Froth texture extraction with deep learning
| dc.contributor.advisor | Auret, Lidia | en_ZA |
| dc.contributor.advisor | Aldrich, C. | en_ZA |
| dc.contributor.advisor | Herbst, B. M. | en_ZA |
| dc.contributor.author | Horn, Zander Christo | en_ZA |
| dc.contributor.other | Stellenbosch University. Faculty of Engineering. Dept. of Process Engineering. | en_ZA |
| dc.date.accessioned | 2018-02-22T11:21:16Z | |
| dc.date.accessioned | 2018-04-09T07:04:03Z | |
| dc.date.available | 2018-02-22T11:21:16Z | |
| dc.date.available | 2018-04-09T07:04:03Z | |
| dc.date.issued | 2018-03 | |
| dc.description | Thesis (MEng)--Stellenbosch University, 2018. | en_ZA |
| dc.description.abstract | ENGLISH SUMMARY: Soft-sensors are of interest in mineral processing and can replace slower or more expensive sensors by using existing process sensors. Sensing process information from images has been demonstrated successfully, but performance is dependent on feature extractors used. Textural features which utilise spatial relationships within images are preferred due to greater resilience to changing imaging and process conditions. Traditional texture feature extractors require iterative design and are sensitive to changes in imaging conditions. They may have many hyperparameters, leading to slow optimisation. Robust and accurate sensing is a key requirement for mineral processing, making current methods of limited potential under realistic industrial conditions. A platinum froth flotation case study was used to compare traditional texture feature extractors with a proposed deep learning feature extractor: convolutional neural networks (CNNs). Deep learning applies artificial neural networks with many hidden layers and specialised architectures for powerful correlative performance through automated training. All information of the input data structure is determined inherently in training with only a limited number of hyperparameters. However, deep learning methods risk overfitting with small datasets, which must be mitigated. A CNN classifier and a framework for unbiased comparison between feature extractors were developed for predicting high to low grade classes of platinum in flotation froth images. CNNs can perform all the functions of a soft-sensor, but this may bias performance comparison. Instead, features were extracted from hidden layers in CNNs and fed into a traditional soft-sensor. This ensured performance measurements were unbiased across all feature extractors. With a full factorial experiment, the following CNN hyperparameters were evaluated: batch size, number of convolutional filters, and convolutional filter size. Accuracy of grade classification was used to score feature extractors. These reference texture feature extractors were compared to CNNs: Local Binary Patterns, Grey-Level Co-occurrence Matrices, and Wavelets. The impact of spectral features (bulk image features such as average colour) was also evaluated, as CNNs can also use spectral image properties to create features, unlike traditional texture extractors. Extractors were tested with input resolutions from 16x16 to 128x128 with two soft-sensor models: Linear Discriminant Analysis, and k-Nearest Neighbour classifiers. Optimal grade classification accuracies were: CNN – 96.5%, LBP – 100%, GLCM – 73.7%, Wavelets – 98.3%, and Spectral – 98.4% Training CNNs to extract features was successful with robust results regardless of hyperparameters selected. The only statistically significant differences obtained during training were that smaller batch size and smaller input resolution gave superior training performance. Results were found to be reproducible for all models. Analysing learned CNN features indicated both textural and spectral features were utilised. Overall results showed spectral features gave good classification performance, potentially adding to CNN performance. CNNs showed comparable performance to other texture feature extractors at all resolutions. This proof of concept implementation shows promise for deep learning methods in mineral processing applications. The resilience of CNNs to changes in imaging and process conditions could not be evaluated due to limited data in the case study. Future work with deep learning methods, while promising, will require larger datasets which are more representative of a variety of process conditions. | en_ZA |
