Algorithm 1 Procedures of the proposed fine-grained image classification with misclassification information method.

Input:    Training images ${\mathit{x}}_n^m$ and labels $y_n$, prelearned classifier $f_c^m$, K testing images.

Output:    The predicted classes of testing images:    Training phase

1: Predict the classes of images with prelearned classifiers using Equation (1);

2: Calculate the misclassification information using Equation (2);

3: Train misclassification classifiers using Equation (4);

4: Concatenate the results for new image representation using Equations (4) and (5);

5: Train the final classifiers using Equations (6)–(8).

Testing phase

6: Calculate the misclassification information with prelearned classifiers using Equations (1) and (2);

7: Concatenate the predicted results of testing images using Equations (4) and (5);

8: Predict the classes of testing images using Equations (6) and (8).

9: return The predicted classes of testing images.