A heterogeneous implementation for plant disease identification using deep learning

In own global economy, the agricultural sector plays a pivotal role in every aspect of our modern life. One of the most important issues that matters in agriculture, that leads to huge economic losses, are the crop diseases. The reliable and accurate diagnosis of plant diseases, even today, remains one of the most difficult tasks. An efficient, accurate and rapid diagnosis of plant disease is active area of research. One of the solutions that has been proposed is Deep Learning (DL). DL is a vital approach in many fields, including agriculture, as it has the potential to reach a high level of accuracy and efficiency. Various authors have investigated DL techniques for agriculture, but most of them examine a very limited dataset or few models and optimizers. In constrast with existing publications, we have performed the most thoroughly examination of all the state of the art DL models resulting in discovering the best models and parameters for utilizing DL in modern agricalture. The experimental results have shown that the DenseNet201 model in combination with the Adam optimization algorithm achieves the highest testing accuracy score of 99.87% surpassing all other DL architectures.