Group members: Chenyu Zhang, Mengmeng Liu, Anzhe Xiao, Zepeng Hu
We are interested in bird image classification problems. The input of our algorithm is a bird image. Then we consider using image classification neural networks like VGG, ResNet to classify which species the bird is.
Data set of 315 bird species (100 species in first trial). 45980 training images, 1575 test images(5 images per species) and 1575 validation images (5 images per species). Each figure has 224 × 224 resolution. For preprocessing, we want to use PCA to decrease the dimensions of each data sample so that we can make data more centralized and decrease noise. In addition, if our model meet underfittting problem, we use shifted graphs to augmenting the total size of data. Ideally, we will do the unsupervised task by dropping labels and implement clustering on the data set.