Plant Classification Using Conv Neural Network

Classifying plants is essential because they play such an important role in the ecosystem and for the various benefits they provide. Automating the process of identifying plants has proven difficult. This is because feature engineering is not automated by machine learning, making the process of applying machine learning to the problem rather laborious. While deep learning presents an opportunity to automate feature engineering, the need for huge datasets presents an additional challenge. We created a convolutional neural network and used it, after some tuning, on a small leaf dataset. It was accurate to the tune of 99.7% of the time. Due to the limited size of the dataset, significant overfitting was seen. In this scenario, data augmentation was done to the dataset itself, with the photos being augmented before being fed into the model. This is in contrast to the common practise of applying augmentation dynamically.