AI Weed Targeting
AI weed targeting has been developed by researchers, and is reducing pesticide use by 80%.
AI and the Pesticide Problem
Weeds compete with farmers' crops for light, nutrients and water. In order to maximise crop yield, and profits, farmers spray entire crops with pesticides to kill these weeds. But there are major drawbacks to pesticide use. Pesticides are known to be extremely toxic substances, many of which are known to cause cancer. When these pesticides are sprayed on the crops, they are absorbed by them, and end up in the human food chain. It has been established that some of these chemicals cause cancer, low sperm count and many other illnesses. Yet somehow they are still allowed to be used. The EU has the most stringent regulations and has banned several of the most toxic pesticides. Other countries, including the USA and Australia have barely any regulation and permit large quantities of toxic chemicals to be sprayed onto our food each year.
AI Weed Targeting
The researchers are from the Insitut National des Sciences Appliquees and the University of New Orleans. They developed AI that can detect weeds from images of beet, bean and spinach crops. The images are taken by drones which then use georeferenced coordinates to check where the photograph was taken. This allows the pesticides to be sprayed only onto the weeds, rather than over the whole field. They hope to utilise agricultural robots to then target the weeds.
Hafiane and his colleagues used a cluster of NVIDIA Quadro GPUs to train the neural networks. Their work was supported by France’s Centre-Val de Loire region.
Deep Learning on Crop Images
To a human looking at the images, it would be impossible for us to determine where the weeds were. A very skilled agriculturalist may be able to do so, but it would take a long time. It would also be almost impossible to then locate the weed accurately, and take so long that it would not be a commercially viable process. But the deep learning, artificially intelligent neural networks are able to achieve it. The dataset used by the system contains tens of thousands of images of each crop: the team used transfer learning based on the popular ImageNet model to develop its deep learning system.
Crops are sown in neat lines, so when there is green visible between those lines, or rows, then it is likely there are weeds present. Detecting the weeds within the crop rows was more challenging and that process is still being improved.
The AI system has an accuracy of 93% on beet crops. The technology is now being developed to utilise in a range of different purposes, including vineyards.
AI weed targeting is deep learning and deep tech at its very best. Solving problems considered impossible and making the world a safer, healthier place.
Read more about clever uses of AI an deep tech
At Volanto we are strategic growth consultants. We are passionate about the power of tech to improve businesses and the world. We report on interesting developments in the tech world, like this AI weed targeting AI.
