Browsing by Author "Mishra, Puneet"
Now showing 1 - 9 of 9
Results Per Page
Sort Options
- An automated deep learning pipeline based on advanced optimisations for leveraging spectral classification modellingPublication . Passos, Dário; Mishra, PuneetNa modelagem de deep learning (DL) para dados espectrais, um grande desafio está relacionado à escolha da arquitetura de rede DL e à seleção dos melhores hiperparmetros. Muitas vezes, pequenas mudanças na arquitetura neural ou seu hiperparômetro podem ter uma influência direta no desempenho do modelo, tornando sua robustez questionável. Para lidar com isso, este estudo apresenta uma modelagem automatizada de aprendizagem profunda baseada em técnicas avançadas de otimização envolvendo hyperband e otimização bayesiana, para encontrar automaticamente a arquitetura neural ideal e seus hiperparmetros para alcançar modelos robustos de DL. A otimização requer uma arquitetura neural base para ser inicializada, no entanto, mais tarde, ajusta automaticamente a arquitetura neural e os hiperparmetros para alcançar o modelo ideal. Além disso, para apoiar a interpretação dos modelos DL, foi implementado um esquema de pesagem de comprimento de onda baseado no mapeamento de ativação de classe ponderada por gradiente (Grad-CAM). O potencial da abordagem foi mostrado em um caso real de classificação da variedade de trigo com dados espectrais quase infravermelhos. O desempenho da classificação foi comparado com o relatado anteriormente no mesmo conjunto de dados com diferentes abordagens DL e quimiométrica. Os resultados mostraram que, com a abordagem proposta, foi alcançada uma precisão de classificação de 94,9%, melhor do que a melhor precisão relatada no mesmo conjunto de dados, ou seja, 93%. Além disso, o melhor desempenho foi obtido com uma arquitetura neural mais simples em comparação com o que foi usado em estudos anteriores. O deep learning automatizado baseado na otimização avançada pode suportar a modelagem DL de dados espectrais.
- Deep calibration transfer: transferring deep learning models between infrared spectroscopy instrumentsPublication . Mishra, Puneet; Passos, DárioCalibration transfer (CT) is required when a model developed on one instrument needs to be transferred and used on a new instrument. Several methods are available in the chemometrics domain to transfer the multivariate calibrations developed using modelling techniques such as partial least-square regression. However, recently deep learning (DL) models are gaining popularity to model spectral data. The traditional multivariate CT methods are not suitable to transfer a deep learning model which is based on neural networks architectures. Hence, this study presents the concept of deep calibration transfer (CT) for transferring a DL model made on one instrument onto a new instrument. The deep CT is based on the concept of transfer learning from the DL domain. To show it, two different CT cases are presented. The first case is the CT between benchtop FT-NIR (Fourier Transform Near Infrared) instruments, and the second case is the CT between handheld NIR (Near Infrared) instruments. In both the demonstrated cases, the transfer was performed standard-free i.e., no common standard samples were used to estimate any transfer function. The results showed that with deep CT, the DL models made on one instrument can be easily adapted and transferred to a new instrument. The main benefit of the deep CT is that it is a standard free approach and does not require any standard sample measurements. Such a standard free approach to transfer DL models between instruments can support a widespread sharing of chemometric DL models between the scientific practitioners.
- Deep chemometrics: validation and transfer of a global deep near‐infrared fruit model to use it on a new portable instrumentPublication . Mishra, Puneet; Passos, DárioRecently, a large near-infrared spectroscopy data set for mango fruit quality assessment was made available online. Based on that data, a deep learning (DL) model outperformed all major chemometrics and machine learning approaches. However, in earlier studies, the model validation was limited to the test set from the same data set which was measured with the same instru ment on samples from a similar origin. From a DL perspective, once a model is trained it is expected to generalise well when applied to a new batch of data. Hence, this study aims to validate the generalisability performance of the earlier developed DL model related to DM prediction in mango on a different test set measured in a local laboratory setting, with a different instrument. At first, the performance of the old DL model was presented. Later, a new DL model was crafted to cover the seasonal variability related to fruit harvest season. Finally, a DL model transfer method was performed to use the model on a new instrument. The direct application of the old DL model led to a higher error compared to the PLS model. However, the performance of the DL model was improved drastically when it was tuned to cover the seasonal variability. The updated DL model performed the best compared to the implementation of a new PLS model or updating the existing PLS model. A final root-mean-square error prediction (RMSEP) of 0.518% was reached. This result supports that, in the availability of large data sets, DL modelling can outperform chemometrics approaches.
