Evaluating infrared thermographic imaging technology in the detection and quantification of tick burdens in cattle

Thesis (MScAgric)--Stellenbosch University, 2022.

ENGLISH ABSTRACT: Ticks and tick-borne diseases (TTBDs) are one of the biggest economic threats to livestock production systems, especially in the sub- and tropical regions of the world. The use of acaricides is currently regarded as the most effective method to control TTBDs. This method is, however, failing to eradicate the problem of TTBDs in cattle. The use of tick tolerant cattle breeds is therefore encouraged. Notably, it remains a challenge to accurately quantify tick infestation levels on cattle to effectively select for tick resistance using conventional tick counting method. The aim of this study was to detect and quantify tick burdens in cattle using infrared thermographic imaging technology (IRT) as an alternative for conventional visual methods. This study evaluated the influence of the interaction of season and vegetation on tick infestation in beef cattle. The convolutional neural networks (CNNs) were applied on the algorithms for detection and quantification of ticks with the use of IRT. Tick counts were conducted once a month under natural challenge over a six-month period (February to July) on 19 Bonsmara and 36 Nguni cattle located at ARC Roodeplaat and Loskop farms. The experiment was conducted throughout both warmer climates and cooler climates. Rectal temperatures were recorded throughout the two seasons from cattle through February 2021 until July 2021. Five tick species were identified, namely R. appendiculatus (7%), R. (Boophilus) species (18%), Hyalomma species (19%), Amblyomma hebraeum (39%) and R. evertsi evertsi (17%). Negative binomial analysis was used to establish the effect of location and season on the overall tick count. The tick counts were then analysed in Rstudio using the fixed effect of General Addative models (GAM). An Akaike Information Criterion (AIC) was used to evaluate the best fit model for analysis. There was a positive correlation (P<0.05) between tick infestation and rectal temperature. The average rectal temperature of 38.83°C and 38.59°C was recorded in ARC Loskop and Roodeplaat Research farms, respectively. Thermographic images of both engorged & unfed females and males ticks were taken from cattle from February 2021 until July 2021. The deep learning models with architectures: “ConvNet '' and “MobileNet” were trained on a dataset of 1124 “thermograms” to detect ticks on cattle. ConvNet model achieved a training and validation accuracy of ~90 and 60%, respectively. Whereas MobileNet scored a training and validation accuracy of ~95 and 75%, respectively. Finally, deep learning was successfully used to detect ticks on cattle using pre-trained convolutional neural networks (CNNs).

AFRIKAANSE OPSOMMING: Bosluise en bosluisoorgedraagde siektes (BBOS) is een van die grootste ekonomiese bedreigings vir veeproduksiestelsels, veral in die sub- en tropiese streke van die wêreld. Die gebruik van mytdoders word tans beskou as die doeltreffendste metode om BBOS te beheer. Hierdie metode slaag egter nie daarin om die probleem van BBOS by beeste uit te roei nie. Die gebruik van bosluisverdraagsame beesrasse word dus aangemoedig. Dit bly veral 'n uitdaging om bosluisbesmettingsvlakke op beeste akkuraat te kwantifiseer om effektief te selekteer vir bosluisweerstand deur gebruik te maak van konvensionele bosluis telmetodes. Die breë doelwit van hierdie studie was om bosluislaste by beeste op te spoor en te kwantifiseer deur gebruik te maak van infrarooi termografiese beeldingstegnologie (IRT) as 'n alternatief vir konvensionele visuele metodes. Hierdie studie het die invloed van die interaksie van seisoen en plantegroei op bosluisbesmetting by vleisbeeste geëvalueer. Die konvolusionele neurale netwerke (KNN's) is toegepas op die algoritmes vir opsporing en kwantifisering van bosluise met die gebruik van IRT. Bosluistellings is een keer per maand onder natuurlike uitdaging oor 'n tydperk van ses maande (Februarie tot Julie) gedoen op 19 Bonsmara- en 36 Nguni- beeste wat op LNR Roodeplaat en Loskop plase geleë is. Die eksperiment is deur beide warmer klimate en koeler klimate uitgevoer. Rektale temperature van beeste is gedurende die twee seisoene aangeteken vanaf Februarie 2021 tot Julie 2021. Vyf bosluisspesies is geïdentifiseer, naamlik R. appendiculatus (7%), R. (Boophilus) spesies (18%), Hyalomma spesies (19%), Amblyomma hebraeum (39%) en R. evertsi evertsi (17%). Negatiewe binomiale analise is gebruik om die effek van ligging en seisoen op die algehele bosluistelling vas te stel. Die bosluistellings is dan in Rstudio ontleed deur gebruik te maak van die vaste effek van General Addative modelle (GAM). 'n Akaike Inligtingkriterium (AIC) is gebruik om die beste geskikte model vir analise te evalueer. Daar was 'n positiewe korrelasie (P<0.05) tussen bosluisbesmetting en rektale temperatuur. Die gemiddelde rektale temperatuur van 38.83°C en 38.59°C is onderskeidelik in LNR Loskop en Roodeplaat Navorsingsplase aangeteken. Termografiese beelde van beide verswelgde en ongevoede wyfies en manlike bosluise is vanaf Februarie 2021 tot Julie 2021 van beeste geneem. Die diep leermodelle met argitekture: "ConvNet '' en "MobileNet" is opgelei op 'n datastel van 1124 "termogramme" om bosluise op te spoor op beeste. ConvNet-model het 'n opleiding- en valideringsakkuraatheid van onderskeidelik ~90 en 60% behaal. Terwyl MobileNet 'n opleiding- en valideringsakkuraatheid van onderskeidelik ~95 en 75% behaal het. Laastens is diep leer suksesvol gebruik om bosluise op beeste op te spoor deur gebruik te maak van vooraf- opgeleide konvolusionele neurale netwerke (KNN's).

Masters