Shiny APP for Bird Recognition Using Convolutional Neural Network
An easy-to-use website for bird image classification. Upload a photo (any resolution will be fine) of bird, then it will tell you the top-5 most likely bird species it belongs to.
Demo
Dataset & Model
Caltech-UCSD Birds-200-2011 (CUB-200-2011) is used as the dataset. It contains 11,788 images of 200 bird species, including Johns Hopkins’s mascot, Blue Jay!!!!!!
To form the classification model, the Resnet-50 and two fully conneted layers are concatenated together. The Resnet-50 was pretrained on ImageNet by Keras. All of its parameters are freezed during training. Data augmentation techniques are used, including rotation, shifting, shearing, zooming, flipping, to avoid overfitting.
On the validation set (20% of the dataset), the model achieves an accuracy of ~ 0.35, which is far lower than those of the benchmarks. The simple and shallow structure of the fully connected layers is to blame. I will take some deep learning courses to build a more robust model in the future.
References
https://gerinberg.com/2019/12/10/image-recognition-keras/
https://www.r-bloggers.com/2018/02/deep-learning-image-classification-with-keras-and-shiny/
https://keras.io/api/applications/
