Pig Breed Detection Using Faster R-CNN
In this paper, convolutional neural network object detection technology has been used to detect pig breeds with high precision from images captured through mobile cameras. The pretrained model is retrained on several images of 6 different pure breed pigs
- PDF / 962,964 Bytes
- 12 Pages / 439.37 x 666.142 pts Page_size
- 110 Downloads / 332 Views
Abstract In this paper, convolutional neural network object detection technology has been used to detect pig breeds with high precision from images captured through mobile cameras. The pretrained model is retrained on several images of 6 different pure breed pigs obtained from organized farms. The Faster R-CNN InceptionResNet-v2 model has been used in transfer learning fashion for the above task. The training accuracy of this model is 100%, and the testing accuracy of this model is 91% with a confidence level of 94%. From the results achieved, it is noted that this model has produced better results compared to detection accuracy on other datasets like dog dataset, flower dataset, etc. Keywords Convolutional neural network · Faster R-CNN · Pretrained model · Confidence level
1 Introduction Pig breeds are classified by their phenotypic and genotypic characterizations. The phenotypic traits of Indigenous pigs like coat color and skin pigmentation, head shape and orientation, ear shape and orientation, tail shape and orientation, body shape, belly type, top line, hoof placement and presence of wattles were observed and recorded by visual observation. If a larger number of pigs carry similar characteristics in some generations, the group of pigs or pig cluster will form a new pig breed. This is time-consuming and laborious work, and some generations have to wait for new pig breed characterizations. Another way of developing new pig breeds is genotyping characterization. If the DNA sequences of a group of pigs are similar and different from existing DNA sequences of registered pig breeds, the group having the new DNA sequences will form a new breed. In genotyping characterization, generally P. Ghosh (B) · S. Mustafi · K. Mukherjee · S. Dan · K. Roy · S. N. Mandal Department of Information Technology, Kalyani Government Engineering College, Kalyani, Nadia 741235, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 D. Bhattacharjee et al. (eds.), Proceedings of International Conference on Frontiers in Computing and Systems, Advances in Intelligent Systems and Computing 1255, https://doi.org/10.1007/978-981-15-7834-2_19
201
202
P. Ghosh et al.
blood samples were taken to extract DNA sequences. But, collecting this is a very difficult, painful and time-consuming task. The testing procedure is also very costly. Only eight pig breed names are registered in India but many pig breeds are still non-descriptive, i.e. they have not been characterized till now. In this paper, six registered pig breeds have been captured through mobile cameras from organized farms, and faster region-based convolutional neural networks (Faster R-CNN) have been used to detect the breeds with a confidence level. The pretrained version of faster region-based convolutional neural networks (Faster RCNN) known as Inception-ResNet-v2 has been used in transfer learning fashion. The feature extraction part of Faster R-CNN remained intact and top levels have been changed for applying on new image sets [1]. The aim of this paper
Data Loading...
