Spatial Pattern Mining Research Articles

Trends in Hotspots of Alien Plant Invasion in Kachchh Biosphere Reserve, India Using Spatial Pattern Mining Tool

Trends in Hotspots of Alien Plant Invasion in Kachchh Biosphere Reserve, India Using Spatial Pattern Mining Tool

A Spatial Fuzzy Co-Location Pattern Mining Method Based on Interval Type-2 Fuzzy Sets

The goal of spatial co-location pattern mining is to find subsets of spatial features whose instances are often neighbors in a geographical space. In many practical cases, instances of spatial features contain not only spatial location information but also attribute information. Although there have been several studies that use type-1 fuzzy membership functions to mine spatial fuzzy co-location patterns, there is great uncertainty associated with such membership functions. To address this problem, we propose a spatial fuzzy co-location pattern mining method based on interval type-2 fuzzy sets. First, we collect the interval evaluation values of the interval data of attribute information from experts to form granular data. Next, the original type-1 fuzzy membership function is extended to a granular type-2 fuzzy membership function based on elliptic curves. We use a gradual method to adjust the parameters of the fuzzy membership function so that its footprint of uncertainty satisfies both the connectivity and the given confidence. Based on this granular type-2 fuzzy membership function, we fuzzify the attribute information of instances and define the concepts of fuzzy features and fuzzy co-location patterns. A fuzzy co-location pattern mining algorithm based on spatial cliques is then proposed, termed the FCPM-Clique algorithm. In order to improve the efficiency of the algorithm, we propose two pruning strategies. In addition, we extend two classical spatial pattern mining algorithms, the Join-based algorithm and the Joinless algorithm, to mine fuzzy co-location patterns based on interval type-2 fuzzy sets. Many experiments on synthetic and real-world datasets are conducted, the performance of the three algorithms is compared, and the effectiveness and efficiency of our proposed FCPM-Clique algorithm is demonstrated.

Analysing the trends in annual forest loss hotspots in the regional landscape of Eastern Ghats, India

Tropical deforestation is one of the widely recognised human-induced factors responsible for climate change. A clear understanding of the trends in forest cover loss hotspots provide baseline information for sustainable forest management. This study has utilised Landsat data and a novel spatial pattern mining tool to identify trends in forest cover loss from 2009 to 2020 in the Araku range of Eastern Ghats, India. This study focuses on identifying the annual rate of deforestation and significant trends in forest cover loss through statistical techniques. Analysis indicates no significant trend in forest loss (P = 0.7 and the trend statistic was Z = 0.27; S = 5) during the study period. Disturbances in the forest cover are mainly attributed to shifting cultivation in the study area. The Emerging hotspot model has found new and sporadic hotspots in the Araku range. However, there are no intensifying hotspots, historical hotspots and persistent hotspots indicate effective management in the majority of the study area. This study helps to identify and prioritise deforested areas for monitoring and preparing appropriate conservation plans and management strategies.

SSGRU: A novel hybrid stacked GRU-based traffic volume prediction approach in a road network

As a potential solution to relieve traffic congestions and help build a more safe traffic system, traffic flow prediction methods are given much attention in recent years. In previous studies, it can be found the machine learning (ML)-based methods are widely used in volume predictions of single roads. However, when applied in a more complicated road network, they usually show low efficiency and need to pay higher computing costs. To solve this problem, an innovative ML-based model, named Selected Stacked Gated Recurrent Units model (SSGRU), is proposed in this paper, which is mainly in allusion to road network traffic flow. There are mainly two parts in this model: one is used to do spatial pattern mining based on linear regression coefficients, and the other one includes a stacked gated recurrent unit (SGRU), which is essential for multi-road traffic flow prediction. As the basic unit, a simple tree structure is adopted to approximate the given road network. Particularly, we implemented our model into both suburban and urban traffic contexts, to prove its high adaptability. The whole evaluation process is based on seven different traffic volume data sets recorded at the 15-min interval, chosen from the England Highways. The results show that our model has higher accuracy than others when applied to a multi-road input infrastructure for all road scenarios.

An Enhanced Human Activity Prediction at Rainy and Night Times

Human activity prediction aims to recognize an unfinished activity with limited motion and appearance information. A generalized activity prediction framework was proposed for human activity prediction where Probabilistic Suffix Tree (PST) was introduced to model casual relationships between constituent actions. Then, each kind of activity in videos was predicted by modeling interactive object information through Spatial Pattern Mining (SPM). This framework mined the temporal sequence patterns. For efficient human activity prediction a Spatio-Temporal Frequent Object Mining (STOM) was proposed in which the spatial, size and motion correlation among objects information were collected along with the temporal information. After the collection of this information, the objects were identified by using Modified Histogram Of Gradient (MHOG) and then the objects were tracked by particle filter technique. The frequent action of detected objects were identified by using Frequent Pattern-growth (FP-growth) which predicted the infrequent action as abnormal human activity in videos. However, MHOG based Object Detection and Tracking-STOM (MHOGODT-STFOM) based human activity prediction is not more effective at night time and rainy time. So in this paper, Enhanced Object Detection and Tracking- STFOM (EODT- STFOM) and Removing Rain Streaks-EODT-STFOM (RSR-EODT-STFOM) are proposed for human activity prediction even at night time and rainy time. In EODT, a modified Contrast Model is used which combined the contrast information and local entropy information to detect object contents present in the current image frame. Then, the objects are tracked by Kalman filter. In RSR-EODT, the rain streaks in the images are removed based on the deep Convolutional Neural Network (CNN). Then the objects are detected and tracked by modified Contrast Model and Kalman filter respectively. After the object detection and object tracking by EODT and RSR-EODT, the frequent actions are obtained by applying STFOM. The frequent actions are considered as normal activities and the infrequent actions are considered as abnormal activities. Thus the proposed EODT-STFOM and RSR-EODT- STFOM methods predict the human activities even at night time and rainy time.

