Proceedings MDPI’s Post

Informative read about what the university is doing to advancing #geospatial artificial intelligence “the GeoAI Research Center … is about enabling machines to process and reason about geospatial data with capabilities that surpass human limitations.”

During my master's program, I focused on developing a robust algorithm for road extraction and digitization from satellite images. Utilizing the power of deep learning, I implemented a UNet-based road segmentation model to extract road mask from satellite images, then breaking down mask into 64x64 tiles. Each tile was analyzed to identify road segments, which were then optimized using a Differential Evolution (DE) algorithm. The final step involved merging these optimized segments to create cohesive shape files, ensuring high accuracy in road mapping. Attached is video showcasing the UNet segmentation results and illustrating the optimization process. This approach not only extracting the roads centerlines but also roads widthes . That paves the way for more efficient infrastructure planning and management. #GeospatialAnalysis #SatelliteImaging #DeepLearning #RoadExtraction #UNet #DifferentialEvolution #RemoteSensing #GIS #MachineLearning #Innovation #DigitalMapping #SpatialData #DataScience

Our new peer-reviewed article is available online. This study introduces a real-time unsupervised monitoring framework for monitoring sinkhole formation events during earth pressure balance (EPB) shield tunneling operations. A feature extractor (FE) is constructed by coupling variational Autoencoders structure with convolutional neural network layers (VAE-CNN) to manage the complexity of EPB operational data, including non-linearity and temporal dependencies. The monitoring framework consists of two main phases: offline modeling and online monitoring. In the offline modeling phase, an FE model is trained using data-intensive techniques to define a subspace characterizing the behavior of multivariate data without sinkhole formations. The squared prediction error (SPE) statistics and the control limits are computed for detection. During the online monitoring phase, unseen EPB data is propagated to generate SPE values and determine sinkhole events based on whether these values surpass the control limit. Sensor validity index violation counts were used to isolate the most influential variables, while the results demonstrated the superiority of the proposed VAE-CNN method, achieving a 100% detection rate and a 0.9% false alarm rate. The influential variables identified include cutter resolutions per minute, jack speed, screw pressure, torque, and cutter seal components. The monitoring system shows great potential for early warnings during EPB operations to mitigate sinkhole formation risks. https://lnkd.in/gt_cverb

Recently published: The challenge of land in a neural network ocean model 👉 https://bit.ly/4jbtJmB Process-based models are powerful tools for understanding and predicting the earth system, but they are computationally expensive. To address this issue, machine learning (ML) techniques are being utilized to provide computationally cheaper models that can predict atmospheric weather with accuracies comparable to leading operational forecast systems. However, to apply ML-based models to ocean prediction, we need to consider the representation of land, which influences the flow but is not modeled itself. In this study, we developed a CNN ocean model and found that while the model has great predictive ability overall, its performance near land is poor. Simple methods used in other studies for representing land are not sufficient. Our results suggest that new ML techniques need to be developed specifically for this area, and their performance must be assessed by focusing on their ability to predict near-land conditions. Overall, this study highlights the potential of ML techniques for ocean prediction and emphasizes the need for further research to improve their accuracy and applicability in complex coastal environments. By Rachel Furner, Peter Haynes, Dani C Jones, Dave Munday, Brooks Paige & Emily Shuckburgh #MachineLearning #oceanography #forecasting #coast

I am pleased to announce that the final "Open Access" version of my article "Real-time unsupervised monitoring of earth pressure balance shield-induced sinkholes in mixed-face ground conditions via convolutional variational autoencoders" is now available online. This study introduces a real-time unsupervised monitoring framework for monitoring sinkhole formation events during earth pressure balance (EPB) shield tunneling operations. A feature extractor (FE) is constructed by coupling variational Autoencoders structure with convolutional neural network layers (VAE-CNN) to manage the complexity of EPB operational data, including non-linearity and temporal dependencies. The monitoring framework consists of two main phases: offline modeling and online monitoring. In the offline modeling phase, an FE model is trained using data-intensive techniques to define a subspace characterizing the behavior of multivariate data without sinkhole formations. The squared prediction error (SPE) statistics and the control limits are computed for detection. During the online monitoring phase, unseen EPB data is propagated to generate SPE values and determine sinkhole events based on whether these values surpass the control limit. Sensor validity index violation counts were used to isolate the most influential variables, while the results demonstrated the superiority of the proposed VAE-CNN method, achieving a 100% detection rate and a 0.9% false alarm rate. The influential variables identified include cutter resolutions per minute, jack speed, screw pressure, torque, and cutter seal components. The monitoring system shows great potential for early warnings during EPB operations to mitigate sinkhole formation risks. Thanks Sean

The future of AI in unlimited

How can AI help solve environmental challenges? 🧠 https://lnkd.in/e2dcz52y 💻 Advancements in data collection technology have made it possible to gather data more quickly and cost effectively than ever before, and machine learning can help us compute comprehensive information to make better decisions. 🔢 Andrew Reicks is a geospatial data scientist with nearly a decade of experience in the field, and he leads the Sky Wave team’s remote sensing and machine learning applications. Drew is a pro at analyzing site data, remote sensing, and developing models that allow experts to compute mass amounts of environmental and infrastructural information to solve the complex site challenges of today. Learn more about #SkyWaveatCDMSmith: https://lnkd.in/eXxurYgw

We are hard at work at building advanced AI Agent framework with powerful spatial searching and operations and spatial analysis and geoprocessing

Here is another deep learning application recently published, with Sagun Kayastha as the first author, from UH Choi's AQF and machine learning group. We developed a deep learning model designed to fill gaps in PM2.5 pollution data, offering a comprehensive view of air quality across the United States, even in regions lacking monitoring stations. By integrating satellite data with variables such as meteorology, land use, and urbanization, our model significantly improves the accuracy of pollution estimates and uncovers previously unnoticed hotspots. This innovation supports better health and environmental decision-making, particularly benefiting underserved communities. https://lnkd.in/gPfw36X9

I am excited to share our latest publication, "Efficient Large-scale Nonstationary Spatial Covariance Function Estimation Using Convolutional Neural Networks," which discusses leveraging Convolutional Neural Networks (#ConvNets) to estimate nonstationary spatial covariance functions, particularly in large-scale geospatial datasets! Our study tackled the challenges of modeling spatial processes in fields like climate science, where spatial nonstationarity often complicates Gaussian Process modeling. Our innovative approach uses ConvNets to dynamically partition spatial regions, identifying subregions that exhibit stationary-like behavior. This enables more accurate parameter estimation than traditional, user-defined methods. Key Contributions: - Developed a data-driven, scalable framework using ConvNets for nonstationary Matérn covariance estimation. - Improved parameter estimation accuracy and handling large datasets more efficiently. - Demonstrated effectiveness with synthetic and real-world geospatial data, achieving more reliable modeling of spatial dependencies. This work opens up new possibilities for large-scale geospatial modeling with real-world applications, from environmental monitoring to climate modeling! The paper was led by Pratik Nag, PhD. Paper: https://lnkd.in/ecFR9i6g Published in Journal of Computational and Graphical Statistics. #SpatialStatistics #MachineLearning #GeospatialModeling #ConvolutionalNeuralNetworks #HighPerformanceComputing