Plant leaf classification using dual path convolutional neural networks

Abstract

Taxon identification is highly needed for a wide variety of research including ecology, agronomy and medicine. As of 1970, classification of plants was introduced into computer vision techniques. Most research conducted in this area focuses on leaves due to their availability as well as their ability to discretize. The most common features researchers base their work on are shape, texture and venation. This research study proposes a dual path, dual feature model for plant leaf identification. We weigh our research on shape and venation features. Sobel operators are used for primary and secondary vein extraction for vein patches generation. Then, a dual path convolutional neural network is employed for feature extraction. This architecture encloses two paths, the first for shape feature extraction and the second for venation feature extraction. The experiment was tested on the Flavia dataset and the results showed an accuracy of 96.8 %.