Operational Parameters, Data Density and Benthic Ecology: Considerations for Image-Based Classification of Multibeam Backscatter

Face recognition has witnessed significant progress due to the advances of deep convolutional neural networks (CNNs), the central task of which is how to improve the feature discrimination. To this end, several margin-based (e.g., angular, additive and additive angular margins) softmax loss functions have been proposed to increase the feature margin between different classes. However, despite great achievements have been made, they mainly suffer from three issues: 1) Obviously, they ignore the importance of informative features mining for discriminative learning; 2) They encourage the feature margin only from the ground truth class, without realizing the discriminability from other non-ground truth classes; 3) The feature margin between different classes is set to be same and fixed, which may not adapt the situations very well. To cope with these issues, this paper develops a novel loss function, which adaptively emphasizes the mis-classified feature vectors to guide the discriminative feature learning. Thus we can address all the above issues and achieve more discriminative face features. To the best of our knowledge, this is the first attempt to inherit the advantages of feature margin and feature mining into a unified loss function. Experimental results on several benchmarks have demonstrated the effectiveness of our method over state-of-the-art alternatives. Our code is available at http://www.cbsr.ia.ac.cn/users/xiaobowang/.

The behavior and operation of integrated assemblies comprised of arrays of active transducers interfaced with complex cast compliant structures depends on their true structural dynamics. Such assemblies are characteristic of control elements for adaptive optical systems, sonar systems, and other continuously deformable structures in which energy and motion are transmitted through complex compliant materials. The mechanical and physical interfaces between active transducers and their supporting and radiating structures relate directly to their dynamics.Sophisticated bonding and processing of compliant and rigid materials in such complex assemblies determine their operational parameters. Understanding the effects of structural anomalies and defects is of critical importance to insuring correct array operation, and a non-destructive method for the visualization of anomalous structural and material characteristics would be of great value in both development and testing of such systems. Also, mechanisms for acoustic and mechanical energy transfer through these structures into adjoining media can be inferred from such visualizations. Holographic interferometry presents itself as a viable and useful method for the non-destructive realization of this type of information.The behavior and operation of integrated assemblies comprised of arrays of active transducers interfaced with complex cast compliant structures depends on their true structural dynamics. Such assemblies are characteristic of control elements for adaptive optical systems, sonar systems, and other continuously deformable structures in which energy and motion are transmitted through complex compliant materials. The mechanical and physical interfaces between active transducers and their supporting and radiating structures relate directly to their dynamics.Sophisticated bonding and processing of compliant and rigid materials in such complex assemblies determine their operational parameters. Understanding the effects of structural anomalies and defects is of critical importance to insuring correct array operation, and a non-destructive method for the visualization of anomalous structural and material characteristics would be of great value in both development and testing of such systems. Also, mechanisms for ac...

A comparison of top-down and bottom-up approaches to benthic habitat mapping to inform offshore wind energy development

In this digital era, the recommendation system is an important tool in helping users to make decisions, including choosing tourist destinations. This study aims to implement the K-Nearest Neighbors (k-NN) algorithm in determining the recommendation pattern for tourist destinations. The k-NN algorithm was chosen because of its ability to classify based on the closest neighbor of a point, so that it can obtain recommendation patterns that are more personal and relevant to user preferences. This implementation method involves data collection, data preparation, model building, classification process, model evaluation, and finally implementation of the model in a recommendation system. The results of this study indicate that a recommendation system based on the k-NN algorithm is able to provide suggestions for tourist destinations that are more in line with user preferences, thus potentially increasing user satisfaction and making a positive contribution to the tourism industry. However, the effectiveness of this model is highly dependent on the quality and quantity of data used. Therefore, good data collection and preparation is very important in the implementation of this algorithm. This research opens opportunities for further research and development in the field of tourist destination recommendation systems using other algorithms or combining several algorithms to obtain more optimal results. The effectiveness of this model is highly dependent on the quality and quantity of data used. Therefore, good data collection and preparation is essential in the implementation of this algorithm. This research opens up opportunities for further research and development in the field of tourist destination recommendation systems by using other algorithms or combining several algorithms to get more optimal results. The effectiveness of this model is highly dependent on the quality and quantity of data used. Therefore, good data collection and preparation is very important in the implementation of this algorithm. This research opens up opportunities for further research and development in the field of tourist destination recommendation systems by using other algorithms or combining several algorithms to get more optimal results.

