UAVs in Agriculture: AI on edge, OTA updates & other things

With continuous growth of world’s population, the need for food is also growing - and it’s natural. There are two ways how higher volumes of food production can be achieved, one of them being clearing more land and thus providing more surface to grow agricultural crops on, and the other one is going smarter with what we already have.

But what is “going smarter” exactly? It is all about optimizing the existing food production processes so the given land can be used to its full potential. This is when smart drones (or UAVs) step into the game.

Drones are used for pest control, plant health monitoring, livestock management, soil analysis, and aerial survey with plant health monitoring is being the best illustrative example. There are a lot of factors that contribute to yield productivity - pathogens, fungus, and insects - which can be easily tracked with aerial ispections.

Modern UAVs are equipped with RBG cameras, thermal imaging, and multi-spectral imaging modules. Once gained, the data is transferred onto a cloud platform (see https://up2date.ritms.online for reference) where it can be further analysed, and if needed, your drone fleet go through a full software & firmware maintenance cycle thus ensuring your UAVs are performing great.

UAVs can collect data massively by being equipped with a large variety of sensors: RGB cameras, UAV LiDARs, hyperspectral sensors as well as lightweight cameras. If you have your own own IoT system, you know that it operates with many end-devices, and lots of data is offloaded to the edge devices. That being said, it requires specific mechanism that will help you manage all the data properly.

Artificial intelligence (AI) can be quite helpful in this case. Although multiple AI technologies for data processing on the edge are limited and a numerous criteria when incorporating AI should be considered, it has become one of the most poplular solutions for networking challenges. AI can be used to predict and classify network traffic as well as allocate the resource.

Further, an object detection system can me implemented and GPU acceleration can be mounted to maximize the efficiency of limited device resources and thus increasing the efficiency of flying missions. To ensure real-time drone video analysis, TensorFlow Lite can be used to enable in-device inference from a mobile device and Mobile SDK. Compared to using only the CPU, the average processing time per frame is less while using GPU which makes it much more time-effective to use.