Limited Sample Set Research Articles

Coarse-to-fine few-shot classification with deep metric learning

Few-shot classification predicts the labels of unseen samples using only a few labeled samples, and employs the samples of the classes disjoint with unseen classes for model training. It faces two primary challenges, i.e., handling sample pairs with different similarity degrees by single classifier, and learning discriminant patterns from very few labeled samples per class. To address them, this work presents a Coarse-to-Fine few-shot classification framework under the guidance of Metric-based Auxiliary learning, abbreviated as CFMA. In particular, it discriminates the image pairs with large difference by capturing global features, and models the similarity relation of the image pairs with small difference by mining the local region of interests. Moreover, CFMA adopts deep metric learning to improve the model adaptivity on the set of limited samples, and generates pseudo labels to dynamically guide the coarse learning in iteration. Empirical studies on several benchmark databases, including mini-ImageNet, tiered-ImageNet, and CUB, demonstrate that our method achieves more promising classification performance compared to many state-of-the-art alternatives.

Semi-supervised medical image segmentation via uncertainty rectified pyramid consistency.

Despite that Convolutional Neural Networks (CNNs) have achieved promising performance in many medical image segmentation tasks, they rely on a large set of labeled images for training, which is expensive and time-consuming to acquire. Semi-supervised learning has shown the potential to alleviate this challenge by learning from a large set of unlabeled images and limited labeled samples. In this work, we present a simple yet efficient consistency regularization approach for semi-supervised medical image segmentation, called Uncertainty Rectified Pyramid Consistency (URPC). Inspired by the pyramid feature network, we chose a pyramid-prediction network that obtains a set of segmentation predictions at different scales. For semi-supervised learning, URPC learns from unlabeled data by minimizing the discrepancy between each of the pyramid predictions and their average. We further present multi-scale uncertainty rectification to boost the pyramid consistency regularization, where the rectification seeks to temper the consistency loss at outlier pixels that may have substantially different predictions than the average, potentially due to upsampling errors or lack of enough labeled data. Experiments on two public datasets and an in-house clinical dataset showed that: 1) URPC can achieve large performance improvement by utilizing unlabeled data and 2) Compared with five existing semi-supervised methods, URPC achieved better or comparable results with a simpler pipeline. Furthermore, we build a semi-supervised medical image segmentation codebase to boost research on this topic: https://github.com/HiLab-git/SSL4MIS.

ASSESSMENT OF CAPILLARY ZONE ELECTROPHORESIS AND SERUM AMYLOID A QUANTITATION IN CLINICALLY NORMAL AND ABNORMAL SOUTHERN WHITE RHINOCEROS (CERATOTHERIUM SIMUM SIMUM) AND SOUTHERN BLACK RHINOCEROS (DICEROS BICORNIS MINOR).

Capillary zone electrophoresis (CZE) and an immunoassay for serum amyloid A (SAA) were used to examine serum samples from clinically normal and abnormal southern white rhinoceros (Ceratotherium simum simum) and southern black rhinoceros (Diceros bicornis minor) under managed care. CZE resolved seven fractions as well as subfractions for α1 globulins. Reference intervals were calculated for white rhinoceros (n = 33) and found to have some differences over previously reported intervals generated using agarose gel electrophoresis (AGE) methods in sera from free-ranging animals. In addition, the coefficient of variation related to fraction quantitation was found to be overlapping or superior to that reported for AGE. No significant differences were observed in CZE measurands and total protein between clinically normal and abnormal rhinoceros. In contrast to CZE, significant differences in SAA levels (P < 0.001) were observed in samples from the white rhinoceros between clinically normal and abnormal animals. In addition, in limited sample sets with repeated measures, SAA provided prognostic value. Future studies should generate more robust reference intervals and delineate the application of both SAA quantitation and CZE in routine health assessments and in prognostication.

