Composite Dataset Research Articles

TPE-xgboost for explainable predictions of concrete compressive strength considering compositions, and mechanical and microstructure properties of testing samples

The use of machine learning (ML) for predicting concrete compressive strength (CCS) has shown promising and accurate results, making it a valuable tool in the field. However, efficient prediction requires not only a robust ML approach but also a comprehensive and well-curated dataset. This study addresses this challenge by investigating an extensive and integrated dataset of (1) concrete compositions (mixture proportions) and (2) testing conditions (mechanical and microstructure properties of concrete testing samples), containing 1525 observations relating to 39 parameters. On algorithms side, this research proposes novel tree-structured parzen estimator based extreme gradient boosting (TPE-xgboost) for accurate and confident CCS predictions. Moreover, SHapley Additive exPlanations (SHAP) analysis was performed to provide a comprehensive understanding of feature importance, dependencies, and interactions. Compared to prior research, this study demonstrates significant improvements in CCS prediction accuracy for separate investigations on concrete compositions (3.77%) and mechanical and microstructure properties (28.57%), while maintaining reasonable accuracy for the integrated dataset despite its complexity and sparseness. SHAP analysis reveals key influential factors, including age and water-to-binder ratio in concrete composition, and peak load and height of testing samples, as well as microstructural characteristics such as mean values of global autocorrelation length and its integral range. This research provides valuable insights into model optimization, predictive performance, and feature dependencies and interactions, contributing to the development of more accurate and reliable CCS prediction models.

MCCA-VNet: A Vit-Based Deep Learning Approach for Micro-Expression Recognition Based on Facial Coding

In terms of facial expressions, micro-expressions are more realistic than macro-expressions and provide more valuable information, which can be widely used in psychological counseling and clinical diagnosis. In the past few years, deep learning methods based on optical flow and Transformer have achieved excellent results in this field, but most of the current algorithms are mainly concentrated on establishing a serialized token through the self-attention model, and they do not take into account the spatial relationship between facial landmarks. For the locality and changes in the micro-facial conditions themselves, we propose the deep learning model MCCA-VNET on the basis of Transformer. We effectively extract the changing features as the input of the model, fusing channel attention and spatial attention into Vision Transformer to capture correlations between features in different dimensions, which enhances the accuracy of the identification of micro-expressions. In order to verify the effectiveness of the algorithm mentioned, we conduct experimental testing in the SAMM, CAS (ME) II, and SMIC datasets and compared the results with other former best algorithms. Our algorithms can improve the UF1 score and UAR score to, respectively, 0.8676 and 0.8622 for the composite dataset, and they are better than other algorithms on multiple indicators, achieving the best comprehensive performance.

Sample Article for Pree Reviewer

Background Large research consortia can generate tremendous amounts of biological information, including high-resolution soil, vegetation, and climate data. While this knowledge stock holds invaluable potential for answering evolutionary and ecological questions, making these data exploitable for modelling remains a daunting task due to the many processing steps required for synthesis. This might result in many researchers to fall back to a handful of ready-to-use data sets, potentially at the expense of statistical power and scientific rigour. In a push for a more stringent approach, we introduce BEpipeR, an R pipeline that allows for the streamlined synthesis of plot-based Biodiversity Exploratories data. Methods BEpipeR was designed with flexibility and ease of use in mind. For instance, users simply choose between aggregating forest or grassland data, or a combination thereof, effectively allowing them to process any experimental plot data of this research consortium. Additionally, instead of coding, they parse most processing information in a user-friendly way through parameter sheets. Processing includes, among others, the creation of a spatially explicit plot-ID template, data wrangling, quality control, plot-wise aggregations, the calculation of derived metrics, data joining to a large composite data set, and metadata compilation. Results With BEpipeR, we provide a feature-rich pipeline that allows users to process Biodiversity Exploratories data in a flexible and reproducible way. This pipeline might serve as a starting point for aggregating the numerous data sets of this and potentially similar research consortia. In this way, it might be a primer for the construction of consortia-wide composite data sets that take full advantage of the consortia’s rich information stocks, ultimately boosting the visibility and participation of individual research projects. Conclusions The BEpipeR permits the user-friendly processing and plot-wise aggregation of Biodiversity Exploratories data. With modifications, this framework may be easily adopted by other research consortia.

A Comprehensive Evaluation of Iris Segmentation on Benchmarking Datasets.

