Rosette plant segmentation with leaf count using orthogonal transform and deep convolutional neural network

Plant image analysis plays an important role in agriculture. It is used to record the morphological plant traits regularly and accurately. The plant growth is one of the key traits to be analyzed, which relies on leaf area (i.e., leaf region or plant region) and leaf count. One of the ways to find the leaf count is counting the leaves using segmented plant region. In this paper, a new plant region segmentation scheme is proposed in the orthogonal transform domain based on orthogonal transform coefficients. Initially, an analysis of orthogonal transform coefficients is carried out in terms of the response of orthogonal basis vectors to extract the plant region. After extracting the plant region, the L*a*b and CMYK color spaces are used for noise removal in the segmentation scheme. Finally, the leaves are counted using fine-tuned deep convolutional neural network models. The proposed scheme is experimented on CVPPP benchmark datasets and also tested with the images taken from mobile phone to ensure its reliability and cross-platform applicability. The experiment results on CVPPP benchmark datasets are promising.