Multi-platform quantitation of alpha-synuclein human brain proteoforms suggests disease-specific biochemical profiles of synucleinopathies

Based on immunostainings and biochemical analyses, certain post-translationally modified alpha-synuclein (aSyn) variants, including C-terminally truncated (CTT) and Serine-129 phosphorylated (pSer129) aSyn, are proposed to be involved in the pathogenesis of synucleinopathies such as Parkinson’s disease with (PDD) and without dementia (PD), dementia with Lewy bodies (DLB), and multiple system atrophy (MSA). However, quantitative information about aSyn proteoforms in the human brain in physiological and different pathological conditions is still limited. To address this, we generated sequential biochemical extracts of the substantia nigra, putamen and hippocampus from 28 donors diagnosed and neuropathologically-confirmed with different synucleinopathies (PD/PDD/DLB/MSA), as well as Alzheimer’s disease, progressive supranuclear palsy, and aged normal subjects. The tissue extracts were used to build a reverse phase array including 65 aSyn antibodies for detection. In this multiplex approach, we observed increased immunoreactivity in donors with synucleinopathies compared to controls in detergent-insoluble fractions, mainly for antibodies against CT aSyn and pSer129 aSyn. In addition, despite of the restricted sample size, clustering analysis suggested disease-specific immunoreactivity signatures in patient groups with different synucleinopathies. We aimed to validate and quantify these findings using newly developed immunoassays towards total, 119 and 122 CTT, and pSer129 aSyn. In line with previous studies, we found that synucleinopathies shared an enrichment of post-translationally modified aSyn in detergent-insoluble fractions compared to the other analyzed groups. Our measurements allowed for a quantitative separation of PDD/DLB patients from other synucleinopathies based on higher detergent-insoluble pSer129 aSyn concentrations in the hippocampus. In addition, we found that MSA stood out due to enrichment of CTT and pSer129 aSyn also in the detergent-soluble fraction of the SN and putamen. Together, our results achieved by multiplexed and quantitative immunoassay-based approaches in human brain extracts of a limited sample set point to disease-specific biochemical aSyn proteoform profiles in distinct neurodegenerative disorders.

Effective Plans during Earthquake Emergencies for Medical Treatment

The collapse of buildings caused by earthquakes can lead to a large loss of life and property. Rapid assessment of building damage with remote sensing image data can support emergency rescues. However, current studies indicate that only a limited sample set can usually be obtained from remote sensing images immediately following an earthquake. Consequently, the difficulty in preparing sufficient training samples constrains the generalization of the model in the identification of earthquake-damaged buildings. To produce a deep learning network model with strong generalization, this study adjusted Convolution Neural Network (CNN) models for extracting earthquake detection, damaged building information and compared their performance. A sample dataset of damaged buildings, earthquake detection, was constructed by using multiple disaster images retrieved from our dataset. These results provide a solution to the rapid extraction of earthquake-damaged building information based on a deep learning network model.

A European multicenter evaluation study to investigate the performance on commercially available selective agar plates for the detection of carbapenemase producing Enterobacteriaceae

The European Food Safety Authority (EFSA) advised to prioritize monitoring carbapenemase producing Enterobacteriaceae (CPE) in food producing animals. Therefore, this study evaluated the performance of different commercially available selective agars for the detection of CPE using spiked pig caecal and turkey meat samples and the proposed EFSA cultivation protocol. Eleven laboratories from nine countries received eight samples (four caecal and four meat samples). For each matrix, three samples contained approximately 100 CFU/g CPE, and one sample lacked CPE. After overnight enrichment in buffered peptone water, broths were spread upon Brilliance™ CRE Agar (1), CHROMID® CARBA (2), CHROMagar™ mSuperCARBA™ (3), Chromatic™ CRE (4), CHROMID® OXA-48 (5) and Chromatic™ OXA-48 (6). From plates with suspected growth, one to three colonies were selected for species identification, confirmation of carbapenem resistance and detection of carbapenemase encoding genes, by methods available at participating laboratories. Of the eleven participating laboratories, seven reported species identification, susceptibility tests and genotyping on isolates from all selective agar plates. Agars 2, 4 and 5 performed best, with 100% sensitivity. For agar 3, a sensitivity of 96% was recorded, while agar 1 and 6 performed with 75% and 43% sensitivity, respectively. More background flora was noticed for turkey meat samples than pig caecal samples. Based on this limited set of samples, most commercially available agars performed adequately. The results indicate, however, that OXA-48-like and non-OXA-48-like producers perform very differently, and one should consider which CPE strains are of interest to culture when choosing agar type.