Iris is one of the most widely used biometric modalities because of its uniqueness, high matching performance, and inherently secure nature. Iris segmentation is an essential preliminary step for iris-based biometric authentication. The authentication accuracy is directly connected with the iris segmentation accuracy. In the last few years, deep-learning-based iris segmentation methodologies have increasingly been adopted because of their ability to handle challenging segmentation tasks and their advantages over traditional segmentation techniques. However, the biggest challenge to the biometric community is the scarcity of open-source resources for adoption for application and reproducibility. This review provides a comprehensive examination of available open-source iris segmentation resources, including datasets, algorithms, and tools. In the process, we designed three U-Net and U-Net++ architecture-influenced segmentation algorithms as standard benchmarks, trained them on a large composite dataset (>45K samples), and created 1K manually segmented ground truth masks. Overall, eleven state-of-the-art algorithms were benchmarked against five datasets encompassing multiple sensors, environmental conditions, demography, and illumination. This assessment highlights the strengths, limitations, and practical implications of each method and identifies gaps that future studies should address to improve segmentation accuracy and robustness. To foster future research, all resources developed during this work would be made publicly available.

Chemical Composition and Bioactivity Dataset Integration to Identify Antiproliferative Compounds in Phyllanthus Plants.

Phyllanthus species are renowned in traditional medicine for their diverse therapeutic properties, including potential anticancer activities. This study explored the antiproliferative potential of six Phyllanthus species by integrating chemical composition with bioactivity assays to identify key antiproliferative compounds. The integration of liquid chromatography-mass spectrometry (LC-MS)-based chemical composition data with antiproliferative activity against three cancer cell lines-PC-3 (prostate adenocarcinoma), SiHa (cervical carcinoma), and A549 (lung carcinoma)-as well as a normal mouse fibroblast line (L929) was performed by covariate analysis. These compounds were subsequently isolated and structurally characterized using spectroscopic methods. Through covariate statistics, seven m/z features were found to be plausible active compounds, and after isolation, they were related to butyrolactone and arylnaphthalide lignans. Among the active isolates, an unreported compound, (+)-phyllanlathyrin 6, was discovered in the aerial part of Phyllanthus lathyroides. The isolated compounds exhibited moderate to good antiproliferative activity (IC50 < 20 µM) with selectivity to SiHa, validating the covariate-based identification approach. These findings highlight the potential of Phyllanthus species as sources of novel anticancer agents, with specific arylnaphthalide lignans showing promising cytotoxic effects that could be further developed into therapeutic leads. Additionally, this study underscores the value of combining advanced analytical techniques with bioactivity testing to uncover bioactive compounds from natural sources. The results contribute to the growing body of evidence supporting the therapeutic relevance of Phyllanthus species and provide a foundation for future drug development efforts targeting cancer treatment.

BEpipeR: a user-friendly, flexible, and scalable data synthesis pipeline for the Biodiversity Exploratories and other research consortia

Abstract Background Large research consortia can generate tremendous amounts of biological information, including high-resolution soil, vegetation, and climate data. While this knowledge stock holds invaluable potential for answering evolutionary and ecological questions, making these data exploitable for modelling remains a daunting task due to the many processing steps required for synthesis. This might result in many researchers to fall back to a handful of ready-to-use data sets, potentially at the expense of statistical power and scientific rigour. In a push for a more stringent approach, we introduce BEpipeR, an R pipeline that allows for the streamlined synthesis of plot-based Biodiversity Exploratories data. Methods BEpipeR was designed with flexibility and ease of use in mind. For instance, users simply choose between aggregating forest or grassland data, or a combination thereof, effectively allowing them to process any experimental plot data of this research consortium. Additionally, instead of coding, they parse most processing information in a user-friendly way through parameter sheets. Processing includes, among others, the creation of a spatially explicit plot-ID template, data wrangling, quality control, plot-wise aggregations, the calculation of derived metrics, data joining to a large composite data set, and metadata compilation. Results With BEpipeR, we provide a feature-rich pipeline that allows users to process Biodiversity Exploratories data in a flexible and reproducible way. This pipeline might serve as a starting point for aggregating the numerous data sets of this and potentially similar research consortia. In this way, it might be a primer for the construction of consortia-wide composite data sets that take full advantage of the consortia’s rich information stocks, ultimately boosting the visibility and participation of individual research projects. Conclusions The BEpipeR pipeline permits the user-friendly processing and plot-wise aggregation of Biodiversity Exploratories data. With modifications, this framework may be easily adopted by other research consortia.

