E-Nose Identification of Milk Somatic Cell Count

Mastitis is a common disease among dairy animals which causes serious economic losses. It can be diagnosed via diverse clinical findings, while milk somatic cell count (SCC) is accepted as a key indicator. However, determination of SCC with traditional methods is time consuming and laborious. This paper focuses on the ability of electronic nose (e-nose) system containing 12 different metal oxide sensors (MOS) to discriminate milks with somatic cell counts (SCC) above a threshold value. Milk samples were collected from dairy farms around Biga district of Canakkale province, Turkey. Forty-six samples were analyzed using standard protocols in laboratory, then exposed to DiagNose II e-nose system. Artificial Neural Networks (ANNs) was used to discriminate between Non-Mastitic (N-M) / Mastitic (M) samples depending on sensor responses. Results showed that 8 of 12 sensors were responded to milk samples. Thus, performances of several ANNs models with different topologies were tested using 8 sensor responses. ANNs was trained using 28 samples, and remaining 18 samples were used in validation step. Among tested models, the results of the lowest overall errors for training and validation steps were found to be 35.71 % and 38.89 % respectively. To improve the performance, Principal Components Analysis (PCA) performed for dimension reduction and three components were selected to be included in ANNs model instead of 8 sensors. Performing of PCA prior to ANNs provided decreased overall errors for training (10.7 %) and validation (0 %). However, the actual performance of the system should be tested using new dataset. Mastitis sagmal hayvanlar arasinda yaygin bir hastalik olup onemli ekonomik kayiplara sebep olur. Hastalik cesitli klinik bulgularla teshis edilebilirken sut somatic hucre sayisi (SHS) kilit gostergelerden biri olarak kabul edilmistir. Bununla birlikte SHS’ nin geleneksel yontemlerle belirlenmesi yogun emek gerektirir ve zaman alicidir. Bu calisma farkli metal oksit sensorler (MOS) iceren elektronik burun (e-burun) sisteminin bir esik degerin uzerinde SHS iceren sutleri ayirt edebilme yetenegi uzerine odaklanmistir. Sut ornekleri Canakkale ili Biga ilcesinde bulunan ciftliklerden toplanmistir. Kirk alti ornek laboratuarda standart protokoller ile analiz edilmis ve ardindan DiagNose-II elektronik burun sistemi olcumune tabi tutulmustur. Sensor tepkilerine gore Mastitik-Olmayan (M-O) / Mastitik (M) sutlerin ayirt edilmesinde Yapay Sinir Aglari (YSA) kullanilmistir. Sonuclar 12 sensor icerisinden 8 sensorun sut orneklerine tepki verdigini gostermistir. Bu nedenle farkli topolojilere sahip cesitli YSA modellerinin performansi 8 sensorun tepkileri kullanilarak test edilmistir. Tum YSAlari 28 ornek kullanilarak egitilmis ve kalan 18 ornek ise gecerlik asamasinda kullanilmistir. Test edilen modeller arasindan egitim ve gecerlik asamalarina iliskin en dusuk hata sonuclari sirasiyla % 35.71 ve % 38.89 bulunmustur. Performansin artirilmasi amaciyla boyutlari azaltmak icin Ana Bilesenler Analizi (ABA) uygulanmis ve 8 sensor yerine 3 bilesen YSA modeline dahil edilmistir. YSA calistirilmadan once ABA uygulanmasi egitim (% 10.71) ve gecerlik (% 0) asamalarindaki hatalarin daha dusuk olmasini saglamistir. Ancak sistemin gercek performansi yeni veri setleriyle test edilmelidir.