Unsupervised Spatial-Spectral CNN-Based Feature Learning for Hyperspectral Image Classification

The rapid development of remote sensing sensors makes the acquisition, analysis, and application of hyperspectral images (HSIs) more and more extensive. However, the limited sample sets, high-dimensional features, highly correlated bands, and mixing spectral information make the classification of HSIs a great challenge. In this article, an unsupervised multiscale and diverse feature learning (UMsDFL) method is proposed for HSI classification, which deeply considers the spatial–spectral features via convolutional neural networks (CNNs). Specifically, after employing the simple noniterative clustering (SNIC) algorithm with the heuristic calculation of superpixel size, the HSIs are segmented into superpixels for feature learning. The unsupervised network is designed with the convolutional encoder and decoder, the additional clustering branch, and the multilayer feature fusion to enhance the distinguishability of feature learning and the reusability of feature maps. Then, the spatial relationships and object attributes in large- and small-scale contexts are learned collaboratively through the unsupervised network to utilize the complementary multiscale characteristics. Moreover, the diverse features of hyperspectral information and nonsubsampled contourlet transform (NSCT) textures are learned simultaneously via the unsupervised network to alleviate the insufficiency of geometric representation. Finally, the random forest (RF) is adopted as the comprehensive classifier for land cover mapping based on the UMsDFL, and superpixel regularization is adopted to optimize the classification results. A series of experiments are performed on three real-world HSI datasets to demonstrate the effectiveness of our UMsDFL approach. The experimental results show that the proposed UMsDFL can achieve the overall accuracy of 79.23%, 96.49%, and 77.26% for Houston, Pavia, and Dioni datasets, respectively, when there are only five samples per class for training.

Profiling Malt Enzymes Related to Impact on Malt Fermentability, Lautering and Beer Filtration Performance of 94 Commercially Produced Malt Batches

A largely defined series of hydrolytic enzymes active during malting and/or mashing, substantially determine the quality, profitability, and efficiency of the brewing process. These enzymes potentially hydrolyze starch, proteins and cell wall non-starch polysaccharides including β-glucan and arabinoxylan. Commercial malts (94) were assayed for the DP enzymes (limit dextrinase, beta/α-amylase), and NSP hydrolyzing enzymes (β-glucanase, xylanase, arabinofuranosidase, β-glucosidase). The levels of enzyme activity were related to conventional measures of malt quality such as extract, fermentability, protein, KI, DP, friability, wort viscosity, FAN, and β-glucan. These parameters were interrelated with less conventional measures of malt quality including coarse extract and fermentability (modified infusion mash 65 °C), lautering efficiency, the Small-scale Wort ‘I’ Filtration Test (SWIFT), and viscosity. Substantial variation was observed between the malt samples for all enzymes assayed. Australian barley, whether malted in Australia (n = 61) or China (n = 24), was observed to be of comparable quality. A limited set of Canadian barley samples (n = 9) were malted in China and produced malts with somewhat higher levels of extract, AAL, and some enzymes. Remarkably, the level of limit dextrinase was observed to be almost double that from previous investigations. Greater levels of steep water aeration were proposed to explain this dramatic increase. The interrelationships between the enzyme activities and malt quality identified, enable potential selection of novel malt quality parameters that are more predictive of a malt’s brewing performance (efficiency and quality) than current measures to provide a malt quality assessment system based on ‘functional’ malt quality.

Triaromatic dinosteroids – Isomeric distributions and their geochemical significance

The distribution of triaromatic dinosteroids in a range of oils and some source rock bitumens was determined by GC–MS selected ion monitoring and GC–MS-MS in order to investigate possible ambiguity in the observed distribution of isomers, including those related to maturation and alteration processes commonly affecting oil accumulations. The results confirm that the eight possible isomers for the triaromatic dinosteroids and their 28-nor equivalents (triaromatic 23,24-dimethylcholesteroids) in post-Paleozoic samples are consistently represented by six peaks on commonly used GC stationary phases, although the fifth eluting is broadened and probably represents a pair of isomers. The presence of a seventh, early eluting, triaromatic dinosteroid isomer occasionally encountered in m/z 245 mass chromatograms in the literature was not substantiated. Closely co-eluting compounds can give the seemingly misleading impression that one or two triaromatic dinosteroid isomers (including the last eluting) are present at low levels in pre-Mesozoic samples and in younger oils from dominantly higher plant sources, so care is required in inferring source age for such samples. The relative abundance and retention time pattern observed for the triaromatic dinosteroids is duplicated by their 28-nor analogues. Triaromatic dinosteroid isomer distributions vary little throughout the oil window, based on data from Norwegian North Sea oils representing a range of mean maturities. Overall abundance of the series seems to decline in a similar way to other biomarkers towards the end of the oil window. In source rock bitumen corresponding to maturities of ≤0.5% vitrinite reflectance, when diasterenes are still abundant, the second eluting triaromatic dinosteroid peak was enhanced and the fifth depleted in the limited set of samples studied (Late Jurassic marine shales from the Norwegian continental shelf). Both dinosteroid series appear particularly resistant towards biodegradation and metamorphism, with no obvious change in isomeric distributions compared to unaltered oil. As a consequence, their relative abundances, as represented by age parameters that ratio each series to their 24-ethyl counterparts, increase when alteration is severe. Lacustrine oils can be difficult to date because post-Paleozoic samples do not always contain detectable dinosteroids and, when they are present, they can vary significantly in abundance. Unambiguous age resolution at the Period level was not possible for the Triassic-Tertiary oils studied, neither could depositional environments be distinguished with confidence.