Development and evaluation of a novel framework to enhance k-NN algorithm’s accuracy in data sparsity contexts

This paper presents a novel framework for implementing the k-NN algorithm, designed to enhance its accuracy in contexts with sparse data. The framework addresses limitations in the algorithm’s training process by optimizing data structures. It employs composite datasets generated from the initial data using a data-driven fuzzy Analytic Hierarchy Process weighting scheme. This approach is designed to enhance the informational content in the initial datasets, thus reducing the entropy and implementation uncertainty. The framework was evaluated using 75 publicly available datasets and 3 generated datasets, demonstrating significant accuracy improvements across various k-parameter values. The findings were rigorously generalized using non-parametric hypothesis tests; while the resulting sensitivity was assessed by applying different distance metrics. By enhancing informational content, the composite data structures contribute to both accuracy improvements and scalability, particularly in data-sparse contexts. This relationship underscores the critical role of entropy in enhancing the performance of explainable machine learning algorithms, providing a valuable and interpretable tool for transforming data structures in sparse data environments.

Cassava disease detection using a lightweight modified soft attention network.

Cassava is a high-carbohydrate crop that is at risk of viral infections. The production rate and quality of cassava crops are affected by several diseases. However, the manual identification of diseases is challenging and requires considerable time because of the lack of field professionals and the limited availability of clear and distinct information. Consequently, the agricultural management system is seeking an efficient and lightweight method that can be deployable to edged devices for detecting diseases at an early stage. To address these issues and accurately categorize different diseases, a very effective and lightweight framework called CDDNet has been introduced. We used MobileNetV3Small framework as a backbone feature for extracting optimized, discriminating, and distinct features. These features are empirically validated at the early intermediate stage. Additionally, we modified the soft attention module to effectively prioritize the diseased regions and enhance significant cassava plant disease-related features for efficient cassava disease detection. Our proposed method achieved accuracies of 98.95%, 97.03%, and 98.25% on Cassava Image Dataset, Cassava Plant Disease Merged (2019-2020) Dataset, and the newly created Cassava Plant Composite Dataset, respectively. Furthermore, the proposed technique outperforms previous state-of-the-art methods in terms of accuracy, parameter count, and frames per second values, ultimately making the proposed CDDNet the best one for real-time processing. Our findings underscore the importance of a lightweight and efficient technique for cassava disease detection and classification in a real-time environment. Furthermore, we highlight the impact of modified soft attention on model performance. © 2024 The Author(s). Pest Management Science published by John Wiley & Sons Ltd on behalf of Society of Chemical Industry.

Rapid shifts in grassland communities driven by climate change

Many terrestrial plant communities, especially forests, have been shown to lag in response to rapid climate change. Grassland communities may respond more quickly to novel climates, as they consist mostly of short-lived species, which are directly exposed to macroclimate change. Here we report the rapid response of grassland communities to climate change in the California Floristic Province. We estimated 349 vascular plant species’ climatic niches from 829,337 occurrence records, compiled 15 long-term community composition datasets from 12 observational studies and 3 global change experiments, and analysed community compositional shifts in the climate niche space. We show that communities experienced significant shifts towards species associated with warmer and drier locations at rates of 0.0216 ± 0.00592 °C yr−1 (mean ± s.e.) and −3.04 ± 0.742 mm yr−1, and these changes occurred at a pace similar to that of climate warming and drying. These directional shifts were consistent across observations and experiments. Our findings contrast with the lagged responses observed in communities dominated by long-lived plants and suggest greater biodiversity changes than expected in the near future.

Quantifying age-specific household contacts in Aotearoa New Zealand for infectious disease modelling.

Accounting for population age structure and age-specific contact patterns is crucial for accurate modelling of human infectious disease dynamics and impact. A common approach is to use contact matrices, which estimate the number of contacts between individuals of different ages. These contact matrices are frequently based on data collected from populations with very different demographic and socio-economic characteristics from the population of interest. Here we use a comprehensive household composition dataset based on Aotearoa New Zealand census and administrative data to construct a household contact matrix and a synthetic population that can be used for modelling. We investigate the behaviour of a compartment-based and an agent-based epidemic model parametrized using these data, compared with a commonly used contact matrix that was constructed by projecting international data onto New Zealand's population. We find that using the New Zealand household data, either in a compartment-based model or in an agent-based model, leads to lower attack rates in older age groups compared with using the projected contact matrix. This difference becomes larger when household transmission is more dominant relative to non-household transmission. We provide electronic versions of the synthetic population and household contact matrix for other researchers to use in infectious disease models.

