Attention Augmented Convolutional Neural Network for Fine-Grained Plant Disease Classification and Visualization Using Stochastic Sample Transformations

What is it about?

It is not only the number of training images that determines recognition performance but also the discriminative power of the features learned in each data point. The local shared feature in the CNN convolution and the global context obtained through the attention mechanism combined can model the benefit of both local and global features for fine-grained plant disease recognition. Besides stochastic training sample transformation addresses the data-hungry nature of deep models.

Featured Image

Why is it important?

The plant disease recognition task however requires the model to learn fine-grained subtle differences within a single image instead of learning only the foreground part which is the whole plant leaf region that contains the disease lesion.

Perspectives