Recently, poultry disease outbreaks have occurred frequently, harming the poultry industry. Currently, the conventional method for monitoring poultry disease is mainly by manually observing poultry posture, feathers, cockscombs, feces, and sounds. However, the problem with manual observation is that large-scale production requires many people to perform regular inspections. It is both time-consuming and labor-intensive, making it harder to detect sick poultry early. Hence, human surveillance has ceased to be a viable solution in livestock farming. Instead, precision Livestock Farming (PLF) has been used to solve these challenges by using efficient automated systems while at the same time maintaining animal welfare.

With the advancement of artificial intelligence and specifically in the branch of the deep convolutional network, this research proposed an automated system that could detect the poultry's body keypoints using a deep convolutional network. Obtaining the body keypoints of poultry is a crucial step in developing an automated system that could detect the poultry's mobility or posture. However, this research focuses on posture estimation rather than mobility due to the type of data that could be gathered.

Below is the list of objectives for this research.

• To produce a poultry dataset that is annotated with the body keypoints.
• To propose a deep learning model that can detect the keypoints of a poultry on a video.
• To compare the accuracy of the proposed models to other models with different hyperparameters and backbones.

bodyparts:

• center
• head
• tail
• leftleg
• rightleg

skeleton:

• • center
  • head
• • center
  • tail
• • center
  • leftleg
• • center
  • rightleg

This repo is for my Project 2 thesis results.

All the codes used here is the common DeepLabCut command, to understand the parameters of each commands visit DeepLabCut docstrings

A poultry dataset to train a DNN that can detect keypoints on poultry with deeplabcut v2. To view the complete poultry chicken dataset, click the onedrive link training dataset and model outcome