Cattle Race Classification Using Gray Level Co-occurrence Matrix Convolutional Neural Networks

In e-Livestock management system, practical and accurate cattle race identification is paramount. This paper presents a cattle race identification system from their images. We propose a deep learning architecture, which is called as Gray Level Co-occurrence Matrix Convolutional Neural Networks (GLCM-CNN), to semi-unsupervisedly identify a cattles race given thousands of its images with complex background settings. We introduce and evaluate GLCM features, i.e., contrast, energy, and homogeneity into CNN learning for GLCMs capability to recognize pattern with diverse variations, robustness to geometric distortion, and simple transformation. Our experiments show that GLCM-CNN is gives higher classification accuracy and requires less number of learning iterations than the original CNN. In our approach, the data input layer has better distinguishing features than the original image. In addition, our method does not require any prior segmentation process. In this paper, we also address the reduction of computation overhead using saliency maps.