Our project aimed to address the challenge of mummy detection in real time, utilizing small object detection techniques in large 4K images taken in-field under various lighting conditions. The client's goal was to develop an efficient and accurate solution to detect mummies for robotic removal, which are an indicator of fruit tree health and pose a threat to orchard productivity.


Our contribution involved conducting extensive research, providing valuable camera and lens recommendations to ensure optimal image quality and capture efficiency, and finally data acquisition and management recommendations needed to build a robust model. We evaluated multiple models and training techniques to identify the most suitable approach for mummy detection. Through rigorous training, testing, and continuous model refinement, we achieved a high level of accuracy and efficiency in real-time mummy detection.


To ensure seamless integration, we deployed the trained models as a service wrapped in a Docker image. This allowed for easy integration into the client's existing infrastructure. Our comprehensive approach encompassed the entire project lifecycle, from inception to deployment.


The project's success can be measured in multiple dimensions. Firstly, our model achieved outstanding accuracy and efficiency, effectively detecting mummies in real time. This provided valuable insights into orchard health and enabled real-time intervention to mitigate potential risks. Secondly, the seamless integration of the model as a Docker image ensured easy deployment. Overall, our collaboration resulted in an innovative and effective mummy detection solution, enhancing orchard productivity and fruit tree health.


Through our expertise in camera research, model evaluation, training, testing, and deployment, we successfully delivered a comprehensive and powerful mummy detection system. We are proud to have contributed to the client's goals and achieved a successful outcome for this project.