Population Pharmacokinetics Profiling to Tailor Prophylaxis in Hemophilia a Patients: A Pilot Feasibility Study

Population Pharmacokinetics Profiling to Tailor Prophylaxis in Hemophilia a Patients: A Pilot Feasibility Study

Metrics and Continuity in Reinforcement Learning

In most practical applications of reinforcement learning, it is untenable to maintain direct estimates for individual states; in continuous-state systems, it is impossible. Instead, researchers often leverage {\em state similarity} (whether explicitly or implicitly) to build models that can generalize well from a limited set of samples. The notion of state similarity used, and the neighbourhoods and topologies they induce, is thus of crucial importance, as it will directly affect the performance of the algorithms. Indeed, a number of recent works introduce algorithms assuming the existence of "well-behaved" neighbourhoods, but leave the full specification of such topologies for future work. In this paper we introduce a unified formalism for defining these topologies through the lens of metrics. We establish a hierarchy amongst these metrics and demonstrate their theoretical implications on the Markov Decision Process specifying the reinforcement learning problem. We complement our theoretical results with empirical evaluations showcasing the differences between the metrics considered.

Revisiting the unrepaired unilateral cleft lip and nasal deformity using 3D surface image analysis: A data-driven model and its implications

Current descriptions of the unilateral cleft lip and nasal deformity (uCLND) are based upon limited sample sets and subjective observations. While those descriptions are inconsistent and contradictory, theoretical models, including Hogan's "tilted tripod" and Fisher's "nasal arch forms", have never been tested. Given that favorable outcomes of treatment remain elusive, detailed study of the deformity is critical in devising better treatments. The purpose of this study was to develop a data-driven three-dimensional (3D) model of uCLND that spans the spectrum of presentation and involves a pervasive underlying mechanism. We studied 3D images of 100 infants with unrepaired cleft lip at 6 months of age. Objective assessment included the landmark positions, anthropometric dimensions, and shape-based measures. Cleft severity was stratified by the lateral displacement of subnasale, so that a model could be developed using linear regression. With progressive deviation of subnasale, the non-cleft alar base moved lateral, the cleft alar base was left posterior, and the nasal dorsum followed the caudal septum (deviating towards the non-cleft side). The "twist" resulted in opposing cleft alar dome collapse, non-cleft alar ring constriction (the non-cleft nasal sill narrowed, lateral genu rose, and alar-cheek junction became more acute), and displacement of the philtrum from midline. Our study not only supports theoretical models of uCLND but also clarifies vectors of change and reveals significant non-cleft side alterations. On the basis of our findings, the objectives of treatment should involve centralization of the columella and philtrum and rebalancing the nose by untwisting the orthogonal displacement of the alar bases.

Environmental influence on the distribution of polychaete families and feeding guilds in benthic communities of the Grand Banks and Flemish Cap (NW Atlantic)

Many biodiversity patterns across the globe can be partially explained by energetics and habitat structure, including in the deep sea. Because of difficulties in logistics, studies focusing on deep-sea benthic systems often have limited sample sets that may be far apart in space. Here, we present analyses based on a well-sampled region, the northwest Atlantic, allowing us to relate patterns of polychaete community structure (using taxonomic families and feeding guilds) against well-resolved environmental gradients. We show that community structure is heterogenous in this area. We observe three major groups within polychaete communities which also show differences in feeding guilds. Several environmental gradients, including long-term variables, explained the differences between the groups. Both energetics and habitat structure variables were important, while past fishing intensity had a weak effect, although there was support for a difference in community composition for communities above and below the historical fishing limit of 1200 m. The effect of long-term environmental factors may indicate that the deep sea is not immune to climate change, and these effects must continue to be considered in future management of exploitation of this ecosystem.

