Search

This article presents our research in the prelaunch phase of the Kanyini mission, which aims to implement an energy-efficient, AI-based system onboard for early fire smoke detection using hyperspectral imagery. Our approach includes three key components: developing a diverse hyperspectral training dataset from VIIRS imagery, groundwork in band selection and AI model preparation, and developing an emulation system. We adapted and evaluated our previously developed lightweight convolutional neural network model, VIB_SD, to meet the computational constraints of satellite deployment. The emulation system tests various onboard AI tasks and processes. Our comprehensive experiments demonstrate the feasibility and benefits of employing onboard AI for fire smoke detection, significantly improving downlink efficiency, energy consumption, and detection speed.

Froth flotation, a widely used mineral beneficiation technique, generates substantial volumes of data, offering the opportunity to extract valuable insights from these data for production line analysis. The quality of flotation data is critical to designing accurate prediction models and process optimisation. Unfortunately, industrial flotation data are often compromised by quality issues such as outliers that can produce misleading or erroneous analytical results. A general approach is to preprocess the data by replacing or imputing outliers with data values that have no connection with the real state of the process. However, this does not resolve the effect of outliers, especially those that deviate from normal trends. Outliers often occur across multiple variables, and their values may occur in normal observation ranges, making their detection challenging. An unresolved challenge in outlier detection is determining how far an observation must be to be considered an outlier. Existing methods rely on domain experts’ knowledge, which is difficult to apply when experts encounter large volumes of data with complex relationships. In this paper, we propose an approach to conduct outlier analysis on a flotation dataset and examine the efficacy of multiple machine learning (ML) algorithms—including k-Nearest Neighbour (kNN), Local Outlier Factor (LOF), and Isolation Forest (ISF)—in relation to the statistical 2σ rule for identifying outliers. We introduce the concept of “quasi-outliers” determined by the 2σ threshold as a benchmark for assessing the ML algorithms’ performance. The study also analyses the mutual coverage between quasi-outliers and outliers from the ML algorithms to identify the most effective outlier detection algorithm. We found that the outliers by kNN cover outliers of other methods. We use the experimental results to show that outliers affect model prediction accuracy, and excluding outliers from training data can reduce the average prediction errors.

Landslides, resulting from disturbances in slope equilibrium, pose a significant threat to landscapes, infrastructure, and human life. Triggered by factors such as intense precipitation, seismic activities, or volcanic eruptions, these events can cause extensive damage and endanger nearby communities. A comprehensive understanding of landslide characteristics, including spatio-temporal patterns, dimensions, and morphology, is vital for effective landslide disaster management. Existing remote sensing approaches mostly use either optical or synthetic aperture radar sensors. Integrating information from both these types of sensors promises greater accuracy for identifying and locating landslides. This study proposes a novel approach, the ML-LaDeCORsat (Machine Learning-based coseismic Landslide Detection using Combined Optical and Radar Satellite Imagery), that integrates freely available Sentinel-1, Palsar-2, and Sentinel-2 imagery data in Google Earth Engine (GEE). The approach also integrates relevant spectral indices and suitable bands used in a machine learning-based classification of coseismic landslides. The approach includes a robust and reproducible training and validation strategy and allows one to choose between five classifiers (CART, Random Forest, GTB, SVM, and Naive Bayes). Using landslides from four different earthquake case studies, we demonstrate the superiority of our approach over existing solutions in coseismic landslide identification and localization, providing a GTB-based detection accuracy of 87–92%. ML-LaDeCORsat can be adapted to other landslide events (GEE script is provided). Transfer learning experiments proved that our model can be applied to other coseismic landslide events without the need for additional training data. Our novel approach therefore facilitates quick and reliable identification of coseismic landslides, highlighting its potential to contribute towards more effective disaster management.

Community wastewater management systems (CWMS) are small-scale wastewater treatment systems typically in regional and rural areas with less sophisticated treatment processes and often managed by local governments or communities. Research and industrial applications have demonstrated that online UV-Vis sensors have great potential for improving wastewater monitoring and treatment processes. Existing studies on the development of surrogate parameters with models from spectral data for wastewater were largely limited to lab-based. In contrast, industrial applications of these sensors have primarily targeted large wastewater treatment plants (WWTPs), leaving a gap in research for small-scale WWTPs. This paper demonstrates the suitability of using a field-based online UV-Vis sensor combined with advanced data analytics for CWMSs as an early warning for process upset to support sustainable operations. An industry case study is provided to demonstrate the development of surrogate monitoring parameters for total suspended solids (TSSs) and chemical oxygen demand (COD) using the UV-Vis spectral data from an online UV-Vis sensor. Absorbances at a wavelength of 625 nm (UV625) and absorbances at a wavelength of 265 nm (UV265) were identified as surrogate parameters to measure TSSs and COD, respectively. This study contributes to the improvement of WWTP performance with a continuous monitoring system by developing a process monitoring framework and optimization strategy.

