Every year, three million tonnes of pesticides are sprayed on agricultural land around the world. Only a fraction of these chemicals are actually needed. It is time to act! Many suppliers of sprayers, tractors, and agricultural robots have recognized this issue and are tackling it in various ways. For example, some sprayers are equipped with cameras and sensors that capture images of the ground and identify weeds, applying pesticides only where necessary; another approach, which works similarly, completely avoids the use of pesticides by having agricultural robots pull out the identified weeds. Both methods help reduce pesticide usage while increasing crop yields.
This example shows that machines with “vision” and “understanding” hold great potential—especially in ensuring global food supply. But the question remains: how do we teach machines to recognize weeds?
Pre-trained AI algorithms empower edge intelligence
Edge intelligence is the technology of the moment. With the help of AI algorithms, data can be processed directly at the point of generation (i.e., near the sensors). Commonly used sensors include 2D and 3D stereo cameras, lidar, and radar. The generated data is processed through a pre-trained neural network during the inference process. The so-called “inference process” is essentially the process by which software independently draws conclusions from the collected data: sensor data is analyzed and evaluated at the edge, while the neural network continuously learns from new data—for example, recognizing new components in a production process.
Sensors, software, and AI-enabled embedded computers
Implementing edge intelligence requires three core components: software, sensors, and an AI-enabled embedded system. The embedded system, as the hardware core, can process data in real-time and make intelligent decisions based on it. Today, technology from chip manufacturer NVIDIA is often used for such inference tasks. The NVIDIA Jetson product family offers System on Modules (SoMs) with varying performance levels, integrating CPU and GPU technology. With their parallel processor architecture, these SoMs can quickly and efficiently run the software of autonomous machines and devices, especially capable of processing data from multiple high-resolution sensors with virtually no latency.
Another reason to choose the NVIDIA Jetson platform is its developer kits. Companies can use these kits to start software development early and then seamlessly complete the development process on production devices. The platform also provides a wealth of libraries and application-specific frameworks, effectively reducing development costs. Additionally, this series of SoMs is compatible with ROS2 middleware, which has become the ideal tool for computer vision applications, used to control and coordinate a large number of nodes. This middleware has a modular structure, not only providing sensor data processing and evaluation functions but also enabling actuator control.
The latest system-on-module from NVIDIA is the Jetson AGX Orin, which boasts 275 TOPS of AI computing power and 64GB of RAM, with power consumption scalable between 15 and 60 watts. The unique combination of a powerful CPU and a GPU based on NVIDIA’s Ampere architecture enables new computer vision applications across various industries. Typical applications include hazard detection, environmental perception, intelligent video analysis, and the control of autonomous systems.
Embedded computers withstand dust and water
For NVIDIA’s AI technology to function in harsh environments, such as those found in production machines or agricultural robots, specially designed hardware is required. There are a handful of vendors worldwide that have taken on the task of adapting NVIDIA’s technology for such environments. These manufacturers offer NVIDIA Jetson-based embedded computers with IP67 protection. For example, Syslogic combines the SoM with its own carrier board, a robust housing, and clever connector technology. The company’s rugged computers are passively cooled and suitable for a wider temperature range. The company also collaborates with sensor manufacturers who share the same requirements for durability and reliability.
As a result, the potential of AI can also be realized in harsh industrial applications. Power over Ethernet (PoE) or Gigabit Multimedia Serial Link (GMSL) interfaces are used to connect sensors such as lidars, radars, and cameras to the embedded systems, enabling sophisticated computer vision applications. This opens up new opportunities for companies around the world.
Keywords
Keywords: Edge AI, NVIDIA, Embedded Systems, Agricultural Technology, Computer Vision