Improved Mask R-CNN Segmentation for Contour Extraction of Individual Cattle from an Image

Abstract

Forest species recognition has typically been approached as a texture classification problem and investigated by employing conventional texturing techniques like Local Convolutional Neural Network (CNN). Procedures have been the subject of current research for classification issues, with cutting-edge technique outcomes for object recognition and other activities, but are not yet commonly employed for texture problems. Several forest datasets, including one with macroscopic photos and another with microscopic images, are examined in this process to see how deep learning strategy, more specifically Convolutional Neural Networks (CNN), may be used to classify textures. Given the better resolution photos of these issues, we provide a strategy that can deal with them in order to get excellent accuracy and steer clear of the load of training and constructing an architecture with a lot of free parameters. On the first dataset, the suggested CNN-based technique superior to cutting-edge technology, which achieves an accuracy of 95.77%. It surpasses the best reported result of 93.2% on the dataset of microscopic pictures, achieving 97.32%.