Transferability of Convolutional Neural Network Models for Identifying Damaged Buildings Due to Earthquake

The collapse of buildings caused by earthquakes can lead to a large loss of life and property. Rapid assessment of building damage with remote sensing image data can support emergency rescues. However, current studies indicate that only a limited sample set can usually be obtained from remote sensing images immediately following an earthquake. Consequently, the difficulty in preparing sufficient training samples constrains the generalization of the model in the identification of earthquake-damaged buildings. To produce a deep learning network model with strong generalization, this study adjusted four Convolutional Neural Network (CNN) models for extracting damaged building information and compared their performance. A sample dataset of damaged buildings was constructed by using multiple disaster images retrieved from the xBD dataset. Using satellite and aerial remote sensing data obtained after the 2008 Wenchuan earthquake, we examined the geographic and data transferability of the deep network model pre-trained on the xBD dataset. The result shows that the network model pre-trained with samples generated from multiple disaster remote sensing images can extract accurately collapsed building information from satellite remote sensing data. Among the adjusted CNN models tested in the study, the adjusted DenseNet121 was the most robust. Transfer learning solved the problem of poor adaptability of the network model to remote sensing images acquired by different platforms and could identify disaster-damaged buildings properly. These results provide a solution to the rapid extraction of earthquake-damaged building information based on a deep learning network model.

Screening of diseases in wild exotic birds on Tahiti Island – implications for French Polynesian conservation

In order to identify potential infectious disease threats to the native avifauna of French Polynesia, an evaluation was performed on the health status of four wild non-native species of birds on the island of Tahiti: common myna, red-vented bulbul, rock dove, and zebra dove. From six locations, a large sample set (151–349 individuals) was tested for several viruses and bacteria, and a small sample set (22–40 birds), because of its proximity to the last remaining population of the critically endangered Tahiti monarch, was checked for more pathogens. Disease-specific screening methods were used. None of the following viruses were found: Newcastle disease virus, avian influenza virus, West Nile virus in 159, 189 and 204 sera; 349 birds examined for poxvirus lesion; avian metapneumovirus and avian adenovirus in 38 and 38 sera; avian polyomavirus in 28 cloacal swabs. The prevalence of bacteria and avian malaria was: Salmonella Heidelberg (5% from 21×10 pooled samples of intestinal contents), Chlamydia spp. (8% on 196 cloacal swabs) including Chlamydia psittaci (3%), Plasmodium relictum – haplotype GRW04 (2% on 205 DNA), Haemoproteus spp. (25% on 205 DNA). In the limited sample set, Klebsiella pneumoniae, Bordetella avium and Riemerella columbina were isolated with a prevalence of 3% each in 40 tracheal swabs. The potential role of introduced birds as vectors of zoonosis in French Polynesia and the crucial finding of Plasmodium relictum with several ubiquitous and dangerous pathogens on Tahiti Island should be given the appropriate attention by local authorities and conservationists.

Machine Learning Regression Techniques for the Modeling of Complex Systems: An Overview

Recently, machine learning (ML) techniques have gained widespread diffusion, since they have been successfully applied in several research fields. This paper investigates the effectiveness of advanced ML regressions in two EMC applications. Specifically, support vector machine, least-squares support vector machine and Gaussian process regressions are adopted to construct accurate and fast-to-evaluate surrogate models able to predict the output variable of interest as a function of the system parameters. The resulting surrogates, built from a limited set of training samples, can be suitably adopted for both uncertainty quantification and optimization purposes. The accuracy and the key features of each of the considered machine learning techniques are investigated by comparing their predictions with the ones provided by either circuital simulations or measurements.

DeepSIC: Deep Soft Interference Cancellation for Multiuser MIMO Detection

Digital receivers are required to recover the transmitted symbols from their observed channel output. In multiuser multiple-input multiple-output (MIMO) setups, where multiple symbols are simultaneously transmitted, accurate symbol detection is challenging. A family of algorithms capable of reliably recovering multiple symbols is based on interference cancellation. However, these methods assume that the channel is linear, a model which does not reflect many relevant channels, as well as require accurate channel state information (CSI), which may not be available. In this work we propose a multiuser MIMO receiver which learns to jointly detect in a data-driven fashion, without assuming a specific channel model or requiring CSI. In particular, we propose a data-driven implementation of the iterative soft interference cancellation (SIC) algorithm which we refer to as DeepSIC. The resulting symbol detector is based on integrating dedicated machine-learning methods into the iterative SIC algorithm. DeepSIC learns to carry out joint detection from a limited set of training samples without requiring the channel to be linear and its parameters to be known. Our numerical evaluations demonstrate that for linear channels with full CSI, DeepSIC approaches the performance of iterative SIC, which is comparable to the optimal performance, and outperforms previously proposed learning-based MIMO receivers. Furthermore, in the presence of CSI uncertainty, DeepSIC significantly outperforms model-based approaches. Finally, we show that DeepSIC accurately detects symbols in non-linear channels, where conventional iterative SIC fails even when accurate CSI is available.

