A Hybrid Convolutional Neural Network for Complex Leaves Identification

Abstract

The classification of leaves has gained popularity through the years, and a great variety of algorithms has been created to target these tasks, among those is the Deep Learning approach, which simplicity of learning from raw imputed data makes this task easy to target. However, not all methods are into the complex leaves classification task. In this work we propose a different approach in the way the leaf’s pictures are used to train the models, this is done by using the front and back face of a leaf as one element of the dataset. These pairs will be inputted into two shared convolutional layers, making the models to learn from a complete leaf. The results obtained in this work overpassed the accuracy obtained in related works. For this, we created a new complex leaves dataset, that consists of 6 different kinds of peach varieties, the dataset is available in this link (https://drive.google.com/drive/folders/1rWCr9DrknoK0HKFhNRavCVgZ5UKjU3hi).