Drone Detection Performance Evaluation via Real Experiments with Additional Synthetic Darkness

Detecting drones is increasingly challenging, particularly when developing passive and low-cost defense systems capable of countering malicious attacks in environments with high levels of darkness and severe weather conditions. This research addresses the problem of drone detection under varying darkness levels by conducting an extensive study using deep learning models. Specifically, the study evaluates the performance of three advanced models: Yolov8, Vision Transformers (ViT), and Long Short-Term Memory (LSTM) networks. The primary focus is on how these models perform under synthetic darkness conditions, ranging from 20% to 80%, using a composite dataset (CONNECT-M) that simulates nighttime scenarios. The methodology involves applying transfer learning to enhance the base models, creating Yolov8-T, ViT-T, and LSTM-T variants. These models are then tested across multiple datasets with varying darkness levels. The results reveal that all models experience a decline in performance as darkness increases, as measured by Precision-Recall and ROC Curves. However, the transfer learning-enhanced models consistently outperform their original counterparts. Notably, Yolov8-T demonstrates the most robust performance, maintaining higher accuracy across all darkness levels. Despite the general decline in performance with increasing darkness, each model achieves an accuracy above 0.6 for data subjected to 60% or greater darkness. The findings highlight the challenges of drone detection under low-light conditions and emphasize the effectiveness of transfer learning in improving model resilience. The research suggests further exploration into multi-modal systems that combine audio and optical methods to enhance detection capabilities in diverse environmental settings.

An innovative approach for QoS-aware web service composition using whale optimization algorithm

With the proliferation of services and the vast amount of data produced by the Internet, numerous services with comparable functionalities but varying Quality of Service (QoS) attributes are potential candidates for meeting user needs. Consequently, the selection of the most suitable services has become increasingly challenging. To address this issue, a synthesis of multiple services is conducted through a composition process to create more sophisticated services. In recent years, there has been a growing interest in QoS uncertainty, given its potential impact on determining an optimal composite service, where each service is characterized by multiple QoS properties (e.g., response time and cost) that are frequently subject to change primarily due to environmental factors. Here, we introduce a novel approach that depends on the Multi-Agent Whale Optimization Algorithm (MA-WOA) for web service composition problem. Our proposed algorithm utilizes a multi-agent system for the representation and control of potential services, utilizing MA-WOA to identify the optimal composition that meets the user’s requirements. It accounts for multiple quality factors and employs a weighted aggregation function to combine them into a cohesive fitness function. The efficiency of the suggested method is evaluated using a real and artificial web service composition dataset (comprising a total of 52,000 web services), with results indicating its superiority over other state-of-the-art methods in terms of composition quality and computational effectiveness. Therefore, the proposed strategy presents a feasible and effective solution to the web service composition challenge, representing a significant advancement in the field of service-oriented computing.

Vertical transport of anthropogenic lead by reversible scavenging in the South Atlantic Ocean

Stable lead (Pb) isotopes have been regarded as tracers of ocean circulation, both in the present time and geological past. Here we present a new dataset of seawater Pb concentrations and isotope compositions for ten depth profiles from the South Atlantic Ocean (GEOTRACES cruises GA02 and GA10). By comparing Pb isotope data collected on the two cruises, and by modelling the distribution of Pb with an extended optimum multiparameter analysis, we find evidence of vertical transport of anthropogenic Pb pollution due to reversible scavenging. Surface to depth transfer of polluted Pb is aided by high suspended particulate matter loads at the Brazil – Malvinas Confluence and along ∼40°S in the South Atlantic. Overall, our findings caution the use of Pb isotope ratios as ventilation tracers in the South Atlantic and emphasize the importance of particle-seawater interaction for biogeochemical cycles.

Prediction of CO2 content in mid-ocean ridge basalts via a machine learning approach

Abstract One of the primary locations of mafic magma production on Earth is the global mid-ocean ridge system. The basalts erupted along ridges probe the upper mantle and can be used to explore the deep carbon cycle. However, mid-ocean ridge basalts (MORBs) degas heavily during magma ascent. Some incompatible-trace-element–depleted and –enriched MORBs avoid heavy degassing, and show a narrow range of CO2/Ba, which have been used to reconstruct the pre-eruptive CO2 content of primitive MORB. With an increasing amount of data, however, it has become apparent that the CO2/Ba ratios of MORBs vary significantly. We compiled a data set of the geochemical compositions of MORB glasses and melt inclusions that are not degassed significantly and used a supervised machine learning model to accurately predict CO2 contents of individual samples from the concentrations of selected elements. This approach reveals that predicted CO2 contents and CO2/Ba ratios of global MORBs are highly variable, highlighting the significance of mantle heterogeneity, which can be attributed to the interactions with deep-sourced plumes or recycled crust (oceanic crust with or without sediments). Our findings underscore the potential of machine learning as a powerful tool for investigating the intricate interplay between carbon, mantle composition, and Earth's long-term geological processes.

