Motivation

  At certain times of the year, large flocks of migratory birds will migrate long distances to find their familiar habitats, and spend the summer and winter there. However, the development and destruction of habitats, such as wetlands, intertidal zones, swamps, etc., and even recent climate changes may make migratory birds unable to find the ideal habitat they were once familiar with, which make the process of migration become a thrilling journey.

  Therefore, we hope to build a set of models through deep learning, and by inputting pictures, we can accurately calculate the number of migratory birds inside, and then be able to monitor changes in the ecological environment.

Research Process

  Because the data is not easy to obtain, there is no dataset to train the model, therefore we labeled the dataset manually.

  Although the model perform well on generating the heat map, but the number of birds the model predicted is still not that accurate.

Issue

  We are studying in new approach, which is not manual, to generate the ground truth of training data, which may bring us more training data.

  The accuracy of the model need to be improved.