The old way: Highly manual, cost heavy fertilisation method
"We wanted to know the bunch count from harvesting images. Currently we are conducting random checking with manual counting. We have millions of photos and we would like to have better capability in bunch-image-counting. In addition to that, we had tree imageries from drones and we would like to have an automated way of extracting data for further analytical purposes. We had heard about AI but would like to be very careful in finding the right partner to harness our data."
Triputra is a large conglomerate in SE Asia, with various business, including agriculture. The Agriculture division of Triputra was looking to more effectively apply fertiliser across their fields of 27 million trees. They had access to millions of images of the fruit yields of their farm but had no way to count them manually. In addition they had imagery from drones and satellite but couldn't use it to monitor their crops. They had heard about AI but were struggling to find the right partner to harness their data.
They wanted AI to help them classify fertilisation needs on a tree by tree basis in order to apply fertiliser more efficiently and approached super.AI with the following challenges:
- How can individual trees be identified from drone images?
- Identification of individual features of trees (crowd diameter, canopy size, estimation of age/height etc)
- How much fertiliser does a tree need?
Moving to automation with super.AI
Triputra reached out to super.AI, since it had heard that the company has simplified the complexity of ML projects and provided its customers with a simple unique interface that allowed them to customize their needs based on three dimensions: quality, speed and cost. Following a Proof of Concept, Triputra was happy with the high degree of data accuracy, provided within a very quick timeframe at the desired cost. They therefore decided to entrust us with solving their problem.
super.AI provided Triputra with an API that can automatically count the fruit yield of their farm. We programmed a customised computer vision model for them that took their drone video footage as an input and provided data input for tree and feature identification. We then created an automatic database of their trees from drone and satellite images powered by AI modelling.
"After the project, we got a better understanding as to what our data meant to us. It gave us insights on some important matters. We also got a valuable insight into what AI can and cannot do"
After taking all the drone footage and running it with our computer vision algorithm, we provided Triputra a map outlining the various fertilisation needs for each tree. Using this map, the company was able to decrease the usage of fertiliser by 20%, while keeping tree growth levels the same. Encouraged by this significant cost saving as well as automation of a highly manual process, the company is now considering how to integrate AI modelling into existing systems to enhance operational capabilities.