Data quality plays a vital role in scientific research and decision-making across industries. Thus, it is crucial to incorporate the data quality control (DQC) process, which comprises various actions and operations to detect and correct data errors. The increasing adoption of machine learning (ML) techniques in different domains has raised concerns about data quality in the ML field. Conversely, ML’s capability to uncover complex patterns makes it suitable for addressing challenges involved in the DQC process. However, supervised learning methods demand abundant labeled data, while unsupervised learning methods heavily rely on the underlying distribution of the data. Active learning (AL) provides a promising solution by proactively selecting data points for inspection, thus reducing the burden of data labeling for domain experts. Therefore, this survey focuses on applying AL to DQC. Starting with a review of common data quality issues and solutions in the ML field, we aim to enhance the understanding of current quality assessment methods. We then present two scenarios to illustrate the adoption of AL into the DQC systems on the anomaly detection task, including pool-based and stream-based approaches. Finally, we provide the remaining challenges and research opportunities in this field.

Industry 4.0 is generally defined as a development system that compels the digitalization of processes to create integrated and autonomous systems. The process tracking of parts is very important in terms of detecting missed faulty products. Some defects that escape from quality control directly affect the end-user. Machine learning algorithms have been used to predict changes in the quality control processes and defective products, toward real-time and effective data processing. Thus, the highest quality of the final product will be delivered to the customer and to reduce the defective production coming out of the manufacturing chain. In this article, the study aims to establish a predictive quality model that can detect defect-free approved but faulty products overlooked during the quality inspection operations. Machine learning methods are used to analyze the relationship between quality control data and customer complaints. For this purpose, we use the last quality stage data of an automobile manufacturer’s brake system from 2018 to 2020. Machine learning models are constructed using logistic regression, ridge regression, support vector machine, random forest classification tree, gradient boost, XGBoost, LightGBM, and CatBoost algorithms. The results of specificity and negative prediction value show that the Gradient Boost and CatBoost algorithms have the best classification benefit for detecting the rare events.KeywordsQuality 4.0Industry 4.0Predictive qualityAutomobile industryMachine learningFault detectionRare event detection

Offshore wind developers obtain extensive geophysical, geotechnical, and habitat data during Site Characterization activities. Integration and delivery of this information to a diverse group of stakeholders and Government agencies is required. We present an integrated benthic habitat mapping approach tailored to regional geology and ground conditions and discuss how various data was utilized to deliver multiple components of the permitting process. Multiple data sets were integrated and presented via a web-based GIS platform to aid delivery, visualization, and communication. Our unified approach to benthic habitat mapping and delivery of products to stakeholders was instrumental in successfully coalescing multiple performers to develop their individual deliverables in a cohesive and rapid manner. This approach reduced risk to schedule and budget, without sacrificing data density or quality. Four annual (2019–2022) benthic surveys were acquired to support Site Characterization and subsequent permitting processes. High-Resolution Geophysical data were collected concomitantly with the 2020 benthic survey data and used to refine subsequent 2021 and 2022 benthic survey designs. Benthic survey data consisted of grab sample tests (grain size), macrofaunal taxonomy, sediment profile and plan view imagery (SPI-PV), video imagery from each grab station, and towed video transects. Acoustic data products were processed and interpreted to create polygons of seafloor sediment coverage over the ASOW study area and ground-truthed with physical sampling, video, and digital still imagery to refine and validate acoustic data into a mappable model of essential fish and benthic habitats. Seafloor morphology and seabed sediment interpretations were coalesced into a benthic habitat model that displayed substrates consisting mostly of mobile sand sheets, with interspersed areas of gravelly sand and discrete patches of gravel. Overlying the substrate model was a range of benthic features and morphologies, including sand ridges, sand waves, megaripples, ripples, areas of depressional marks, hummocky seafloor, interbedded surficial sediments, irregular seafloor, and localized relief features. From these data, classified maps of Coastal Marine Ecological Standard (CMECS) substrates and fish habitats were made. Additional CMECS classification of benthic biotic components were mapped, showing the taxonomic communities that are present in each substrate. Seabed sediment modeling and morphological trends were dynamically studied and compiled into an interpreted and GIS-friendly dataset that enabled rapid online transfer to subject matter experts tasked with quantifying the benthic ecosystem across the development area. The methods and modeling that were produced by expert refinement of geophysical data to reflect the physically observed habitat structures allowed for dynamic minimum mapping unit variability while also isolating and identifying key areas of interest for benthic researchers and regulators. This mapping process led to an efficient and unified approach for all teams, saving project time and expense.

