Weedbot works faster and more accurate with Enot technology

Photo: Weedbot
Photo: Weedbot

Enot, a developer of neural network optimization tools, has partnered with Weedbot. Weedbot’s Lumina, which uses laser, can process imagery faster and with more accuracy thanks to Enot’s neural network optimization technology.

Lumina is a laser weeding implement developed by Weedbot. It has a modular design, with one module taking care of one ridge of the crop. Each laser weeding module contains all necessary components to operate fully independently – lasers, cameras and an embedded GPU processing unit.

Laser beam

A camera captures an image from the top of the ridge, whereupon a neural network based algorithm finds all plants in the image and separates weeds from the crop seedlings. As a final step, a laser beam is applied to the weeds so the water in the leaves boils up. After the treatment, the weed’s leaves wilt.

The baseline segmentation model uses high resolution RGB images to be able to detect the smallest parts of the plants.

Weedbot and Enot set out to speed up and improve the model. Enot’s framework enabled Weedbot to find the optimal operations, neural network depth and width, as well as input resolution that meets the required speed of operation at the best achievable accuracy.

Processing speed and accuracy

According to both companies, integration of Enot’s neural network optimization technology in Weedbot’s laser weeding machinery, increased the image processing speed by 2.72 times and achieved a 25% accuracy improvement.

Text continues below image

The baseline segmentation model uses high resolution RGB images to be able to detect the smallest parts of the plants. Enot suggests model parameters to achieve a higher recognition accuracy. - Photo: Enot
The baseline segmentation model uses high resolution RGB images to be able to detect the smallest parts of the plants. Enot suggests model parameters to achieve a higher recognition accuracy. - Photo: Enot

Enot developed two different AI framework versions, with the following goals:

  • V1 – processing time decreased from 22 to 8 ms/img, with the accuracy not decreasing, but increasing.
  • V2 – the recognition accuracy further increased by almost 20%, while the processing time decreased from 22 ms/img to 14ms/img.

By integrating Enot’s framework, Weedbot was able to achieve the following:

  • V1 – processing time decreased from 22 to 8 ms/img, with the accuracy not decreasing, but increasing.
  • V2 – the recognition accuracy further increased by almost 20%, while the processing time decreased from 22 ms/img to 14ms/img.

Figure 1: maximized acceleration
Figure 1: maximized acceleration

Figure 2: maximized accuracy
Figure 2: maximized accuracy
Claver
Hugo Claver Web editor for Future Farming



Beheer