Reason-A Technical Journal (Formerly Reason-A Technical Magazine)

Satyendra Nath Mandal ¹, Sanket Dan ¹, Pritam Ghosh ¹, Subhranil Mustafi ¹, Kunal Roy ¹, Kaushik Mukherjee ¹, Dilip Kumar Hajra ², Santanu Banik ³

Affiliations
1 Kalyani Government Engineering College, Kalyani, Nadia, West Bengal 741235, IN
2 Department of Agronomy, UBKV, Pundibari, Cooch Behar, West Bengal 736165, IN
3 ICAR-National Research Centre on Pig, Rani, Guwahati, Assam 781131, IN

Abstract

Diversity in domestic animals in most of the species is depicted in the form of breeds. Phenotypic and genotypic characterizations are the tools for breed identification of livestock species. Variation within breed or similar looking breeds make it difficult to confirm breed identity of individual animal. An experiment was conducted with the aim of identification of breed of an individual goat by the help of its image using Inception model v3; a convolutional neural network. More than 500 digital images of individual goat captured in restricted (to get similar image-background) and unrestricted (natural) environment without imposing stress to animals. Six different purebred goats (Blackbengal, Beetal, Jamunapari, Barbari, Jakhrana and Sirohi) which have been reared and maintained by reputed government research organizations in India were used for training and testing the model. 10% of the captured images were used for testing the trained model. Breed confirmation was made by seeing the value (probability) in output terminals corresponding to six different breeds under study which best described an input image. 56 images out of the 60 images used in the test were successfully interpreted for breed identity by the trained model and thus the model was 93.33% accurate. Output probability of more than or equal to 0.95 was taken as minimum confidence limit for determination of breed. Value less than 0.95 was considered as unsuccessful test. Upon testing with images from breeds for which the model was not trained on, the output values could not provide confirmatory result. Therefore, the technique has great potential to solve confusion on breed identity. It would also be useful in implementation of Global Plan ofAction for animal genetic resource (AnGR).

Keywords

Livestock, Goat Breed Identification, Deep Learning, Convolutional Neural Network, Confidence Level.

Full Text