Topic modelling is important for tackling several data mining tasks in information retrieval. While seminal topic modelling techniques such as Latent Dirichlet Allocation (LDA) have been proposed, the ubiquity of social media and the brevity of its texts pose unique challenges for such traditional topic modelling techniques. Several extensions including auxiliary aggregation, self aggregation and direct learning have been proposed to mitigate these challenges, however some still remain. These include a lack of consistency in the topics generated and the decline in model performance in applications involving disparate document lengths. There is a recent paradigm shift towards neural topic models, which are not suited for resource-constrained environments. This paper revisits LDA-style techniques, taking a theoretical approach to analyse the relationship between word co-occurrence and topic models. Our analysis shows that by altering the word co-occurrences within the corpus, topic discovery can be enhanced. Thus we propose a novel data transformation approach dubbed DATM to improve the topic discovery within a corpus. A rigorous empirical evaluation shows that DATM is not only powerful, but it can also be used in conjunction with existing benchmark techniques to significantly improve their effectiveness and their consistency by up to 2 fold.

Smoke plumes are the first things seen from space when wildfires occur. Thus, fire smoke detection is important for early fire detection. Deep Learning (DL) models have been used to detect fire smoke in satellite imagery for fire detection. However, previous DL-based research only considered lower spatial resolution sensors (e.g., Moderate-Resolution Imaging Spectroradiometer (MODIS)) and only used the visible (i.e., red, green, blue (RGB)) bands. To contribute towards solutions for early fire smoke detection, we constructed a six-band imagery dataset from Landsat 5 Thematic Mapper (TM) and Landsat 8 Operational Land Imager (OLI) with a 30-metre spatial resolution. The dataset consists of 1836 images in three classes, namely “Smoke”, “Clear”, and “Other_aerosol”. To prepare for potential on-board-of-small-satellite detection, we designed a lightweight Convolutional Neural Network (CNN) model named “Variant Input Bands for Smoke Detection (VIB_SD)”, which achieved competitive accuracy with the state-of-the-art model SAFA, with less than 2% of its number of parameters. We further investigated the impact of using additional Infra-Red (IR) bands on the accuracy of fire smoke detection with VIB_SD by training it with five different band combinations. The results demonstrated that adding the Near-Infra-Red (NIR) band improved prediction accuracy compared with only using the visible bands. Adding both Short-Wave Infra-Red (SWIR) bands can further improve the model performance compared with adding only one SWIR band. The case study showed that the model trained with multispectral bands could effectively detect fire smoke mixed with cloud over small geographic extents.

Water quality monitoring is an essential component of water quality management for water utilities for managing the drinking water supply. Online UV-Vis spectrophotometers are becoming popular choices for online water quality monitoring and process control, as they are reagent free, do not require sample pre-treatments and can provide continuous measurements. The advantages of the online UV-Vis sensors are that they can capture events and allow quicker responses to water quality changes compared to conventional water quality monitoring. This review summarizes the applications of online UV-Vis spectrophotometers for drinking water quality management in the last two decades. Water quality measurements can be performed directly using the built-in generic algorithms of the online UV-Vis instruments, including absorbance at 254 nm (UV254), colour, dissolved organic carbon (DOC), total organic carbon (TOC), turbidity and nitrate. To enhance the usability of this technique by providing a higher level of operations intelligence, the UV-Vis spectra combined with chemometrics approach offers simplicity, flexibility and applicability. The use of anomaly detection and an early warning was also discussed for drinking water quality monitoring at the source or in the distribution system. As most of the online UV-Vis instruments studies in the drinking water field were conducted at the laboratory- and pilot-scale, future work is needed for industrial-scale evaluation with ab appropriate validation methodology. Issues and potential solutions associated with online instruments for water quality monitoring have been provided. Current technique development outcomes indicate that future research and development work is needed for the integration of early warnings and real-time water treatment process control systems using the online UV-Vis spectrophotometers as part of the water quality management system.

Learning analytics (LA), like much of machine learning, assumes the training and test datasets come from the same distribution. Therefore, LA models built on past observations are (implicitly) expected to work well for future observations. However, this assumption does not always hold in practice because the dataset may drift. Recently, algorithmic fairness has gained significant attention. Nevertheless, algorithmic fairness research has paid little attention to dataset drift. Majority of the existing fairness algorithms are "statically" designed. Put another way, LA models tuned to be "fair" on past data are expected to still be "fair" when dealing with current/future data. However, it is counter-intuitive to deploy a statically fair algorithm to a nonstationary world. There is, therefore, a need to assess the impact of dataset drift on the unfairness of LA models. For this reason, we investigate the relationship between dataset drift and unfairness of LA models. Specifically, we first measure the degree of drift in the features (i.e., covariates) and target label of our dataset. After that, we train predictive models on the dataset and evaluate the relationship between the dataset drift and the unfairness of the predictive models. Our findings suggest a directly proportional relationship between dataset drift and unfairness. Further, we find covariate drift to have the most impact on unfairness of models as compared to target drift, and there are no guarantees that a once fair model would consistently remain fair. Our findings imply that "robustness" of fair LA models to dataset drift is necessary before deployment.