Aetina Smart Farming Solutions: Edge AI for Agriculture, Forestry, and Aquaculture

Manually identifying crop diseases in the field is costly, inconsistent, and slow. Inspectors must physically examine individual plants, and their assessments vary based on experience level and fatigue. Even when disease is correctly identified, the time needed to reach a laboratory for confirmation often means the intervention comes too late to prevent significant crop loss.

Developed in collaboration with an ISV partner, this solution uses Aetina's AIE-KN42 edge AI computing platform combined with the NVIDIA TensorRT inference SDK to deliver instant, on-site health diagnostics for mango leaves. A camera captures leaf images directly in the field. The AI model — trained on a comprehensive disease dataset and optimized for TensorRT inference — analyzes color, texture, and morphological patterns in real time to classify the leaf as healthy or to identify the specific disease present, without any cloud upload or laboratory delay.

Because the AIE-KN42 performs all inference locally, latency is minimal and the system operates independently of internet connectivity. The same trained pipeline can be retrained for other crop species with additional labeled datasets, giving the solution broad applicability across different agricultural contexts.

• Instant, on-site diagnostics — farmers receive actionable results in the field without waiting for laboratory analysis.
• Better crop management decisions — timely identification of disease enables targeted interventions before spreading occurs.
• Reduced chemical use — precise detection prevents blanket pesticide application, lowering both cost and environmental impact.
• Higher yield and quality — early disease mitigation directly improves harvest outcomes.

The global food production shortfall cannot be solved by recruiting more human labor alone. Rural demographics are aging, interest in agricultural work is declining among younger generations, and birth rates in many farming regions are falling. Robotic harvesting systems offer a scalable path forward — but only if the onboard computing platform can handle the real-time AI workloads required for precise object detection, path planning, and mechanical actuation.

Aetina's Smart Farm System is built around the AIB-MX22, a MegaEdge MXM-format carrier board powered by the NVIDIA Jetson AGX Orin module. Delivering up to 275 TOPS of AI inference throughput, the AIB-MX22 provides server-class edge computing in a compact, field-deployable form factor. The system supports LTE/5G (M.2 B-Key), Wi-Fi/Bluetooth (M.2 E-Key), and NVMe storage (M.2 M-Key), enabling connectivity and local data storage regardless of network availability.

Four PSE (Power Sourcing Equipment) PoE ports connect directly to field cameras mounted on the harvesting robot, while a 10GbE port provides 10× the bandwidth of standard Gigabit Ethernet for transferring high-resolution image and sensor data at full speed. The wide input power range of 9–36 VDC ensures compatibility with a variety of mobile and battery-powered agricultural platforms. Aetina provides one-stop project support from initial evaluation through to after-sales training, accelerating time-to-deployment for agricultural robot manufacturers.

• Lower workforce overheads — robots replace repetitive manual harvesting labor without quality degradation.
• Higher crop yield — AI-guided harvesting minimizes damage and missed crops.
• Smaller operational footprint — agricultural robots occupy less space than conventional machinery.
• Fewer errors in planting, irrigation, and pesticide application — AI-driven precision eliminates over- and under-treatment.
• Lower pesticide consumption — targeted application reduces chemical costs and environmental load.

Traditional timber stack measurement is manual, labor-intensive, and hazardous — requiring workers to leave vehicles in harsh weather conditions to physically assess stacks. Dralle A/S, a Danish forestry technology company operating across Europe, reached the limits of their legacy x86 platform: the system's classic computer vision algorithms (watershed and Hough transform) could not support modern machine learning workloads, and the hardware's physical size was incompatible with their compact camera box design.

Aetina partnered with Dralle to migrate their sScale timber measurement system from the legacy x86 platform to an ARM-based NVIDIA Jetson edge device. Aetina's DeviceEdge platform — featuring M.2 and GigE expansion ports for seamless camera integration — provided the hardware foundation, while Aetina's engineering team facilitated the full migration in just three months. The transition enabled Dralle to deploy sophisticated machine learning models for real-time log end detection and measurement, replacing the previous classical computer vision approach entirely.

The results were dramatic: the new system detects up to 1,000 unique log ends per frame in just 200 milliseconds, compared to the multi-second processing times of the previous platform. Measurement accuracy improved significantly due to the shift from heuristic algorithms to trained neural networks, with minimal false positives. Automated measurements mean workers no longer need to exit vehicles to inspect stacks, eliminating exposure to weather and terrain hazards. Dralle's sScale system now provides all-weather, all-season, day and night, off-grid timber measurement at the accuracy levels required for commercial forestry trade.

• Detection speed: up to 1,000 log ends per frame in 200 ms.
• Accuracy: machine learning significantly outperforms classical watershed/Hough transform approaches.
• Worker safety: automated measurement eliminates the need to leave the vehicle.
• Migration time: full transition from x86 to Jetson ARM platform completed in three months.
• Operating conditions: all-weather, all-season, day/night, off-grid capability.

Modern fish farms can hold up to 200,000 fish in a single pen. Accurately tracking the biomass, health, and growth of individual fish at that scale is impossible with manual sampling — the traditional method. Manual counting and visual assessment are slow, imprecise, and stressful for fish, leading to over- or underfeeding, higher operational costs, welfare issues, and unsustainable practices.

Optoscale, a Norwegian aquaculture technology company, solved this by integrating Aetina's NVIDIA Jetson AGX Xavier-based edge AI platform into their underwater biomass estimation system. The Jetson device was placed inside an underwater camera housing, where it processes continuous high-resolution video streams of fish in real time — entirely at the edge, without cloud dependency. The porting process from Optoscale's previous system took approximately six months and required no customization of the carrier hardware, enabling a seamless integration with existing infrastructure.

The system achieves biomass estimates within 1% of actual weight — a level of precision that was previously unattainable with manual methods. The Jetson platform's advanced GPU capabilities allow Optoscale to run complex neural network models that assess fish size, shape, and behavioral patterns at the individual level, enabling feeding decisions to be made in real time. For remote fish farms without reliable internet access, edge processing is not just advantageous — it is a fundamental requirement.

• Biomass accuracy: estimates within 1% of actual weight — far beyond manual sampling.
• Scale: supports pens with up to 200,000 fish using automated measurements.
• No cloud required: operates fully at the edge in remote offshore environments.
• Reduced feeding waste: real-time data enables precise feeding schedules, cutting cost and environmental impact.
• Improved fish welfare: continuous non-invasive health monitoring reduces stress compared to manual handling.
• Sustainable operations: data-driven decisions support long-term aquaculture sustainability.