Classification situations are ubiquitous and arise in a range of practical applications. Very often classification problems do not occur in isolation but in clusters where several classification problems need to be solved. Examples of these include employee contract renewal in large organizations, patient prognosis in hospitals, and crowdsourced problem solving. With any form of classification, however, there are invariably inaccuracies occurring in different forms, especially when multiple classification tasks need to be performed. Since typical classifiers are not free from errors, classification errors tend to accumulate, and having frequent misclassifications are often unacceptable. Moreover, in unsupervised learning situations, there are typically no pre-determined ground truth classes; in such a situation the ground truth class is determined by the view of the majority of classifiers. In this paper, we study the situation of multiple classification within the Naive Bayes framework, where the ground truth is determined by the decision of most classifiers, and where there is finite resources requiring decisions to be made within limited budget. Here, we represent the classification tasks as a random walk process, and perform a stochastic analysis of the situation and predict the occurrence of erroneous classification. We obtained closed-form solutions for the probability of classification inaccuracies and derive useful error bounds as a function of the budget. We find that by raising the budget, the probability of error in classification can be controlled, and the extent of the reduction can be quantified. Experimental measurements are also performed which show good agreement with the theoretical results.

Chapter 2 - Data, People, and Systems

Report: Release of the International Life Sciences Institute Crop Composition Database Version 5

Support Vector Machines (SVMs) have been widely used for classification tasks, but face significant challenges including sensitivity to noise and outliers, and the critical need for effective feature selection. This paper proposes a robust Mixed Integer Linear Programming (MILP) model based on Support Vector Machines (SVMs) that simultaneously addresses both issues. Our model, Pin-FS-SVM, incorporates a budget constraint to limit feature selection and employs the pinball loss function for noise and outlier handling in the classification process. Experiments on multiple datasets show that Pin-FS-SVM outperforms existing methods in robustness to noise and outliers, achieves better feature reduction, and maintains competitive performance. Impact Statement-Classification tasks are fundamental in modern data analysis, but real-world datasets often face challenges with noise, outliers and high-dimensional feature spaces that hinder model performance and interpretability. Traditional SVMs struggle with these issues, requiring extensive data preprocessing. Our Pin-FS-SVM model addresses these challenges through a combination of feature selection under a budget constraint and robust handling of noise and outliers, reducing the need for preprocessing and enhancing performance. This approach significantly lowers the number of features while maintaining competitive performance, making it ideal for sensitive applications where model interpretability is crucial such as medical diagnostics and financial risk assessment. By improving robustness and interpretability, Pin-FS-SVM enables broader adoption of machine learning in domains where data quality is uncertain, paving the way for more transparent and reliable decision-making systems in both technological and societal contexts.