Using linkage to assess coverage of population estimates

ObjectivesThe Demographic Index (DI) comprises of five linked administrative datasets, used for population estimation. Current linkage methods are not ideal to utilise the power of this asset. Using the 2021 England and Wales Census, we developed an innovative composite linkage method to fully utilise the power of the DI. ApproachUsing non-greedy deterministic and probabilistic linkage methods, we linked the DI to the Census at a composite level where we believe links exist – i.e., linking a Census record with a DI cluster (consisting of linked records from the data sources used to make the DI). Next, pairwise linkage of records within these clusters was performed to ensure that every DI record within the cluster was a link. Clerical review of a sample of the full linkage was performed using clerical resolution and search to resolve uncertain and conflicting links and to inform the quality of our linkage. Due to data quality errors in the administrative data, clerical techniques were not always able to determine true match status, so clerical identified ‘uncertain’ links to estimate precision and recall ranges rather than a set estimate. Results A quality analysis to estimate linkage accuracy was conducted and suggests that the linkage has a precision of between 99.3% and 99.7%, and recall between 99.1% and 99.7%. ConclusionWe have linked the ONS’ composite population-level dataset to the 2021 England and Wales Census. The presentation will showcase the methods developed and how we ensured the highest quality possible. *** Please note that this was subbmited in 2022 but the presenters were unable to present due to the UK Queen's death.

Late time cosmic acceleration through parametrization of Hubble parameter in [formula omitted] gravity

This paper explores the rapid expansion of the Universe during its later stages within the context of f(R,Lm) gravity theory. We present a novel parametrization of the Hubble parameter in a manner independent of any specific model and employ it to analyze the Friedmann equations within the FLRW Universe framework. Using a Markov Chain Monte Carlo (MCMC) approach, we derive the model parameters by analyzing a composite dataset comprising 31 Cosmic Chronometers (CC) data points, 26 independent Baryonic Acoustic Oscillations (BAO) measurements, and 1701 data points from Pantheon+SH0ES dataset. The trajectory of the deceleration parameter highlights the shift from deceleration to acceleration in the evolution of the Universe. Additionally, we delve into the dynamics of fundamental cosmological variables for two nonlinear f(R,Lm) models, including energy density, pressure, equation of state (EoS) parameter, and energy conditions.

Chemical and isotopic constraints on fluid origin and genesis of geothermal systems in the Tingri-Tangra Yumco rift, southern Tibetan Plateau

Numerous geothermal systems are hosted by extensional rifts that transect the Himalayas and Lhasa block in the Himalayan–Tibetan orogen. However, the relationships between hydrogeological processes and geothermal fluid circulation in different tectonic units remain unclear. Here, we report an integrated dataset of chemical and isotopic compositions (including major and trace elements, δD, δ18O, and 87Sr/86Sr) of thermal spring water from the Tingri-Tangra Yumco rift to assess their origins and circulation processes. δ18O (− 21.3 to − 17.0‰) and δD (− 166 to − 135‰) values of thermal springs indicate dominant recharge of meteoric waters from areas with elevation of > 6000 m and minor addition of magmatic fluids. Meteoric water could infiltrate to depths of about 1700–2900 m along the faults, whereby it is influenced by geothermal gradient and/or conductive heat transfer of magmatic fluids. The thermal spring waters are mainly Na-HCO3 type and are controlled by dissolution of silicate and carbonate minerals and mixing with deep fluids. The results of chemical and multicomponent geothermometers indicate reservoir temperatures of 115 − 195 ℃, corresponding to a convection heat flux of 3.96 × 105 J/s to 1.78 × 107 J/s from geothermal systems, which are comparable to that of the low-enthalpy geothermal systems in southern Italy. Geochemical modeling is conducted to assess the water–mineral equilibria in the reservoir. Trace elements and 87Sr/86Sr data suggest spatially variable controlling factors for the rift-related geothermal systems: (1) interaction with granitoid and carbonate in the Himalayas; (2) cold groundwater mixing with that leaching from granite and volcanic rocks in the Lhasa block; (3) the input of vapors from magmatic degassing. The geochemistry of thermal springs associated with extensional rift is largely induced by the interaction between fluid and different reservoir rocks in the Himalayas and Lhasa block. Based on these findings, a genetic model is proposed for exploration and development of geothermal resources in the Tingri-Tangra Yumco rift.