| dc.description.abstract | AFRIKAANSE OPSOMMING: Inferensiële waarneming is van belang in die mineraalverwerkingsveld. Dit kan stadiger of duurder aanlynsensors vervang met bestaande prosesveranderlike sensors. Waarneming van proses inligting uit beelde is suksesvol gedemonstreer, maar werkverrigting is afhanklik van die kenmerk-ekstraksie metode. Teksturele eienskappe wat ruimtelike verhoudinge binne beelde gebruik geniet voorkeur as gevolg van hul veerkragtigheid teenoor veranderings in beeldopname en prosesomstandighede. Tradisionele tekstuurkenmerk-ekstraksiemetodes benodig iteratiewe ontwerp en is sensitief vir veranderinge in beeldopnameomstandighede. Tekstuurkenmerk-ekstraksiemetodes kan baie hiperparameters hê wat stadige optimalisering veroorsaak. Robuuste, akkurate waarneming is 'n belangrike vereiste vir minerale verwerking. Huidige tegnieke van tekstuur-ekstraksie het beperkte potensiaal onder realistiese industriële toestande. ’n Platinumflottasieskuim gevallestudie is gebruik om tradisionele tekstuurkenmerk-ekstraksie te vergelyk met ’n voorgestelde diep-leer kenmerk-ekstraksiemetode: konvolusionele neurale netwerke (KNNs). Diep-leer pas kunsmatige neurale netwerke, met versteekte lae en gespesialiseerde argitektuur, toe. Dit maak voorsiening vir ‘n kragtige korrelatiewe prestasie en geoutomatiseerde opleiding. Alle inligting rakende die insetdata struktuur word inherent bepaal tydens opleiding met slegs ’n beperkte hoeveelheid hiperparameters. Klein datastelle kan egter lei na oormatige passing in diep-leermetodes. Dié risiko moet versag word. ’n KNN-klassifiseerder en ‘n raamwerk vir onbevooroordeelde vergelyking tussen kenmerk-ekstraksiemetodes is hier ontwikkel om lae tot hoë platinumgraadklasse van flotasieskuimbeelde te voorspel. KNNs kan dieselfde funksies as inferensiële sensors verrig, maar dit kan sydig wees in prestasiemetings. Om dit te voorkom is kenmerke van versteekte lae in die KNNs onttrek en as insette in die tradisionele inferensiële sensor gebruik. Dit het verseker dat kenmerk-ekstraksie nie sydig was as gevolg van die korrelerende vermoëns van KNNs nie. Die volgende KNN-hiperparameters is evalueer deur ’n vol-faktor eksperiment: bondelgrootte, toenemende of konstante aantal konvolusiefilters, en konvolusiefiltergrootte. Tekstuurkenmerk-ekstraksiemetodes is gegradeer volgens die klassifikasie-akkuraatheid van platinumgraad. Hierdie tradisionele tekstuurkenmerk-ekstraksiemetodes is vergelyk met KNNs: Lokale Binêre Patrone (LBP), Grysskaalmede-aanwesigheidsmatrikse (GSMMs), en Golfie-transformasies. Die impak van spektrale eienskappe (byvoorbeeld massa eienskappe soos gemiddelde kleur) is geëvalueer, aangesien KNNs spektrale beeldkenmerke ook kan toepas om eienskappe te skep, anders as tradisionele tekstuur-ekstraksiemetodes. Die kenmerk-ekstraksiemetodes is getoets in 'n reeks insetresolusies van 16x16 tot 128x128 met een van twee sagte-sensor modelle: Lineêre Diskriminant Analise (LDA) en k-Naaste Buurman (k-NB) klassifiseerders. Die beste klassifikasie-akkuraatheid vir elke metode was as volg: KNN – 96.5%, LBP – 100%, GSMM – 73.7%, Golfie-transformasies – 98.3%, en Spektraal– 98.4% Die opleiding van KNNs vir kenmerk-ekstraksie was suksesvol, met robuuste resultate ongeag die gekose hiperparameters. Die enigste statisties beduidende resultate wat behaal is, is dat kleiner bondelgrootte en kleiner insetresolusie 'n beter opleidingsprestasie het. Herhalingstoetsing het bevind dat opleidingsresultate reproduseerbaar was. Ontleding van geleerde KNN-kenmerke het aangedui dat beide teksturele en spektrale kenmerke gebruik is. Oor die algemeen het inferensiële toetse getoon dat spektrale kenmerke tot uitstekende klassifikasie-prestasie gelei het, wat moontlik die KNN se prestasie verbeter. KNNs het vergelykbare prestasie getoon met ander tekstuurfunksie-ekstrakte by alle beeldresolusies, en het klassifiekasie-uitslae geproduseer wat geskik is vir beheer- en moniteringdoeleindes. Met hierdie bewys-van-konsep implementering, toon diep-leermetodes belofte vir gebruik in minerale verwerkingsprobleme. Die veerkragtigheid van KNNs teen verandering in beeld- en prosesomstandighede kon egter nie geëvalueer word nie, as gevolg van beperkte data in die gevallestudie. Verdere werk met diep-leermetodes, terwyl belowend, sal groter datastelle benodig wat meer verteenwoordigend is van ‘n verskeidenheid prosesomstandighede. | af_ZA |
| dc.format.extent | xv, 101 pages ; illustrations | en_ZA |
| dc.identifier.uri | http://hdl.handle.net/10019.1/103622 | |
| dc.language.iso | en_ZA | en_ZA |
| dc.publisher | Stellenbosch : Stellenbosch University | en_ZA |
| dc.rights.holder | Stellenbosch University | en_ZA |
| dc.subject | Flotation | en_ZA |
| dc.subject | Ore dressing | en_ZA |
| dc.subject | Imaging systems in chemistry | en_ZA |
| dc.subject | Convolutional neural networks | en_ZA |
| dc.subject | Machine learning | en_ZA |
| dc.subject | Texture mapping | en_ZA |
| dc.subject | UCTD | |
| dc.title | Froth texture extraction with deep learning | en_ZA |
| dc.type | Thesis | en_ZA |