Bestemmelse av kjøttprosent, og automatisk multivariat klassifisering av vevstyper i levende svin og svinekjøtt

The implementation of the project "Determination of meat percent, and automatic multivariate classification of tissues in live pigs and pork (PIGCOMP)". Collection of raw data (WP1) was completed without any major deviations from the plan. This means that the was collected data from 240 half Norwegian pig carcasses. The carcasses are stratified into 16 different groups according to weight and fatness. For all carcasses the usual slaughter data (i.e. the data recorded for all pigs slaughtered in Norwegian slaughterhouse) are also registered. All these 240 carcasses were scanned with two different CT scanners, one Danish and one Norwegian. In the Norwegian scanner with three different voltage levels. These CT data is the main material that are still being developed. 20, the carcass is also manually dissected.   The PhD-student, Lars Erik Gangsei (LEG), completed his PhD-thesis at the Norwegian University of Life Sciences on 24 May 2016. A central part of LEGs thesis was incorporation of CT results as a "supportive" dissection. This is thoroughly explained in the article published in Acta agriculturae Scandinavica, Section A - Animal Science (cf. "Publications"). This issue also initiated two more theoretical articles in the PhD thesis submitted to Communucation in Statistics. The methodology developed in these articles have a far wider application than the prediction of meat percentage of pig. In these articles was an expectation skewed estimator, known as James-Stein estimator, analyzed a problem with missing data. Every year the company Topigs Norsvin CT-scans approximately 3500 Norwegian male pigs that are appropriate to use in breeding. CT provides very valuable knowledge for breeding, but the value will increase significantly if we can find an effective method for identifying various organs, prime commercial cuts etc. in CT images. Atlas segmentation is one such method. The atlas can be seen as an average pig. The underlying idea is that one can transform ( 'squeeze') the individual subjects in the atlas (atlas form), where bodies, prime cuts etc. are defined in advance. Thus, one identifies the various organs / prime beef cuts in individual. The transformation is a multivariate linear prediction, where the predicted values ??consist of coordinates in the atlas room and explanatory variables are based on the basic functions of the coordinates of individual rooms. To construct the atlas, and transformations, we based ourselves on landmarks. These are points with known coordinates in both the atlas room and in individual rooms. Skeletal structure, and the surface (skin), were used to identify these points. Paper IV (published in Computers and Electronics in Agriculture) describes solely an algorithm for automatic segmentation and identification of the main bones of the skeleton. This algorithm is important for paper V (submitted to Computers and Electronics in Agriculture) where we describe how the atlas is constructed and used for segmentation. The whole paper V is based on standard image analysis techniques.

PIGCOMP will address challenges regarding online classification of pork and the establishment of dissection by CT as reference method. The aim is to update the fundamental premises for valuation of pork by developing a new model for classification of lean meat, fat and bone in the carcasses. It is a great challenge to improve the routines for analysis of CT pictures. Both for CT-dissection of carcasses and CT-analysis of live animals for breeding purposes, are there a need for improved and faster readings based on sound statistical analyses and as automatized as possible. Algorithms for segmentation and classification of tissue in digital pictures will be developed. The project will be a collaboration between Norwegian, Swedish and Danish partners, and th e project is also linked up to the activities in the COSTaction FAIM.Industrial end-users are well integrated in the consortium and will imidiately put the results into action. The results will have great impact on producer's, industry's and breeding comp any's economy and improvement of routines. PIGCOMP tar for seg utfordringer innen områdene online klassifisering av svineslakt og etablering og anvendelse av CT-disseksjon som referansemetode. Målet er å oppdatere grunnforutsetningene for verdisetting a v svineslakt ved å utvikle ny modell for å bestemme innholdet av kjøtt, fett og ben i slaktene. Det er en stor utfordring å forbedre analyserutinene av CT-bilder. Både til CT-disseksjon av slakt og til CT-analyse av levende dyr i avlsformål det behov for bedre og raskere avlesning basert på sunne statistiske analyser og mest mulig automatiske rutiner.Det skal derfor utvikles nye algoritmer for segmentering og klassifisering av vev i digitale bilder. Prosjektet skal gjennomføres i samarbeid med svenske og danskefagmiljøer, og bygger på aktivitetene i COSTaction FAIM. Industrien er godt integrert i konsortiet og vil sikre at resultatene raskt tas i bruk.Reultatene vil ha stor betydning for økonomien i verdikjeden.