Composite fault diagnosis of gearbox based on deep graph residual convolutional network

In gearbox systems, a composite fault diagnosis resulting from mutual interference among different components poses a significant challenge. The traditional composite fault diagnosis methods based on conventional signal analyses and feature extractions often suffer from low sensitivity to fault characteristics and difficulty in effectively identifying composite faults. On the other hand, composite fault diagnosis research via deep learning and data-driven approaches typically faces issues such as incomplete training datasets and insufficient exploration of feature correlation information, leading to an underutilization of the fault information. Therefore, this paper proposes a deep graph residual convolutional neural network (DGRCN) based on feature correlation mining for composite fault diagnosis in gearboxes. First, Pearson correlation coefficients are utilized to explore the relationships among features in the traditional feature set, transforming these relationships into a graph-structured feature set. Next, a deep graph residual convolutional network is constructed by integrating deep graph structures into a residual framework. This network globally extracts composite fault subgraph features and explores local feature correlations. Finally, the model is trained via various composite fault datasets under complex working conditions, achieving the diagnosis and identification of composite faults under the constraint of limited samples. The experimental results demonstrate that the proposed method significantly improves composite fault diagnosis accuracy, outperforming commonly used methods in this field.

Research of real-time corn yield monitoring system with DNN-based prediction model.

The real-time monitoring of corn yield by a combine harvester is a critical data source for constructing the yield histogram, which significantly benefits precision management and decision-making in modern precision agriculture. While widely used, the current photoelectric sensor-based yield monitoring method has limitations. It detects the corn height on each scraper and calculates the yield through a geometric formula. However, it neglects the noticeable difference in the corn stacking patterns affected by factors such as feeding volume, terrain, and driving speed. This oversight often results in low accuracy and poor stability in the prediction of corn yield, highlighting the need for a more advanced approach. To resolve this, we employ EDEM discrete element simulation to demonstrate the large difference of corn stacking patterns on the scraper of the elevator corresponding to feeding volume. Then, we develop a real-time monitoring system on our self-developed double elevator testing rig for carrying out a composite dataset for training three machine learning algorithm-based models, namely Deep Neural Networks (DNN), Gradient Boosting Machine (GBM), and Random Forest (RF). Importantly, these models have undergone rigorous validation under various feeding volumes, ensuring their robustness and reliability. The auxiliary elevator speed is meticulously set at 150r/min, 225r/min, and 450r/min, providing a comprehensive performance assessment. The results denote that the DNN model performs best and is stable, with a coefficient of determination (R2) of 0.998, root mean square error (RMSE) of 0.526, and mean absolute error (MAE) of 0.425. The paper also performs field experiments to test the proposed three prediction models and the system. The results also denote the DNN-based prediction model's best performance for the lowest relative error of 2.29% and the highest average accuracy of 97.85%. Consequently, the proposed real-time corn yield monitoring system achieves high accuracy and reliability for the combine harvester applications.

Chemometrics methods, sensory evaluation and intelligent sensory technologies combined with GAN-based integrated deep-learning framework to discriminate salted goose breeds

The authenticity of salted goose products is concerning for consumers. This study describes an integrated deep-learning framework based on a generative adversarial network and combines it with data from headspace solid phase microextraction/gas chromatography–mass spectrometry, headspace gas chromatography-ion mobility spectrometry, E-nose, E-tongue, quantitative descriptive analysis, and free amino acid and 5′-nucleotide analyses to achieve reliable discrimination of four salted goose breeds. Volatile and non-volatile compounds and sensory characteristics and intelligent sensory characteristics were analyzed. A preliminary composite dataset was generated in InfoGAN and provided to several base classifiers for training. The prediction results were fused via dynamic weighting to produce an integrated model prediction. An ablation study demonstrated that ensemble learning was indispensable to improving the generalization capability of the model. The framework has an accuracy of 95%, a root mean square error (RMSE) of 0.080, a precision of 0.9450, a recall of 0.9470, and an F1-score of 0.9460.