Ensemble of Deep Neural Networks for Rice Leaf Disease Classification

Rapid disease identification is critical for prompt treatment and reduces crop losses. However, in developing countries, rice disease diagnosis is still mainly performed manually. Currently, deep learning is developing significantly and its applications in agriculture are undeniable. So we decide to propose a method using deep learning in which we ensemble CNN models based on the combination of two neural network architectures ResNet and DenseNet. We get a high accuracy result in our experiments to classify rice leaf disease images. The experimental results show that our combination is compatible with the CNN model and gets an average test F1 score of 0.96 for 10 trials. We believe that our initial results in this project can demonstrate the possibility of wide application of artificial intelligence for agriculture and contribute efficiently to the economy.