposed application.