Towards Smart Agriculture: A Deep Learning based Phenotyping Scheme for Leaf Counting

Abstract

Plant phenotyping is a smart technique in which plant features data is collected and analyzed using computer vision, robotics and machine learning techniques to increase agricultural production. We propose a leaf segmentation and leaf counting technique based on learning without using the denotation of the leaf center and the data on the plant segmentation given in the LCC CVPPP 2017 dataset. After required segmentation, noise removal and enhancement, as well as the transformation of leaf pixel data, a deep neural network architecture based on Alexnet, was used on a total of 783 plant images by dividing the dataset into 70% for training, 15% for validation and 15% for testing. The result thus obtained showed significant improvement based on four evaluation parameters such as Count Difference, Absolute Count Difference, Percentage of Agreement and Mean Square Error when compared with contemporary works.