Agriculture tech (AgTech) is one of the fastest-growing industries in the last five years, accumulating a 410% growth since 2017. Something that can be attributed to recent technological advancements as well as the constant pressure to produce more, with lower costs and in a sustainable way. More precisely, in 2020, AgTech companies received investments of $6.1B, an increase of 60% over 2019, while the growth for 2021 was even more significant, 137% translating into $12.23B. AgTech includes various technologies such as robotics and computer vision combined to solve complex agricultural problems, one of those being disease detection.
Whether caused by a fungus, bacteria, or virus, plant diseases can cause significant economic damage to farmers. On the one hand, the damage caused can be direct such as yield reduction due to stunted growth or even damage to the fruit, making it non-harvestable. On the other hand, indirect economic losses can occur through product quality downgrade (e.g., small fruit, spots). As a result, a large amount of the production costs is allocated to disease inspection, prevention, and countermeasures. All these actions increase production costs and agriculture’s environmental footprint while exposing farmers to dangerous chemicals putting their health at risk. Artificial Intelligence and computer vision hold the promise to automate disease detection using cameras and at a later stage feed this information to agricultural machinery. Thus, facilitating the transformation of uniform blanket applications into targeted ones, saving money and the environment simultaneously. This becomes possible by spraying only the diseased parcel of the field, reducing the amount of plant protection products used.
Additionally, thorough monitoring of crops through imagery allows for detailed assessing at cm level of the disease pressure, enabling data-driven decision making about the optimal course of action. Disease detection models have been under development for over a decade and for various plants and diseases. Examples are Apple scab and black rot for apples, early blight and Cercospora for Celery, Leaf blight, and Esca for grapes. While in the infancy, those models were restricted to lab conditions, lately, more and more models are trained and applied under real condition environments with promising results that, on some occasions, reached up to 100% accuracy. The technology and hardware to develop a commercial system are already available. However, there is still something missing for the commercialization of such models, data, data, data. A vast data pool is needed to create a robust, field invariant disease detection system. This pool should consist of a variety of field images, acquired under various illumination and weather conditions, plant and disease growth stages, and different varieties, not to mention that the orientation and distance of the camera to and from the point of interest will need to be consistent and match the conditions where the system will be deployed.
We are one step away from commercializing disease detection systems as both software and hardware have reached the minimum requirements. The secret ingredient that is missing is data.
References: