Pelt Pattern Classification of New Born Lambs Using Image Processing and Artificial Neural Network

In this study a method to determine the pelt pattern of Zandi sheep lambs using image processing and neural network is presented. Data were collected from Zandi sheep breeding center located in the North East of Tehran (Khojir). In the lambing season, pelt pattern (including regular and irregular patterns) of 300 newborn lambs along with other important characteristics were determined by qualified appraisers. Simultaneously, some photos were taken from each lamb pelt using digital camera. Due to the difference in image resolution and variety of pelt patterns, a total of 170 high quality pictures of lambs were selected and used for final assessment. Two independent image processing scenarios were developed in MATLAB GUI environment. In both scenarios, after the necessary image transformation and segmentation the relevant features were extracted from each image. In the first scenario, some morphological and texture features were extracted from images to classify the pelt pattern of lambs. In the second scenario, the original image firstly was divided into four equal sub-images, and in addition to the texture and morphological features which were extracted in the first scenario, variances and correlations between four sub-images were calculated and added to the features vector. The selected features were used as an input data to the artificial neural network to classify pelt pattern quality of lambs. Input data to the neural network in the first scenario included 21 morphological and texture features, while in the second scenario included 44 features. In both scenarios a three layers Percepteron artificial neural network with feed forward back-propagation algorithms were used. The regular and irregular pattern of lamb pelts were detected by the neural network with accuracy of 92 % and 100% in the first and second scenarios, respectively. The results showed that determination of pelt pattern of lambs based on proposed image processing method is feasible, and substitution of this new method instead of human appraisal method is achievable.