- Deep learning for near-infrared spectral data modelling: Hypes and benefitsPublication . Mishra, Puneet; Passos, Dário; Marini, Federico; Xu, Junli; Amigo, Jose M.; Gowen, Aoife A.; Jansen, Jeroen J.; Biancolillo, Alessandra; Roger, Jean Michel; Rutledge, Douglas N.; Nordon, AlisonDeep learning (DL) is emerging as a new tool to model spectral data acquired in analytical experiments. Although applications are flourishing, there is also much interest currently observed in the scientific community on the use of DL for spectral data modelling. This paper provides a critical and compre-hensive review of the major benefits, and potential pitfalls, of current DL tecnhiques used for spectral data modelling. Although this work focuses on DL for the modelling of near-infrared (NIR) spectral data in chemometric tasks, many of the findings can be expanded to cover other spectral techniques. Finally, empirical guidelines on the best practice for the use of DL for the modelling of spectral data are provided.
- Deep multiblock predictive modelling using parallel input convolutional neural networksPublication . Mishra, Puneet; Passos, DárioIn the domain of chemometrics, multiblock data analysis is widely performed for exploring or fusing data from multiple sources. Commonly used methods for multiblock predictive analysis are the extensions of latent space modelling approaches. However, recently, deep learning (DL) approaches such as convolutional neural networks (CNNs) have outperformed the single block traditional latent space modelling chemometric approaches such as partial least-square (PLS) regression. The CNNs based DL modelling can also be performed to simultaneously deal with the multiblock data but was never explored until this study. Hence, this study for the first time presents the concept of parallel input CNNs based DL modelling for multiblock predictive chemometric analysis. The parallel input CNNs based DL modelling utilizes individual convolutional layers for each data block to extract key features that are later combined and passed to a regression module composed of fully connected layers. The method was tested on a real visible and near-infrared (Vis-NIR) large data set related to dry matter prediction in mango fruit. To have the multiblock data, the visible (Vis) and near-infrared (NIR) parts were treated as two separate blocks. The performance of the parallel input CNN was compared with the traditional single block CNNs based DL modelling, as well as with a commonly used multiblock chemometric approach called sequentially orthogonalized partial least-square (SO-PLS) regression. The results showed that the proposed parallel input CNNs based deep multiblock analysis outperformed the single block CNNs based DL modelling and the SO-PLS regression analysis. The root means squared errors of prediction obtained with deep multiblock analysis was 0.818%, relatively lower by 4 and 20% than single block CNNs and SO-PLS regression, respectively. Furthermore, the deep multiblock approach attained ~3% lower RMSE compared to the best known on the mango data set used for this study. The deep multiblock analysis approach based on parallel input CNNs could be considered as a useful tool for fusing data from multiple sources.
- Multi-output 1-dimensional convolutional neural networks for simultaneous prediction of different traits of fruit based on near-infrared spectroscopyPublication . Mishra, Puneet; Passos, DárioIn spectral data predictive modelling of fresh fruit, often the models are calibrated to predict multiple responses. A common method to deal with such a multi-response predictive modelling is the partial least-squares (PLS2) regression. Recently, deep learning (DL) has shown to outperform partial least-squares (PLS) approaches for single fruit traits prediction. The DL can also be adapted to perform multi-response modelling. This study presents an implementation of DL modelling for multi-response prediction for spectral data of fresh fruit. To show this, a real NIR data set related to SSC and MC measurements in pear fruit was used. Since DL models perform better with larger data sets, a data augmentation procedure was performed prior to data modelling. Furthermore, a comparative study was also performed between two of the most used DL architectures for spectral analysis, their multi-output and single-output variants and a classic baseline model using PLS2. A key point to note that all the DL modelling presented in this study is performed using novel automated optimisation tools such as Bayesian optimisation and Hyperband. The results showed that DL models can be easily adapted by changing the output of the fully connected layers to perform multi-response modelling. In comparison to the PLS2, the multi-response DL model showed ~13 % lower root mean squared error (RMSE), showing the ease and superiority of handling multi-response by DL models for spectral calibration.