A deep learning approach to anomaly detection in geological carbon sequestration sites using pressure measurements

Carbon capture and storage (CCS) has been extensively investigated as a potential engineering measure to reduce anthropogenic carbon emission to the atmosphere. Real-time monitoring of the safety and integrity of carbon storage reservoirs is a critical aspect of any commercial-scale CCS deployment. Pressure-based sensing is cost effective, suitable for real-time monitoring, and scalable to large monitoring networks. However, questions remain on how to best harness intelligent information from the high-frequency pressure monitoring sensors to support real-time decisions. This work presents a deep-learning-based framework for analyzing and detecting anomalies in pressure data streams by using a convolutional long short-term memory (ConvLSTM) neural network model, which allows for the fusion of both static and dynamic reservoir data. In ConvLSTM, the convolutional neural network (CNN) is used for spatial pattern mining and the LSTM is used for temporal pattern recognition. The performance of the ConvLSTM model for real-time anomaly detection is demonstrated using a set of pressure monitoring data collected from Cranfield, Mississippi, an active enhanced-oil-recovery field. The anomaly detection model is trained using bottom-hole pressure data acquired from the base experiment (without leak event) and then tested on pressure data collected during a series of controlled CO2 release experiments (with artificially created leak events). Results show that the ConvLSTM neural network model successfully detected anomalies in the pressure time series obtained from the controlled release experiments. Inclusion of static information into the model further improves the robustness of ConvLSTM.

Pattern Deep Region Learning for Crack Detection in Thermography Diagnosis System

Eddy Current Pulsed Thermography is a crucial non-destructive testing technology which has a rapidly increasing range of applications for crack detection on metals. Although the unsupervised learning method has been widely adopted in thermal sequences processing, the research on supervised learning in crack detection remains unexplored. In this paper, we propose an end-to-end pattern, deep region learning structure to achieve precise crack detection and localization. The proposed structure integrates both time and spatial pattern mining for crack information with a deep region convolution neural network. Experiments on both artificial and natural cracks have shown attractive performance and verified the efficacy of the proposed structure.

A Novel Filtered Based Grid partitioning multiple reducers skyline computation using Hadoop framework

Due to the exponential growth of multi-dimensional skyline objects, the computational memory and efficiency of the traditional skyline measures also increases on large spatial datasets. Skyline query computation has attracted a major research problem recently, due to its exponential time and space complexities over imbalanced datasets. A large number of sequential skyline query processing models have been implemented to evaluate the spatial pattern discovery on limited datasets. It is practically expensive and time consuming process to predict various spatial patterns from a large spatial candidate sets. Another major limitation with the sequential spatial pattern mining models is that a large number of spatial candidate sets are generated with duplicate event sets. In the proposed model, a novel parallel skyline processing using MapReduce framework is implemented on large spatial uncertain datasets. In this model, a filtered based k-nearest neighbor approach is used to eliminate the sparsity or empty patterns using the hadoop framework. Experimental results proved that the proposed model has high computational efficiency in terms of time and candidate sets are concerned.

Domain-driven co-location mining

Co-location mining is a classical problem in spatial pattern mining. Considering a set of boolean spatial features, the goal is to find subsets of features frequently located together. It has wide applications in environmental management, public safety, transportation or tourism. These last years, many algorithms have been proposed to extract frequent co-locations. However, most solutions do a "data-centered knowledge discovery" instead of a "expert-centered knowledge discovery". Successfully providing useful and interpretable patterns to experts is still an open problem. In this setting, we propose a domain-driven co-location mining approach that combines constraint-based mining and cartographic visualization. Experts can push new domain constraints into the mining algorithm, resulting in more relevant patterns and more efficient extraction. Then, they can visualize solutions using a new concise and intuitive cartographic visualization of co-locations. Using this original visualization approach, they identify new interesting patterns, and use uninteresting ones to define new constraints and refine their analysis. These proposals have been integrated into a prototype based on PostGIS geographic information system. Experiments have been done using a real geological datasets studying soil erosion, and results have been validated by a domain expert.

Spatial Pattern Mining for Soil Erosion Characterization

The protection and the maintenance of the exceptional environment of New Caledonia are major goals for this territory. Among environmental problems, erosion has a strong impact on terrestrial and coastal ecosystems. However, due to the volume of data and its complexity, assessment of hazard at a regional scale is time-consuming, costly and rarely updated. Therefore, understanding and predicting environmental phenomenons need advanced techniques of analysis and modelization. In order to improve the understanding of the erosion phenomenon, this paper proposes a spatial approach based on co-location mining and GIS. Considering a set of Boolean spatial features, the goal of co-location mining is to find subsets of features often located together. This system provides useful and interpretable knowledge based on a new interestingness measure for co-locations and a new visualization of the discovered knowledge. The interestingness measure better reflects the importance of a co-location for the experts, and is completely integrated in the mining process. The visualization approach is a simple, concise and intuitive representation of the co-locations that takes into consideration the spatial nature of the underlying objects and the experts practice.