Thermal enhanced oil recovery in deep heavy oil carbonates: Experimental and numerical study on a hot water injection performance

Detailed knowledge of the physical and biological conditions of the seafloor is key to the development of economic, environmental and cultural activities in the marine environment. Full coverage information on substrate and benthic habitat relies almost exclusively on acoustic remote-sensed data provided by multibeam echosounders, subsequently ground-truthed using visual observations and physical sampling. Predictive habitat mapping provides a means to develop models of seafloor habitat in regions where only acoustic data are available. The diverse geomorphology and benthic ecology in the Bay of Plenty makes the area a useful case study for developing regional and local models of benthic habitat: water depths range 0 - 3300 m with geomorphological features including channel, canyons, seamounts and ridges at a variety of scales. Active geology is demonstrated by the intense seismicity, dense fault network, and ubiquitous submarine volcanic activity that results in venting hydrothermal fluids at, for example, the Calypso vent fields. This environment provides habitat for benthic communities, whose structure and distribution can be expressed at a variety of scales. Multiscale seafloor topography classification is strongly dependent on the quality of the bathymetric data. In this study we combined multibeam echosounder (MBES) data from a variety of systems, data collected on transit, and dedicated scientific and hydrographic surveys, with data from the New Zealand national bathymetry model derived from historic and modern single-beam echosounder (SBES) data for areas without MBES coverage. We integrated marine geological and oceanographic information in Benthic Terrain Models of the Bay of Plenty and the Calypso vent fields, i.e. at regional and local scales using the ESRI ArcGIS Benthic Terrain Modeler (BTM) tools, with a modified classification catalog to suit the highly variable terrain of the Bay of Plenty. Classification of the geomorphology was based on a number of morphometric parameters (e.g., slope, Bathymetric Positioning Index, curvature, rugosity) derived from the bathymetry gridded at 25 m for the region, and 5 m for the Calypso vent fields. A 13 geomorphological class catalogue was used that differentiated large (broad slopes, basins) and local (knolls, narrow ridges/valleys) scale features. Seventy-seven percent of the 25,000 km2 study area is broad flat, the remainder is dominated by local ridges and narrow depressions. Ground-truthing data will enable us to associate biological observations to each class, and predict benthic habitat in region where no observations exist.

Accurate dosimetry measurements are essential to the maximum success of commercial application of radiation processing. Such dosimetry must be involved in three different areas to assure proper final quality in the processing program. These are; 1) dosimetry related to the basic research on the radiation process involved, 2) dosimetric calibration of the radiation facility at the time of its commissioning and each major modification, and 3) radiation monitoring of the routine production process. Although careful attention is usually given to the first area and sometimes the second, the third area is often given too little importance. The research on a given radiation process should include, as a minimum, the measurement of such radiation quantities as absorbed dose, dose rate, apd dose distribution in the material. The facility calibration should define the complete radiation field characteristics such as radiation type, energy, dose rates, geometric distributions, and depth dose profiles for typical materials. The quality assurance monitoring of the production process should include measurement of the appropriate radiation quantities as well as operational parameters like dwell time, conveyor speed, etc. In particular, where the safety of the product is dependent upon the absorbed dose, such as in sterilization or food preservation, dosimetric monitoring is of vital importance. Quality assurance of the production process can be maintained by monitoring the radiation beam or field; or by measurement of the absorbed dose in the product; or by product testing after processing.

Hyperspectral images of high spatial and spectral resolutions are employed to perform the challenging task of brain tissue characterization and subsequent segmentation for visualization of in-vivo images. Each pixel is high-dimensional spectrum. Working on the hypothesis of pure-pixels on account of high spectral resolution, we perform unsupervised clustering by hierarchical non-negative matrix factorization to identify the pure-pixel spectral signatures of blood, brain tissues, tumor and other materials. This subspace clustering was further used to train a random forest for subsequent classification of test set images constituent of in-vivo and ex-vivo images. Unsupervised hierarchical clustering helps visualize tissue structure in in-vivo test images and provides a inter-operative tool for surgeons. Furthermore the study also provide a preliminary study of the classification and sources of errors in the classification process.