Sophisticated AI models for complex crop scenes of plants with overlapping leaves in agricultural Use Cases do not yet exist.
First, collecting consistent, high-quality real-world images (with cameras) is unexpectedly difficult due to seasonality, environmental and weather conditions and variable growth stages.
Second, annotating each object in complex images, at the pixel level, is impossible for humans to do with any accuracy and no significant annotated datasets exist today. AgriSynth has a unique software system that can generate synthetic (artificial) images of biologically accurate crop and weed plant species, together with objects on the leaves of those plants (pests, diseases, etc.) and objects on the soil background (slugs, stones, etc.).
Because the software generates the images, every image can be annotated or labelled with 100% pixel-perfect accuracy. Annotation is where each pixel is allocated to an object, e.g. plant species, plant leaf, plant stem, soil, stone, pest, or disease.
Such annotation allows an AI model to be trained on what each object in the synthetic image is, and to apply this learning to identify such objects in the real world.
AgriSynth can generate synthetic datasets that are more robust and complete than any real-world agricultural datasets that exist today. These datasets will be used to train game-changing AI Models. By enabling the improvement of agricultural AI models, AgriSynth will allow our first-stage target clients (corporates in the seed, fertiliser, and herbicide industries and agri-robotic companies) to dramatically improve their product and service efficiency.
