Ullmann Newman precision cultivates in-row
The Newman in-row cultivator from Czech manufacturer Ullmanna is making strides across Europe and beyond. They are known not only for their own Newman cultivator but also for their weed recognition technology used in other brands’ cultivators.
Few may have seen the blue and yellow cultivator with its striking blue air tanks online, but Ullmanna’s Newman in-row cultivator from the Czech start-up remains largely unknown. Founded in 2019 by Martin Ullmann and his father-in-law Jindrich Ullmann, Martin previously worked as an IT specialist in the Czech military, and Jindrich was involved in agriculture. They now share ownership of the company with two others. This year, a zero series of six units of the cultivator was produced.
The Newman uses cameras to identify plants in the row to cultivate around them. While other smart cultivators do the same, this machine impresses with its level of plant recognition. The cultivator uses AI to recognize individual plants with cameras, a component Ullmanna also provides to other cultivator manufacturers like Feldklasse, K.U.L.T., and Samo. Ullmanna collaborates closely with Czech universities on AI technology.
Lower and go
Once the machine is started and the crop is selected, you can lower the cultivator and start immediately. Nothing else is needed, everything else is automatic. You do not need to input plant spacing or size. We tested this in a cabbage crop, a challenging crop for ‘intelligent’ cultivators because the center of the leaf mass and the root position can vary. To handle this quickly, the algorithm – the set of instructions with which the software recognizes cabbage plants – was first trained by a Ullmanna employee. Currently, algorithms for over 35 crops are available.
An algorithm for an unknown crop can be added within half a day. As with any algorithm, the more images you feed and train the algorithm with, the better the crop recognition becomes. For this reason, the system also uploads every photo taken to Ullmanna’s server, allowing the company to assess whether the AI made the right decision and adjust the crop image database if necessary. So far, more than 300,000 photos of sugar beets have been evaluated! If the machine temporarily has no internet connection, it stores the photos in the interim.
The Newman only focuses on the weeds in the row, between the crop plants. By combining the Newman with a cultivator that cultivates between the rows, a full-field cultivator is created. This is the case with Samo from Austria, one of Ullmanna’s shareholders and also a partner, and with K.U.L.T.
Text continues below picture
What’s in each stainless steel box
Of the 3-meter-wide test machine, we used four of the six rows in the cabbage crop. Wider machines are technically not a problem, but the manufacturer wants to first focus on the zero series and the first production series. More than three rows require a compressor with a minimum capacity of 500 l/min. The technology for recognizing plants is housed in a stainless steel box (box) per row, which you can replace entirely with another box if necessary. According to the manufacturer, the electronics are active as soon as there is voltage on the machine. Without voltage, the electronics switch to standby or off. Graphic cards are absent. Image analysis is done per row with an Nvidia Jetson Nano chip, a type often used for AI applications.
Each box contains components for image recording, height and midpoint detection, and image analysis. Specifically, a 2D camera, a 3D camera, and four strong flash lamps for illuminating the plants. Thanks to these flash lamps, a closed (dark) enclosure is not necessary. Moreover, according to the manufacturer, dust does not affect the accuracy of plant recognition. Each 2D camera takes a photo fifteen times per second, synchronized with the frequency of the flash lamps. These photos serve as the basis for plant recognition. With image analysis, the system determines the center of the rows and controls the hydraulic sideshift accordingly. The 3D camera determines the height of the plants and the distance to the ground. This information is processed to determine and set the correct height of the cultivating blades for each row. For now, the cultivating blades move linearly in the row. Work on a parallelogram suspension is ongoing.
Text continues below picture
Surprisingly good plant recognition
Almost in real-time, all the photos taken are split into about twelve different layers. These include color tones like green, blue, red, black, and white. Based on this, the AI determines whether a crop plant is visible in the photos. According to the manufacturer, this is possible from the one- or two-leaf stage of a crop. Our experience was that this plant recognition and also the midpoint detection functioned surprisingly well in an already well-developed crop like white cabbage. Varying driving speeds of 3 to 6 km/h or a wide variety in plant size do not negatively affect the quality of the cultivation.
Additionally, two cultivating blades per row pneumatically move in and out of the row. This can also be hydraulic and electric, for example, when the Newman box is mounted on a different brand of cultivator. On the associated wirelessly connected tablet, you set how large the movement of the cultivating blades should be and how far they should stay from the plant. Isobus is absent and not needed.
Text continues below picture
Detailed image analysis
The working depth is set electrically. This can be done per row, allowing you to cultivate rows next to a track deeper or shallower. According to Ullmanna, the camera technology works flawlessly up to speeds of 18 km/h. However, the driving speed in practice is limited by the plant spacing in the row and the construction of the moving cultivating blades. Experience shows that these can open up to 100,000 times per hectare in sugar beets. The manufacturer is still working on the robustness of this mechanism and is also testing different cultivating blades.
Based on the detailed image analysis, Ullmanna is also working on recording/mapping individual plants and details. This includes recording plant size, precise plant position, plant distribution, and mapping missing plants. The system is suitable for plant spacings from 5 centimeters and the minimum row spacing is 25 centimeters.
Co-author: Sönke Schulz