- Realizing transfer learning for updating deep learning models of spectral data to be used in new scenariosPublication . Mishra, Puneet; Passos, DárioThis study presents the concept of transfer learning (TL) to the chemometrics community for updating DL models related to spectral data, particularly when a pre-trained DL model needs to be used in a scenario having unseen variability. This is the typical situation where classical chemometrics models require some form of re-calibration or update. In TL, the network architecture and weights from the pre-trained DL model are complemented by adding extra fully connected (FC) layers when dealing with the new data. Such extra FC layers are expected to learn the variability of the new scenario and adjust the output of the main architecture. Furthermore, three approaches of TL were compared, first where the weights from the initial model were left untrained and the only the newly added FC layers could be retrained. The second was when the weights from the initial model could be retrained alongside the new FC layers. The third was when the weights from the initial model could be re-trained with no extra FC layers added. The TL was shown using two real cases related to near-infrared spectroscopy i.e., mango fruit analysis and melamine production monitoring. In the case of mango, the model needs to be updated to cover a new seasonal variability for dry matter prediction, while, for the melamine case, the model needs to be updated for the change in the recipe of the production material. The results showed that the proposed TL approaches successfully updated the DL models to new scenarios for both the mango and melamine cases presented. The TL performed better when the weights from the old model were retrained. Furthermore, TL outperformed three recent benchmark approaches to model updating. TL has the potential to make DL models widely useable, sharable, and scalable.
- A synergistic use of chemometrics and deep learning improved the predictive performance of near-infrared spectroscopy models for dry matter prediction in mango fruitPublication . Mishra, Puneet; Passos, DárioThis study provides an innovative approach to improve deep learning (DL) models for spectral data processing with the use of chemometrics knowledge. The technique proposes pre-filtering the outliers using the Hotelling’s T2 and Q statistics obtained with partial least-square (PLS) analysis and spectral data augmentation in the variable domain to improve the predictive performance of DL models made on spectral data. The data augmentation is carried out by stacking the same data pre-processed with several pre-processing techniques such as standard normal variate, 1st derivatives, 2nd derivatives and their combinations. The performance of the approach is demonstrated on a real near-infrared (NIR) data set related to dry matter (DM) prediction in mango fruit. The data set consisted of a total 11,961 spectra and reference DM measurements. The results showed that removing the outliers and augmenting spectral data improved the predictive performance of DL models. Furthermore, this innovative approach not only improved DL models but attained the lowest root mean squared error of prediction (RMSEP) on the mango data set i.e., 0.79% compared to the best known RMSEP of 0.84%. Further, by removing outliers from the test set the RMSEP decreased to 0.75%. Several chemometrics approaches can complement DL models and should be widely explored in conjunction.
- A tutorial on automatic hyperparameter tuning of deep spectral modelling for regression and classification tasksPublication . Passos, Dário; Mishra, PuneetDeep spectral modelling for regression and classification is gaining popularity in the chemometrics domain. A major topic in the deep learning (DL) modelling of spectral data is the choice and optimization of the deep neural network architecture suitable for the specific task of spectral modelling. Although there are several recent research articles already available in the chemometric domain showing advanced approaches to deep spectral modelling, currently, there is a lack of hands-on tutorial articles in this space that supply the non-expert user with practical tools to learn and implement advanced DL optimization methodologies aimed a spectral data. Hence, this tutorial article aims a reducing the gap between the non-expert user of DL in the chemometric community and the implementation of DL models for daily usage. This tutorial supplies a quick introduction to the state-of-the-art deep spectral modelling and related DL concepts and presents a set of methodologies aimed a DL hyperparameters' optimization. To this end, this tutorial shows two practical examples on how to implement and optimize two DL models for spectral regression and classification tasks. The models are implemented in python and Tensorflow and the complete code is supplied in the form of two complementary notebooks.