A study of duck detection using deep neural network based on RetinaNet model in smart farming

Area of Science:

• Agricultural Technology
• Computer Vision
• Animal Science

Background:

• Duck welfare in cages is crucial for optimal growth and requires continuous monitoring.
• Adverse events such as overturned, fallen, or deceased ducks negatively impact the farm environment.
• Traditional cage management is labor-intensive and may miss critical welfare issues.

Purpose of the Study:

• To propose and evaluate an object detection algorithm for automated monitoring of duck welfare in cages.
• To improve the efficiency and accuracy of identifying critical events affecting duck health and growth.
• To lay the groundwork for smart farm integration in duck husbandry.

Main Methods:

• Collected and augmented image data from duck cages over two years (2021-2022).
• Utilized object detection algorithms for classification and localization of ducks and potential issues.
• Trained and verified the model using a 9:1 data split, with final validation on unseen images.

Main Results:

• The object detection method successfully identified key objects and states within the duck cage environment.
• Visual confirmation demonstrated the algorithm's capability in real-world scenarios.
• The study established the feasibility of using computer vision for duck welfare monitoring.

Conclusions:

• The proposed object detection method offers a promising solution for automated duck cage management.
• This technology can significantly reduce the need for manual labor in monitoring duck welfare.
• Implementation of this system can contribute to the development of smart duck farms.