Artificial Intelligence has been pervasive in this decade and with the advent of Covid19, progressive businesses are actively opting for higher degree of automation than ever. Agriculture/horticulture sector is fundamental for New Zealand economy and it is highly reliant on human-centric operations. Despite the fact that most of the farming is now fully mechanised yet there are many tasks that require skilled human resources. Pollination is one of the most important farm management operations which is naturally done by honeybees and insects. Kiwifruit, nevertheless cant rely on that because of its unattractive flowers. This means commercial scale pollination has to be performed by skilled workers. Kiwifruit farmers pay good fortune to ensure timely pollination, maximum farm coverage and least waste of expensive pollens.
Researchers at AiWorx are investigating the use of Artificial Intelligence based solutions that could help the Kiwifruit farmers in the pollination process. For example, realtime Kiwifruit flowers identification for mobile Pollinator system. The system is designed to consume low power and smaller in size to be installed on mobile platforms like agri drones.
The system can operate both in autonomous and semi automated modes. In fully automated mode of operation, autopilot module navigates through the farm and scans each vine and flower using multispectral sensors and cameras. The aligner module tracks the target kiwifruit flower and steers the spraying nozzle in the best position and pollens are sprayed on the detected female flowers. The system also records all the valuable parameters in a timestamped video for later play. It can log pollen and spraying stats, vine data, its appearance/Hue, flower count and their distribution in various locations of the farm. This data could be very useful for better estimation, growth and quality enhancement.
We are using realtime Deep Learning algorithms for high accuracy detection. The following shows the performance of our flower detector. The Pink region in the following images is the annotated Female flower which is passed to the detector. The detector recognises and locates every single female flower w.r.t the spraying system.
The software tracks every detail of the flower in fraction of a second and also adjusts itself for the surrounding conditions. This ability makes it suitable for mobile/vehicle operation and equally serviceable for variable working conditions available at the farm.