Automated recognition and harvesting of cabbages may be one step closer to reality, after Japanese researchers came up with methods to tackle some of the key barriers in developing robots for this purpose.
Automated cabbage harvesting is seen as a key target for ‘farming 4.0’ developments in Japan, where an ageing population makes finding sufficient farm labour difficult.
However, technologies proposed to carry out this work have encountered several pitfalls: machine learning models working from camera data have not been able to properly identify cabbages in the evening sun due to areas of light and shade. Additionally, cabbages in the back row of plots which are not meant to be harvested have been incorrectly selected, and harvesting equipment has struggled to follow its correct path in clayey and soft soil, where cabbages are grown following rice cultivation.
Masaki Asano and his colleagues at The University of Tokyo took on a troubleshooting exercise to try and address these issues.
Aiming low for a clearer view
The team carried out a number of tests to improve the visual recognition of cabbages by the models, settling on a system which focused on the lower half of the vegetables for identification where images had a backlight. In field experiments, this approach performed 25 percentage points better than using the whole cabbage to make the identification, while slightly outperforming an approach using the upper half.
In addition, they employed a distance threshold based on images from the system’s RGB-D camera, to exclude the selection of cabbages from the back row of plots. Using this adjustment, the harvesting robot harvested 98.9% of the intended crop.
To address the issue of the harvester’s movement on soft soil, the scientists employed a method of control known as sliding mode control. This was able to account and adjust for differences between expected and actual trajectories in the softer terrain, allowing the robot to move along the desired path.
Overcoming other conditions
Despite the successes, further improvements are needed for cabbage farmers to have a labour-saving tool at their disposal which can be used in a wide range of conditions, Asano wrote in the journal Advanced Robotics.
“Although automatic cabbage harvesting became more robust against environmental changes, there are problems that recognition accuracy decreases due to the snow and effects of different varieties on colour and shape. We will address these issues,” he added.
