Cargando…

Deep ensemble learning for automatic medicinal leaf identification

The therapeutic nature of medicinal plants and their ability to heal many diseases raises the need for their automatic identification. Different parts of plants that help in their identification include root, fruit, bark, stem but leaf images have been widely used as they are an abundant source of i...

Descripción completa

Detalles Bibliográficos
Autores principales: Sachar, Silky, Kumar, Anuj
Formato: Online Artículo Texto
Lenguaje:English
Publicado: Springer Nature Singapore 2022
Materias:
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9373896/
https://www.ncbi.nlm.nih.gov/pubmed/35975198
http://dx.doi.org/10.1007/s41870-022-01055-z
Descripción
Sumario:The therapeutic nature of medicinal plants and their ability to heal many diseases raises the need for their automatic identification. Different parts of plants that help in their identification include root, fruit, bark, stem but leaf images have been widely used as they are an abundant source of information and are also easily available. This work explores the branch of Artificial Intelligence, called deep learning, and proposes an Ensemble learning approach to rapidly detect medicinal plants using the leaf image. The medicinal leaf dataset consists of 30 classes. Transfer learning approach was used to initialize the parameters and pre-train Neural networks namely MobileNetV2, InceptionV3, and ResNet50. These component models were used to extract features from the input images and the softmax layer connected to the Dense Layer was used as the classifier to train the models on the concerned dataset. The obtained accuracies were validated using threefold and fivefold cross-validation. The Ensemble Deep Learning- Automatic Medicinal Leaf Identification (EDL-AMLI) classifier based on the weighted average of the component model outputs was used as the final classifier. It was observed that the EDL-AMLI outperformed the state-of-the-art pre-trained models such as MobileNetV2, InceptionV3, and ResNet50 by achieving 99.66% accuracy on the test set and average accuracy of 99.9% using threefold and fivefold cross validation.