Medicinal plant leaf information extraction using deep features

In today's digital world of ubiquitous and Internet of thinks, medicinal plant identification is a challenging but very useful task in computer vision (CV) helping agro-community to recognize the unknown species more rapidly. The tchnological improvements in feature representation deep convolutional neural network (DCNN) is promisingly used in several applications like object recognitions, natural language processing and computer graphics. In this paper, we propose a knowledge transfer from object identification to plant species identification where the raw plant leaf image is represented into deep features. These deep features are experimentally proved to out-perform the state-of-the-art in plant species recognition. These paper presents a new and efficient technique for leaf acquisition. Secondly, the image is transformed to device independent lαß color space that is further used to compute VGG-16 feature map. This feature map is re-projected to PCA subspace to optimize the performance for species recognition. To prove the robustness, the paper uses two different types of plant leaf datasets.