Analysis of expression, epigenetic, and genetic changes of HNF1B in 130 kidney tumours

Hepatocyte nuclear factor 1 beta (HNF1B) is a transcription factor which plays a crucial role in nephronogenesis, and its germline mutations have been associated with kidney developmental disorders. However, the effects of HNF1B somatic exonic mutations and its role in the pathogenesis of kidney tumours has not yet been elucidated. Depending on the type of the tumour HNF1B may act as a tumour suppressor or oncogene, although the exact mechanism by which HNF1B participates in the process of cancerogenesis is unknown. Using an immunohistochemical approach, and methylation and mutation analysis, we have investigated the expression, epigenetic, and genetic changes of HNF1B in 130 cases of renal tumours (121 renal cell carcinomas, 9 oncocytomas). In the subset of clear cell renal cell carcinoma (ccRCC), decreased HNF1B expression was associated with a higher tumour grade and higher T stage. The mutation analysis revealed no mutations in the analysed samples. Promoter methylation was detected in two ccRCCs and one oncocytoma. The results of our work on a limited sample set suggest that while in papillary renal cell carcinoma HNF1B functions as an oncogene, in ccRCC and chRCC it may act in a tumour suppressive fashion.

Decreased plasma α-synuclein in idiopathic Parkinson’s disease patients after adjusting hemolysis factor

Lewy bodies are pathological hallmarks for the diagnosis of Parkinson’s disease, where the core components are composed of aggregated forms of α-synuclein (α-syn). Although α-syn has been investigated as a potential biomarker for PD, its usage has been limited and still remains controversial. An accurate enzyme-linked immunosorbent assay (ELISA) was developed for the detection of α-syn in plasma. The hemolysis score was calculated to eliminate the additive α-syn levels from RBCs. Human plasma samples were collected in heparinized blood tubes from idiopathic Parkinson disease (IPD) and healthy control (HC). Hemolysis score had a strong correlation with the level of plasma α-syn. From the limited set of samples in this preliminary study, decreased α-syn concentrations were observed in patients with IPD in comparison to HC after adjusting for hemolysis factor. Similar results with a commercial ELISA kit were found for measuring α-syn from the same set of samples with lower correlation and the reduced accuracy of diagnosis than the current study. The adjustments for the hemolysis factor would be indispensable, supporting plasma α-syn as a potential surrogate biomarker for distinguishing IPD from HC.

Analysis of Raman spectra for the verification of Australian grass‐ and grain‐fed beef using principal component analysis and partial least square models

AbstractThe diet of cattle has been shown to alter both subcutaneous fat and muscle fatty acid composition, and these differences cause a change in the Raman spectra of the fat due to the different chemical bonds associated with these biochemical changes. Because biochemical changes only lead to subtle changes in the Raman spectra, statistical methods based upon chemometric modelling are necessary to extract indicative information on the type of feed consumed. This investigation undertook a feasibility study using a limited sample set derived from two producers in one location to ascertain the potential for Raman spectroscopy to accurately discriminate grain‐fed cattle from grass‐fed cattle and investigated the usefulness of principal component analysis (PCA) and partial least squares discriminate analysis (PLS‐DA) to best discriminate between feeding groups using the information from the Raman spectra. The first two principal components accounted for 83% of the variation in the spectra and demonstrated discrimination of samples by feed type. PLS‐DA resulted in a model that was able to accurately predict grain‐ and grass‐fed carcases with a misclassification rate of 3.5%. Changes in beef cattle diets lead to subtle changes in subcutaneous fat; using Raman spectroscopy with chemometric modelling, these changes can be identified and used to identify the production system of beef products. This study successfully uses Raman spectroscopy as an automated, nondestructive and rapid technique in the range of 600–2000 cm−1 in combination with pattern recognition of unsupervised (PCA) and supervised (PLS‐DA) techniques to classify the production system of cattle between grain and grass fed.